assessment of pedestrian discomfort glare from urban led

Assessment of pedestrian discomfortglare from urban LED lightingC Villa PhD, R Bremond PhD and E Saint-Jacques MScLaboratoire Exploitation, Perception, Simulateurs et Simulations (LEPSIS),Institut Francais des Sciences et Technologies des Transports, del’Amenagement et des Reseaux IFSTTAR, Champs-sur-Marne France

Received 1 July 2016; Revised 7 September 2016; Accepted 19 September 2016

Due to different visual tasks and gaze patterns, the discomfort glare experiencedby pedestrians may differ from that experienced by drivers. This paper investigatesthe discomfort glare experienced by pedestrians under various urban LEDluminaires through psychovisual experiments conducted on a test track. Theability of state-of-the-art models to predict the level of discomfort glare, measuredon the de Boer rating scale, for this application is also investigated. With oneexception, the models all overestimate the mean subjective discomfort glarecompared to the experimental data. Models proposed by Lin et al. (2015) andBullough et al. (2008, 2011) perform well. However, the implementation of thesemodels is not straightforward because choices are needed to estimate some of thevariables such as the background luminance and the glare source area.

Nomenclature

I80 luminous intensity at 808 from the

vertical (in cd)I88 luminous intensity at 888 from the

vertical (in cd)F luminous surface of the luminaire seen

at 808 from the vertical (in m2)h’ adjusted luminaire height (luminaire

height – 1.5m (observer’s eye height))

(in m)p number of poles per kilometre� eccentricity, i.e. the angle between the

line of sight and the light source (in arc

minute in Sch74, in degrees in the other

models)La adaptation luminance (in cd/m2)Lb background luminance (in cd/m2)Ll source luminance (in cd/m2)

El vertical illuminance due to the source

(in lx)Es vertical illuminance due to the immedi-

ate surround of the source (in lx)Ea vertical ambient illuminance (in lx)Eb vertical illuminance due to the back-

ground (in lx)! solid angle from the observer (in sr)

Ebaffle vertical illuminance measured when

the illuminance meter is covered with a

circular baffle to remove the direct illu-

minance from the source (in lx)ELantern vertical illuminance due to the lantern

(in lx)ELED vertical illuminance due to the LED

module (in lx)LLantern average vertical luminance of the lan-

tern (in cd/m2)LLED average vertical luminance of the LED

module (in cd/m2)Lmax maximum luminance of the source (in

cd/m2)LRZ average luminance of the road zone (in

cd/m2)

Address for correspondence: Celine Villa, IFSTTAR/LEPSIS,Cite Descartes, 20 Boulevard Newton, 77447 Champ-sur-Marne, Marseille, France.E-mail: [email protected]

Lighting Res. Technol. 2016; 0: 1–26

� The Chartered Institution of Building Services Engineers 2016 10.1177/1477153516673402

at INRETS on October 10, 2016lrt.sagepub.comDownloaded from

http://lrt.sagepub.com/

LDZ average luminance of the disc zone (308circular zone) (in cd/m2)

ERZ vertical illuminance computed fromLRZ (in lx)

EDZ vertical illuminance computed fromLDZ (in lx)

Lblack average luminance of the whole imagecaptured when all the lanterns areswitched off (in cd/m2)

Eblack vertical illuminance computed fromLblack (in lx)

1. Introduction

1.1. Context

Urban lighting is designed for all roadusers, including pedestrians. As far as pedes-trians are concerned, the main factors influ-encing lighting quality are feeling of personalsecurity, facial recognition and visibility.1

However, the performance of a lightinginstallation must be balanced against itscosts in terms of financial cost, energy con-sumption, light pollution and glare.Nowadays, pedestrian comfort has become aresearch topic,2–5 especially with the increas-ing use of LED luminaires. Indeed, the smallsize of LED chips can lead to high luminancesthat may produce discomfort glare.6 Kosticand Djokic conducted a field experiment tocompare subjective judgments on LED andmetal halide lighting with respect to variouscriteria about the light quality (illuminance,colour) and pedestrians’ experience (security,comfort).2 Discomfort glare was evaluated ona five-point scale while walking along thepath, looking straight ahead. Luo et al.3

studied pedestrian zone lighting by collectingpreference judgments and eye fixations fromfive participants, but no glare was reported bythe participants in this experiment. Milleret al.4,5 investigated pedestrian-friendly out-door lighting. To avoid pedestrian glare, theauthors recommended limiting the maximumluminance, and avoiding strong luminous

intensity/luminance variations with the view-ing angle. They advised using frosted anddiffusing optics to avoid a direct view of theLEDs. In addition, they raised methodo-logical problems for discomfort glare assess-ment, including the photometric descriptionof the scene (source size, background areadefinition). They also highlighted the need fornew investigations on pedestrian discomfortglare, especially about the influence of theluminance distribution, the number of sourcesand the temporal changes occurring as thepedestrian walks along the path.

There is no consensus in previous workabout the effect of the number of sources ondiscomfort glare. Schmidt-Clausen andBindels7 tested multiple glare sources veryclose to each other at low eccentricities (i.e.small angle between the line of sight of thedriver and the light sources) and predictedthat discomfort glare increases with thenumber of sources (additivity). In assessingroad lighting installations, de Boer andSchreuder8 found an increase of þ0.5 on thede Boer scale, when the number of visiblelight sources was about halved (from 11 to 6luminaires) for a driver’s point of view.Bullough et al.9 addressed outdoor lightingindependently of the participant’s status(driver or pedestrian) and found no effect ofa second distant source, when this source is atleast 9 degrees away from the line of sight.Zhu et al.10 found no effect of the number ofsources on discomfort glare either; they used avertical panel of LEDs to simulate roadlighting in the laboratory. Pedestrian studiesalso suggest that observers feel the discomfortglare due to one luminaire at a time,5 andjudge the lighting quality with respect to theluminaire closest to them.11

In order to quantify the quality of alighting installation in terms of comfort,one would like to predict the discomfortglare level. For that purpose, variousmodels7–9,12–14 have been developed mostlyin laboratories (i.e. with simplified and

2 C Villa et al.



controlled visual scene photometry) to predictdiscomfort glare from automotive7 and/oroutdoor lighting.9,12–14 These models predictthe discomfort glare level on the de Boerscale8 from the photometric and geometriccharacteristics of the visual scene. Thesemodels estimate the discomfort glare experi-enced by a static observer, with a fixed gazedirection.

As pointed out by Miller and McGowan,5

discomfort glare is somehow different forpedestrians compared to drivers. Althoughpedestrians and drivers share the same visualsystem, their visual experience (includingglare) differs in some respect. First, they donot perform the same visual tasks, which isknown to impact visual performance15 as wellas discomfort glare.16 Second, their regions ofinterest (in which information to perform thetask is collected) are different; driver gazehas been studied intensively in the last20 years,17,18 while pedestrian gaze allocationstill is in its infancy.19 This impacts both thesource eccentricity (which in turn impacts

glare) and the adaptation luminance, which isa proxy for the luminance in the area wherepeople stare, that is, the Area of Interest(AoI). Also, a typical driver gaze is muchmore constrained to look towards specificAoI (tangent point, vanishing point, otherroad users, etc.) than a pedestrian’s gaze,which is expected to increase the discomfortglare.20 Another issue is that some level ofoverhead glare is experienced by pedestrians,not by drivers, because of the cut-off of thecar roof. Finally, the speed of movement isdifferent and dynamic effects on glare may bedifferent.

The models for predicting discomfort glarelevel have each been developed using datafor a given range of variables (see Table 1).Many of these, especially the range of testedeccentricities, do not cover all the valuesappropriate for a pedestrian situation. Inaddition, these models have not been usedso far for outdoor lighting from a pedestrianpoint of view, and therefore deserve to beassessed in such a scenario.

Table 1 Range of values for variables used for modelling discomfort glare in previous work.7–9,12–14 The last columnshows the experimental data used in this paper

Geometric andphotometricvariables

GCMGlare ControlMark (datafrom De Boerand Schreuder,1967)8

Sch74(Schmidt-Clausenand Bindels,1974)7

Lin14(Lin et al.,2014)13

Lin15(Lin et al.,2015)14

Bul08/Bul11(Bulloughet al., 2008;Bulloughet al., 2011)9,12

Our pedestriancase study

Sourceeccentricity

6 or 11 poles;scale 1:50

0.16–1.58 108, 158, 208 28, 48, 88, 168 n.a. Lum1: 22–628Lum2: 9–208

Sourcephotometry

I80: 100–5000 cd(five stages)

10–3 lx to 10–1 lx 104 cd/m2 101 lx to 102 lx(20–300 lx)

101 lx to 102 lx104 cd/m2 to105 cd/m2

Es¼ 10–2 lx

100 lx to 101 lx(4–26 lx)104 cd/m2

I80: 12–200 cdI88: 0 cd

Backgroundphotometry

0.34–6.7 cd/m2

(four stages)10–3 cd/m2 to

10–1 cd/m2100 cd/m2 to

101 cd/m20 lx, 10 lx,

200 lx(0.01–1.6 lx)

10–2 lx to100 lx

10–2 lx to 100 lx10–2 cd/m2 to10–1 cd/m2

Source size F: 0.007 m2,0.07 m2 or0.25 m2

n.a. 10–5 sr n.a. (distance1.5 m)

n.a. (distance3–20 m)

10–5 sr to 10–3 srF: 0.0075 m2

or 0.08 m2,0.15 m2 or0.24 m2

Discomfort glare experienced by pedestrians 3



1.2. Discomfort glare models

Previous work on lighting design investi-gated the link between geometric and photo-metric characteristics of the illuminated sceneand discomfort glare. The main photometric/geometric factors identified in the literatureare the source luminance, the backgroundor adaptation luminance,21 the vertical illu-minance at the eye of the observer,22 thesource eccentricity (angle between the obser-ver’s gaze direction and the source) and thesolid angle of the source from the observerpoint of view. Based on these factors, variousmodels have been proposed.7,9,12–14,23 In out-door lighting, most of them predict a level ofdiscomfort glare on a subjective scale, suchas the de Boer scale.8 The de Boer scale hasbeen the most widely used scale in previousoutdoor lighting studies7–10,13,14,21,22,24–26 andtherefore was implemented in our experimentin order to compare predictions from differ-ent models to our experimental data. This iswhy our overview below only focuses on themodels predicting a de Boer rating. The rangesof values of the variables employed in theoriginal papers to fit each of these models areshown in Table 1. A more comprehensivereview of discomfort glare models was con-ducted by Gibbons and Edwards.27

De Boer and Schreuder8 introduced amodel to determine the level of discomfortglare in road lighting installations usingthe de Boer scale: the Glare Control Mark(GCM) index.8,28 The model was developedwith data collected from participants whojudged the discomfort glare from a driver’spoint of view moving along a 300mlighting installation, simulated on a 1:50scale dynamic mock-up. It accounts for thelight distribution (cut-off) by means of lumi-nous intensity levels in two directions (808and 888 from the vertical), the luminoussurface, the average road luminance, andthe number of sources (height and polespacing), see equation (1). In this experiment,the authors found no impact of the

non-uniformity of the road luminance ondiscomfort glare.

GCM¼ 13:84� 3:31 log I80þ 1:3 logI80I88

� �0:5

� 0:08 logI80I88þ 1:29 logFþ 0:97 logLb

þ 4:41 logh0 � 1:45 logp

ð1ÞSchmidt-Clausen and Bindels7 introduced

a model to predict the discomfort glare on thede Boer scale due to automotive lighting(the model is hereafter referred to as Sch74).It takes as input the source vertical illu-minance at the observer’s eye and the adap-tation luminance (which is taken as theaverage luminance of the road surface);see equation (2). As in the GCM model, theadaptation luminance is not easy to estimate.Moreover, as in all models with sourceeccentricity, glare is undefined when lookingdirectly towards the light source (�1 on thede Boer scale).

Sch74 ¼ 5:0� 2:0 log10

�Xi

Eli

0:003� 1þffiffiffiffiffiffiLa

0:04

q� ��0:46i

0BB@

1CCAð2Þ

In order to overcome these limitations,Bullough et al.9 introduced an index foroutdoor lighting, predicting the discomfortglare on the de Boer scale from three verticalilluminance levels at the observer’s eye (themodel is hereafter referred to as Bul08), whenlooking directly at the source. These illumin-ance levels are respectively associated to theobserved light source (El), the surround (Es)and the ambient term (Ea) (see DG inequation (3)); the total vertical illuminanceEtot at the observer’s eye is the sum of El,Es and Ea. To develop their model,

4 C Villa et al.



Bullough et al. carried out a set of experi-ments in the laboratory and outdoor withmetal halide or mercury vapour lamps infront of the observers. Later, Bullough et al.12

highlighted the limits of their previous modeland proposed a model extension (Bul11 inequation (3)), which includes as input thesource luminance Ll when the source subtendsmore than 0.38 of visual angle (the modelis hereafter referred to as Bul11).

DG ¼ log10 El þ Esð Þ þ 0:6 log10El

Es

� �

� 0:5 log10 Eað Þ

Bul08 ¼ 6:6� 6:4 log10 DGð Þ

Bul11 ¼ 6:6� 6:4 log10 DGð Þ

þ 1:4 log1050 000

Ll

� �

8>>>>>>>>>>>><>>>>>>>>>>>>:

ð3Þ

More recently, Lin et al.13 introduced amodel derived from laboratory data obtainedwith a 3000K LED source located 2m aheadof the observers (the model is hereafterreferred to as Lin14, see equation (4)). Themodel was also tested in the laboratory with a6500K LED source and outdoors with a5000K LED source. The authors found anoffset in average judgments related to thecorrelated colour temperature, the discomfortglare being higher for higher correlated colourtemperatures, but they did not introduce thisfactor into their model.

Lin14 ¼ 3:45� log10ðLl !Þ

2:21

L1:02b �1:62

!ð4Þ

Lin et al.14 proposed another model in 2015(the model is hereafter referred to as Lin15,see equation (5)). The Lin15 model is basedon the illuminance due to the source and

background illuminance. Unlike Lin14, theluminance and the solid angle of the sourceare not needed. Moreover, the Lin15 modelwas based on data collected with a largerrange of background photometry (0 lx, 10 lxand 200 lx) than the Lin14 model (around 1–10 cd/m2), see Table 1. The main differencebetween these models is the value of theconstant which allows for rescaling the logmagnitude to the de Boer scale (3.45 in Lin14vs. 7.09 in Lin15): Lin14 predicts higherdiscomfort glare than Lin15.

Lin15 ¼ 7:09� log10E2:21l

E1:02b �1:62

!ð5Þ

1.3. Model implementation: State of the art

These models were mostly developed inlaboratory conditions with uniform back-grounds and light sources. In heterogeneousreal-world scenes, evaluating luminancevalues requires some choices. Previous workassessing these models13,29 has been done bylaboratory experiments and followed themethod described in the original papers.Very few studies have implemented thesemodels outdoors.30,31

1.3.1. Background luminanceWhen the visual scene is complex, the

background luminance depends on the geo-metric definition of the background area andon the computational method employed toreduce a luminance distribution into a singleluminance value. In previous work,7,32 thebackground luminance is often defined as theaverage luminance of the road surface.However, drivers and pedestrians look atspecific and different AoI. The adaptationluminance roughly reflects the luminance inthese areas of interest, and thus they areexpected to be different. More specifically, theAoI of pedestrians is expected to be larger,and farther away from the light sources and




their gaze is expected to be less constrained.20

Researches are in progress3,33 to understandthe shape, size and location of each roaduser’s adaptation area,34,35–37 and someresearches focus on pedestrians, investigatingpedestrians gaze patterns with eye-track-ers.3,33,34,37 Davoudian et al. found thatpedestrians spend 40–50% of the time lookingat the footpaths.33 Recently, Uchida et al.introduced a simulation method to computeperipheral adaptation luminance in an area ofmeasurement (AOM) in mesopic outdoorscenes, from a static luminance distributionand an eye movement distribution.35 Theauthors tested various AOM sizes on theroad surface with drivers’ or pedestrians’ eyemovement distributions, and they comparedthe simulation with simpler proxies, such asthe average luminance in the AOM and theaverage luminance in the whole image. Theyfound that the adaptation luminance could bepredicted with the average luminance in theAOM, especially when eye movements are notlarge. For pedestrians, the footpath surfaceaverage luminance is a fair predictor of theadaptation luminance, except when manyhigh luminance sources are in the surround-ings. Cengiz et al. assumed a disc-shapedvisual adaptation field, centred on the meangaze direction.34 They computed the averageluminance in circular fields extending 18 to208 from the line of sight of drivers.Maksimainen et al. compared average meso-pic luminance values in the AOM, withvarious sizes and shapes (road surface, circu-lar, elliptical). For each street illustrationsemployed in this work, they obtained astandard deviation of the average luminancescomputed in various AOM of around 21%.36

To implement discomfort glare models forpedestrian applications, Kohko et al. testedaveraging the luminance on the road surfaceand in a circular visual field (208 or 408), ‘butnone of them showed acceptable fitting’.30

Finally, they used a constant backgroundluminance for all stimuli.

In some models,7,14 the backgroundphotometry is measured by switching off thelight source. However, in a pedestrian casestudy, urban lighting contributes to bothglare and adaptation (indirect lighting).Therefore, in such a case study, the indirectcomponent of the sources needs to betaken into account in the background termestimation.

1.3.2. Light source photometrySome models use the source vertical illu-

minance as input. In this approach, the wholelantern defines the light source. The larger thesource region, the higher the illuminance andtherefore the discomfort glare.

Other models use the product of the sourceluminance and its solid angle as input,but they do not specify the region whichshould be considered and how to computeluminance and solid angle of a LED source.Indeed, LED light sources can be non-uniform compared to other common sources(e.g. High Pressure Sodium, Metal-Halide,Incandescent, Fluorescent) because they asso-ciate many very small sources (LED chips) ona LED module, fitted with optical compo-nents. The non-uniformity of LED luminairesmay contribute to discomfort glare.6,38–40

From a luminance map, a computationalmethod has to be chosen to provide a singlevalue as input to the models. In most previouswork, the average luminance is used as theinput value for the source luminance. Theaverage may not be the most relevant way toestimate the luminance of a non-uniformsource; other estimates may also be con-sidered, such as the maximum luminance or aweighted sum.40 In addition, whatever themodel, the predicted discomfort glareincreases with the source luminance as wellas with the source solid angle. Therefore, insome cases, model predictions are sensitive tothe geometric definition of the light source inan unpredictable way. For example, in urbanLED lanterns, the LED module is usually

6 C Villa et al.



surrounded by optical components (e.g. clearglass, frosted glass). In this case, the sourcearea may be described in at least twoways: either the whole lantern or restricted tothe LED module (see Figure 3(b)). The aver-age luminance is higher in the LED modulearea than in the lantern area, whereas the solidangle of the LED module area is lower thanthe solid angle of the lantern area.

1.3.3. Implementation of Bullough’s modelThis model addresses the discomfort glare

from each light source in the lighting instal-lation. Bullough’s model needs as inputs thedirect source illuminance El, the illuminancefrom the area surrounding the source Es andthe ambient illuminance Ea. Ea is the verticalilluminance measured when the sources areswitched off. El cannot be measured directly.Bullough et al.9,31 proposed to deduce El fromthe total vertical illuminance (Etot) and thevertical illuminance measured when the directcomponent of the source is hidden from theilluminance meter. It is not possible to ‘cover’the light source up a 4m high pole asproposed by Bullough et al.9 in a laboratorysetup. This is why, Brons et al.31 proposed toshield the illuminance meter with a circularbaffle, to remove the direct illuminance of thesource El from the measurement and deduce itby computation: El¼Etot – Ebaffle.

The illuminance from the area surroundingthe source Es is not easy to define in acomplex scene. It is defined by Bulloughet al.9 as the remainder found when subtract-ing (ElþEa) from Etot. For pedestrian appli-cations, the same authors31 defined thesurrounding illuminance Es as the differencebetween the illuminance measured whenthe lighting installation is switched on exceptthe tested luminaire, and the sum of the directilluminance levels from all other luminaires(Es includes the ambient illuminance, unlikeBullough’s (2008) definition9).

Whatever the computational definition, weunderstand that the surrounding illuminance

is introduced in the model to quantify theadaptation illuminance due to the indirectcomponent of the lighting installation.

1.3.4. Decomposition of direct and indirectcontributions from the light source

We have seen above that models based onilluminance (i.e. Sch74, Bul08, Lin15)7,9,14

require the estimation of the direct illumin-ance from the light source, as well as theindirect illuminance due to light reflections inthe environment. Unfortunately, measuringthe vertical illuminance at the observer’s eyemerges these variables. The measurement canbe split into two parts, for example, using abaffle on the illuminance meter.31

The two components of the vertical illu-minance may also be computed from lumi-nance values. Indeed, the direct illuminancedue to any region of a luminance map,captured with an imaging luminance meas-urement device (ILMD), can be calculatedfrom pixel luminance, solid angle and eccen-tricity, assuming that the gaze is at the centreof the map.

Finally, the background vertical illumin-ance from a given region may also beestimated from the average luminance overthis region assuming the reflecting surfaces tobe Lambertian. This assumption is supportedby de Boer, who found no influence of theroad luminance non-uniformity8 and a directcorrelation between the average illuminanceand average luminance of a lightinginstallation.41

1.4. Objectives

A pilot outdoor study was conducted toinvestigate pedestrian discomfort glare incomplex visual scenes (as opposed to con-trolled laboratory experiments with uniformbackground and source luminance). Theobjectives of this work were twofold:

1) To investigate a protocol for assessing thepedestrian discomfort glare from LEDurban lighting;




2) To implement and test the relevance ofpreviously developed models for predict-ing the pedestrian discomfort glare fromLED urban lighting.

The second objective was limited to thosemodels which predict a score on the de Boerscale.

Two experimental protocols were tested.The ‘Static’ protocol involved collecting judg-ments under various lighting installations,from specific positions and gaze directions.The ‘Dynamic’ protocol involved collecting aglobal judgment of discomfort glare afterwalking under the poles. This protocol inte-grates temporal variations in the glare experi-ence during the walk. Installations werejudged with one or two luminaires switchedon in order to study the influence of thenumber of sources. In order to collect robustdata, some variability was introduced in thevisual environment by using four different

lanterns. Viewpoints and gaze directions werealso varied.

Mean glare ratings were compared tovarious model predictions in order to evaluatetheir relevance for a pedestrian application.This comparison was based on the RootMean Square Error (RMSE) and on theSpearman correlation (R2). RMSE quantifiesa global error between the participants’ meanratings and the model predictions. R2 quan-tifies the ability of the model to predictordinal results (ranking).

The remainder of the paper is organized asfollows. Section 2 is dedicated to the materialand methods of our experiment and to theimplementation issues of the various models.Section 3 presents the psychovisual results.Section 4 presents the evaluation of eachpredictive model for a pedestrian application.Findings are discussed in Section 5 whileconclusions and future work are presented inSection 6.

14m

2.5m

4.5m

6.5m

8.5m

2.5m

13.3m

FACE

Lum 2

Lum 1

Targets 10° Target 0°

Figure 1 Experimental setup with both luminaires switched on (left) and illustration of the experimental settings(right). The geometrical shapes (circle, square, triangle and star) correspond to the viewing positions in the ‘Static’protocol; targets 08 and 108 are used to control the participants’ gaze direction

8 C Villa et al.



2. Material and methods

A psychovisual experiment was carried outon a test track in Les Andelys, France (seeFigure 1) in order to collect discomfortglare ratings (on the de Boer scale) underdifferent LED urban lighting installations.The street section was approximately 100mlong and 5.5m wide. The road surface reflec-tion properties was classified according to theCIE standards:42 the road surface was classR3 (with a lightness of Q0¼ 0.07)42 and W3(Q0¼ 0.02) for a wet road.

2.1. Description of the urban lighting

installations

The main photometric and geometric char-acteristics of the lighting installations areprovided in Table 2. The spacing betweenthe poles was 13.3m and the height of theluminaires was 4m, which is consistent withthe European standard NF-EN1320143 (typeS2) (according to DIALUXa2 computations).Urban lighting poles are usually 3–8m inheight, and the spacing is defined in order toensure an average illuminance of between10 lx and 15 lx and a uniformity level U0 (i.e.Minimum/Average) higher than 0.4 on theroad. Four lanterns were tested (A, B, C andD in the following). The correlated colour

temperature was 4000K and the CRI Ra¼ 70for all lanterns. The luminous flux was chosenso as to ensure similar average horizontalilluminance under each installation.

2.2. Panel

Thirty-three participants were involved inthis study (14 men, 19 women), with agesbetween 21 and 60 years (M¼ 42.2,SD¼ 13.7). Their vision was tested with anErgoVision device (Essilor) prior to theexperiments to check binocular visual acu-ity in photopic and mesopic conditions,contrast sensitivity and time of recoveryafter glare. Table 3 provides the character-istics of the panel. Participants read andsigned an informed consent form; theyreceived a financial compensation after theexperiment.

2.3. Experimental setting and protocol

2.3.1. Static protocolIn order to investigate urban lighting, the

experiment was based on common use condi-tions, rather than on a factorial plan withphotometric/geometric variables (e.g. eccen-tricity, luminance, etc.). Participants wereasked to rate the discomfort glare on thede Boer scale8 from four positions illustratedin Figure 1: 8.5 m (circle), 6.5m (square),

Table 2 Lighting characteristics. Average illuminances were measured on a 13.3 m�5.5 m grid, according to the NF-EN1320143 standard. In condition 2 S (two luminaires), Lum1 and Lum2 were located on the left corners of therectangular grid.43 In condition 1 S (one luminaire), Lum1 was located in the middle of the left side of the rectangulargrid

Lantern A B C D

Optics Cover opalescentglass

Cover frostedglass

Cover clear glass Flat glass

LED module LED HighPower 18 LEDs

LED HighPower 18 LEDs

LED MidPower30 LEDs

LED HighPower 12 LEDs

Lantern luminous flux 3000 lm 2500 lm 2500 lm 2000 lmLuminous efficiency 70 lm/W 79 lm/W 84 lm/W 85 lm/WAverage illuminance

in condition 1 S[min;max]

12 lx [2 lx; 57 lx] 11 lx [2 lx; 28 lx] 11 lx [4 lx; 40 lx] 12 lx [2 lx; 34 lx]

Average illuminancein condition 2 S[min;max]

13 lx [3 lx; 58 lx] 13 lx [3 lx; 29 lx] 14 lx [7 lx; 42 lx] 14 lx [3 lx; 34 lx]




4.5m (triangle) and 2.5m (star) ahead ofthe first pole (the closest to the observers).The de Boer scale as follows was translatedinto French:

� 1 Unbearable (Insupportable)� 2� 3 Disturbing (Derangeant)� 4� 5 Just admissible (Supportable)� 6� 7 Satisfactory (Acceptable)� 8� 9 Unnoticeable (Negligeable)

They were asked to look at one of two targets(see Figure 1). The first target was straightahead 14m away from the first luminaire(Lum1 in Figure 1); the second target was also14m away from Lum1 but 108 to the left (onefor each of the four positions).

In order to study the impact of the numberof luminaires on discomfort glare, each lan-tern was judged with one (1 S) or two lumin-aires (2 S) switched on. In condition 1 S, theluminaire closest to the observer (Lum1)was switched on; in the condition 2 S, bothluminaires were switched on. Sixty-fourratings (4 lanterns� 4 positions� 2 gazedirections� 2 numbers of luminaires) werecollected in the Static protocol for eachparticipant.

2.3.2. Dynamic protocolIn the dynamic, more realistic condition,

participants were asked to walk from the firststatic position (8.5m from Lum1) to thetarget ahead, in front of them (target 08 inFigure 1), looking continuously at this target.The gaze direction was thus parallel to theroad axis. Then, they were asked to rate onthe de Boer scale the global discomfort glarethey felt during the walk. Each participantmade eight such ratings (4 lanterns� 2 num-bers of luminaires).

2.4. Experimental design, procedure

and statistical analysis

In the ‘Static’ protocol, the four experi-mental factors are the lantern (A, B, C or D),the viewpoint (8.5m, 6.5m, 4.5m or 2.5mfrom Lum1), the gaze direction (08, 108 fromthe street axis) and the number of luminaires(one or two). In the ‘Dynamic’ protocol, thetwo experimental factors are the lantern andthe number of luminaires.

Participants came two by two. Each par-ticipant was handed a booklet where theindividual order of the experimental condi-tions was specified. Lanterns and conditions1 S/2 S were presented in a balanced randomorder. For each lantern, the order of theStatic and the Dynamic tasks was reversedbetween condition 1 S and 2 S for each par-ticipant. For the Static Protocol, based on abalanced random experimental plan, pos-itions and directions were presented in a

Table 3 Characteristics of the panel

Panel (%)

GenderMale 42Female 58

Age535 33.335–50 33.3450 33.3

Corrected visionYes 60No 40

Visual acuity [mesopic in brackets]12/10 70 [0]�10/10 79 [6]�8/10 94 [46]�6/10 94 [70]�5/10 100 [70]�4/10 100 [97]�2/10 100 [100]

Contrast sensitivityVery good (0 err.) 48Good (1–2 err.) 12Medium (3–4 err.) 3Bad (5–9 err.) 15Very bad (410 err.) 21

Recovery time after glare525 s 6125–50 s 33450 s 6

10 C Villa et al.



Tab

le4

Ph

oto

me

tric

cha

ract

eri

stic

so

fth

est

imu

li

Ecc

en

tric

ity

LM

TB

52

0K

on

ica

-Min

olt

aC

A-2

00

0A

Ca

no

nE

OS

70

Dþ

Ph

oto

lux

La

nte

rnP

osi

tio

nD

ire

ctio

n

� Lu

m1

(8)

� Lu

m2

(8)

Eto

t

(lx

)

1S

Eto

t

(lx

)

2S

LR

Z

(cd

/m2)

1S

LD

Z

(cd

/m2)

1S

LR

Z

(cd

/m2)

2S

LD

Z

(cd

/m2)

2S

Eb

lack

(lx

)

LL

ED

(cd

/m2)

Lu

m1

LL

ED

(cd

/m2)

Lu

m2

LL

an

tern

(cd

/m2)

Lu

m1

LL

an

tern

(cd

/m2)

Lu

m2

Lm

ax

(cd

/m2)

Lu

m1

Lm

ax

(cd

/m2)

Lu

m2

EL

an

tern

(lx

)

(Lu

m1

)

EL

an

tern

(lx

)

(Lu

m2

)

A8

.5m

08

22

.69

.25

.25

.60

.19

70

.09

60

.34

90

.24

70

.05

01

.49

Eþ

04

5.4

0Eþ

03

3.4

0Eþ

03

1.7

9Eþ

03

2.5

5Eþ

04

8.6

0Eþ

03

4.6

0.3

4

108

30

.41

7.7

4.8

5.1

0.1

13

0.0

56

0.2

05

0.1

20

.06

04

.20

.33

6.5

m08

28

.51

0.1

7.9

8.4

0.1

44

0.0

55

0.3

20

.21

70

.04

71

.29

Eþ

04

6.0

0Eþ

03

4.1

0Eþ

03

1.9

8Eþ

03

2.8

4Eþ

04

1.0

8Eþ

04

7.7

0.4

7

108

36

.31

8.6

7.1

7.7

0.0

99

0.0

50

.21

10

.12

20

.06

06

.60

.41

4.5

m08

38

.21

1.2

12

.31

3.1

0.1

08

0.0

42

0.3

17

0.2

14

0.0

47

1.3

0Eþ

04

7.7

0Eþ

03

3.9

0Eþ

03

2.1

7Eþ

03

3.5

0Eþ

04

1.1

6Eþ

04

12

.20

.60

108

45

.71

9.6

10

.71

1.5

0.0

85

0.0

43

0.2

26

0.1

24

0.0

53

10

.60

.55

2.5

m08

54

.71

2.6

15

.51

6.2

0.0

72

0.0

34

0.3

08

0.2

06

0.0

50

1.1

0Eþ

04

9.8

0Eþ

03

4.3

0Eþ

03

2.9

7Eþ

03

3.3

0Eþ

04

1.3

0Eþ

04

13

.60

.79

108

62

.12

0.9

11

.21

2.7

0.0

66

0.0

44

0.2

26

0.1

29

0.0

53

10

.20

.80

B8

.5m

08

22

.69

.25

.55

.70

.22

10

.11

70

.34

10

.23

20

.04

71

.39

Eþ

04

2.5

4Eþ

03

4.6

0Eþ

03

1.0

8Eþ

03

4.7

0Eþ

04

4.7

0Eþ

03

5.4

0.2

0

108

30

.41

7.7

5.0

5.3

0.1

16

0.0

68

0.1

88

0.1

20

.05

74

.70

.16

6.5

m08

28

.51

0.1

8.3

8.7

0.1

71

0.0

71

0.3

15

0.1

99

0.0

50

1.3

6Eþ

04

2.2

9Eþ

03

4.7

0Eþ

03

9.5

0Eþ

02

5.9

0Eþ

04

4.2

0Eþ

03

8.4

0.2

6

108

36

.31

8.6

7.6

7.8

0.1

15

0.0

65

0.2

05

0.1

24

0.0

57

7.4

0.2

3

4.5

m08

38

.21

1.2

12

.51

3.0

0.1

31

0.0

51

0.3

09

0.1

83

0.0

47

1.6

9Eþ

04

4.1

0Eþ

03

4.4

0Eþ

03

1.7

3Eþ

03

5.6

0Eþ

04

7.7

0Eþ

03

11

.80

.37

108

45

.71

9.6

10

.71

1.3

0.1

04

0.0

59

0.2

16

0.1

22

0.0

53

10

.40

.37

2.5

m08

54

.71

2.6

13

.81

4.4

0.0

99

0.0

44

0.2

97

0.1

74

0.0

47

1.2

2Eþ

04

7.5

0Eþ

03

4.2

0Eþ

03

1.8

1Eþ

03

4.9

0Eþ

04

1.3

5Eþ

04

12

.80

.63

108

62

.12

0.9

10

.61

1.6

0.0

89

0.0

58

0.2

14

0.1

24

0.0

53

9.1

0.6

0

C8

.5m

08

22

.69

.22

0.9

22

.30

.31

40

.23

70

.63

70

.54

20

.05

71

.29

Eþ

05

4.9

0Eþ

03

1.6

9Eþ

04

1.2

7Eþ

03

2.5

5Eþ

05

1.2

0Eþ

04

19

.10

.32

108

30

.41

7.7

19

.82

1.0

0.1

59

0.1

04

0.3

34

0.2

05

0.0

82

18

.40

.31

6.5

m08

28

.51

0.1

24

.92

6.5

0.2

07

0.1

10

.57

70

.45

40

.05

71

.26

Eþ

05

8.8

0Eþ

03

1.5

0Eþ

04

2.7

1Eþ

03

2.2

8Eþ

05

2.8

4Eþ

04

25

.20

.66

108

36

.31

8.6

22

.62

5.9

0.1

48

0.1

0.3

76

0.2

20

.07

92

1.9

0.6

6

4.5

m08

38

.21

1.2

22

.52

5.6

0.1

61

0.0

80

.60

30

.44

20

.06

09

.30

Eþ

04

3.7

0Eþ

04

8.4

0Eþ

03

5.3

0Eþ

03

1.6

4Eþ

05

6.3

0Eþ

04

24

.21

.88

108

45

.71

9.6

20

.02

2.7

0.1

44

0.0

98

0.4

36

0.2

38

0.0

82

20

.71

.77

2.5

m08

54

.71

2.6

15

.41

9.6

0.1

44

0.0

69

0.6

04

0.4

41

0.0

63

5.5

0Eþ

04

6.2

0Eþ

04

5.5

0Eþ

03

8.8

0Eþ

03

9.4

0Eþ

04

1.1

8Eþ

05

16

3.6

0

108

62

.12

0.9

12

.11

6.7

0.1

39

0.0

96

0.4

43

0.2

34

0.0

79

12

.63

.50

D8

.5m

08

22

.69

.26

.86

.60

.25

70

.14

50

.38

30

.27

10

.05

76

.40

Eþ

04

4.0

0Eþ

02

1.2

0Eþ

04

2.3

8Eþ

02

7.8

0Eþ

04

6.1

0Eþ

02

4.5

0.0

16

108

30

.41

7.7

6.0

6.2

0.1

40

.07

90

.21

40

.12

90

.08

54

.20

.01

6

6.5

m08

28

.51

0.1

16

.01

6.1

0.2

05

0.0

83

0.3

67

0.2

22

0.0

60

1.1

6Eþ

05

5.7

0Eþ

02

2.9

3Eþ

04

2.8

7Eþ

02

1.9

7Eþ

05

8.0

0Eþ

02

14

.50

.02

3

108

36

.31

8.6

14

.61

4.7

0.1

40

.07

30

.24

10

.12

70

.08

51

2.6

0.0

27

4.5

m08

38

.21

1.2

21

.82

2.6

0.1

65

0.0

60

.36

50

.21

0.0

60

1.5

2Eþ

05

8.1

0Eþ

02

2.9

8Eþ

04

4.2

0Eþ

02

2.3

7Eþ

05

1.1

7Eþ

03

24

0.0

37

108

45

.71

9.6

20

.52

0.2

0.1

28

0.0

67

0.2

58

0.1

33

0.0

79

23

.20

.04

1

2.5

m08

54

.71

2.6

21

.32

2.4

0.1

16

0.0

42

0.3

46

0.1

81

0.0

60

1.1

3Eþ

05

9.0

0Eþ

02

1.9

4Eþ

04

4.8

0Eþ

02

2.4

6Eþ

05

1.6

5Eþ

03

22

0.0

61

108

62

.12

0.9

18

.32

0.0

0.1

03

0.0

61

0.2

52

0.1

24

0.0

85

16

.90

.06

1




different order to each participant. Theexperimental sessions lasted about 1 hour.A short practice was carried out at the start ofthe session to familiarize the participantswith the experimental protocol and the deBoer scale.

For data analysis, repeated measures anal-yses of variance (ANOVAs)44 were performed(normality was satisfied according to

Kolmogorov–Smirnov and Jarque–Beratests). The Mauchly test was employed totest the sphericity.45 The Greenhouse–Geissercorrection was employed when the sphericitywas not satisfied. The significance thresholdwas set to p¼ 0.05.

2.5. Photometric description of the visual

stimuli

This section describes the photometricmeasurements needed to describe the stimuliand implement the glare models. Table 4summarizes the photometric and geometriccharacteristics of each stimulus. All measure-ments and calculations were made consider-ing an observer with eye height at 1.50m.

2.5.1. Vertical illuminanceFor each stimulus, the vertical illuminance

was measured at 1.50m height with a LMTB520 illuminance meter aimed towards therespective target, and is reported in Table 4. Inaddition, Figure 2 shows the profile of verticalilluminance along the walk for the fourlanterns A–D when only Lum1 is switched on.

35

Distance (m)

A B C D

0 1 2 3 4 5 6 7 8 9 10 11 12

Ver

tical

illu

min

ance

(1x

)

30

25

20

15

10

5

0

Figure 2 Vertical illuminance along the walk in condition1 S, for lanterns A, B, C and D. The four black symbols(circle, square, triangle and star) correspond to the ratingpositions in the static protocol (see Section 2.3.1). The sunsymbol marks the position of pole Lum1

(a) (b)

Lantern area LED modulearea

Figure 3 (a) Areas for the background luminance estimation. DZ (disc zone): Disc 308 in diameter. RZ (road zone),rectangle including the road surface and the targets. (b) Areas for the source luminance estimation: LED module areaand lantern area

12 C Villa et al.



2.5.2. Background luminance and illuminanceBased on previous work (see Section 1.3.1),

we have considered two ways of computing thebackground luminance from luminance maps(see Figure 3(a)). The first method uses theaverage luminance in the road zone (LRZ); thesecond one uses the mean luminance in a disczone 308 in diameter (LDZ) centred at thetarget which the observers were looking at (seeFigure 3(a)). Thirty degree corresponds to thecentral visual field as defined by Bardy et al.46

when studying central and peripheral vision inpostural control while walking. A KonicaMinolta CA2000A imaging photometer witha wide lens (8mm, 1:1.4, FOV:468� 468,980� 980) captured luminance maps of thescenes (see Figure 3(a)). The average back-ground luminance was extracted from thesemaps, using either LRZ or LDZ.

Background illuminance values ERZ andEDZ were computed from LRZ and LDZ,respectively, assuming the road and the targetswere Lambertian surfaces (see Section 1.3.4).

A luminance map was also captured withall the luminaires switched off to characterizean average ambient luminance Lblack. Theilluminance Eblack was deduced from this map(Eblack¼�Lblack).

2.5.3. Source luminance and illuminanceThe source luminance was measured with a

calibrated Canon EOS 70D camera (Sigma4.5mm, 1:2.8; FOV: 1808, 3648� 2432) withfish-eye lens, connected to the Photoluxsoftware. This device makes it possible tocapture a 1808 luminance map of the visualscene. Source areas were manually selected tocompute the solid angle and mean luminanceof each source. Two methods were consideredfor selecting the source areas: the selectionencompassed either the LED module alone(LLED in the following) or the whole lantern(LLantern in the following), see Figure 3(b).

The illuminance due to any area in theluminance map can be computed from theluminance, solid angle and eccentricity from

the centre of the image (which corresponds tothe gaze direction) of the pixels in the area.Vertical illuminance due to the source wasthus computed, leading to two illuminancevalues, respectively ELED and ELantern.

3. Psychovisual results

3.1. Preliminary analyses

A cluster analysis was first carried out tolook for potential groups of observers whoshare an opinion which differs from that ofother groups, as well as potential outliers.Neither was found with hierarchical ascend-ant clustering (conducted with the percentdisagreement distance and average linkage).Therefore, all data were considered in furtherstatistical analyses.

Because of weather conditions at the time ofthe experiment, the road surface was wet for20 participants and dry for 13 participants.A preliminary analysis was conducted to checkif road surface condition had an effect on dis-comfort glare, in which case the data wouldpossibly be split according to the road surfacecondition. Repeated measures ANOVA wasrun on the data from each protocol (Static orDynamic), with the ‘Stimulus’ as an intra-subject factor (64 modalities in the Staticprotocol, eight modalities in the Dynamicprotocol) and the ‘Road surface condition’ asan inter-subject factor (two modalities: dry orwet road surface). No significant effect ofthe road surface condition was found eitherin the Static protocol (Road surfacecondition: F(1,31)¼ 1.20, p¼ 0.28, Road sur-face condition*Stimuli: F(63,1953)¼ 0.44,p¼ 1.00) or in the Dynamic protocol (Roadsurface condition: F(1,31)¼ 1.35, p¼ 0.25,Road surface condition*Stimuli: F(7,217)¼0.64, p¼ 0.72). Therefore, road surface con-dition was not considered in furtheranalyses. Only the luminance maps capturedin dry conditions were considered in furthercomputations.




3.2. Static protocol

Discomfort glare ratings were collected foreach type of lantern, with one or two lumin-aires switched on, from participants in fourdifferent positions with two gaze directions.A repeated measures ANOVA was run on thedata (after Greenhouse–Geisser correction44).The lantern (F(2.45,78.39)¼ 19.02,p50.0001, �2¼ 0.373), the gaze direction(F(1,32)¼ 7.79, p50.05, �2¼ 0.196), thenumber of luminaires (F(1,32)¼ 5.62,p50.05, �2¼ 0.149) and the position(F(2.24,71.57)¼ 5.19, p50.05, �2¼ 0.139),all had statistically significant main effectson discomfort glare. The variation of discom-fort glare with the positions is significantlydifferent depending on the lantern(‘Lantern*Position’: F(3.99,127.68)¼ 16.09,p50.0001, �2¼ 0.335). No other significantinteraction was found (‘Lantern*Direction’:F(2.92,93.37)¼ 1.037, p¼ 0.3796;‘Position*Direction’: F(2.27,72.76)¼ 0.229,p¼ 0.876; ‘Lantern*Number of luminaires’:F(2.50,80.04)¼ 0.87, p¼ 0.460); ‘Number ofluminaires*Position’: F(2.28,73.00)¼ 1.731,p¼ 0.166; ‘Number of luminaires*Direction’:F(1,32)¼ 0.134, p¼ 0.717).

Table 5 shows the mean and standarddeviation values for each factor. The higherthe rating, the lower the discomfort glare.Post hoc Tukey tests44 were conducted inorder to compare the lantern performanceand showed no significant difference betweenLanterns A and B, which produced signifi-cantly less glare than lantern D, which in turn

produced significantly less glare than LanternC. Lanterns A and B, with opalescent andfrosted glass, produced significantly less glarethan the two others, with an ‘acceptable’average rating across positions (see Table 5).Discomfort under Lantern D was judged onaverage ‘acceptable’ 8.5 m from Lum1, but‘just admissible’ nearer (4–6m). Finally,Lantern C which produces the highest lumi-nance (see Table 4) was the most glaring.

According to the post hoc Tukey tests,participants felt less comfortable when look-ing along the road than when looking awayfrom the luminaires. More discomfort wasexperienced at 6.5m and 4.5m from Lum1than at 8.5m or 2.5m. In addition, lightinginstallations were rated as significantlyless disturbing when two luminaires wereswitched on. The situations with the mostglare were at 6.5m and 4.5m from Lum1looking straight ahead, with a single lumin-aire (M¼ 5.95 [1S, 6.5m, 08]; M¼ 5.92 [1S,4.5m, 08]).

3.3. Dynamic protocol

A repeated measures ANOVA (afterGreenhouse–Geisser correction) and post hocTukey tests were conducted. An effect of theLantern was found (F(2.51,80.21)¼ 12.07,p50.0001, �2¼ 0.274). There was no effectof the number of luminaires (F(1,32)¼ 1.30,p¼ 0.2633), however, an interaction wasfound (F(2.60,83.34)¼ 3.97, p¼ 0.010350.05, �2¼ 0.110).

Table 5 Mean values of the glare ratings in the different experimental conditions (standarddeviation values are in brackets)

Factor Lantern Number of luminaires Gaze direction Position

Static protocol A: 6.9 [1.6]B: 6.8 [1.7]C: 5.6 [2.1]D: 6.2 [2.1]

1 S: 6.2 [2.0]2 S: 6.5 [1.9]

08: 6.3 [2.0]108: 6.5 [1.9]

8.5 m: 6.5 [2.1]6.5 m: 6.2 [2.0]4.5 m: 6.2 [1.9]2.5 m: 6.5 [1.8]

Dynamic protocol A: 7.4 [1.5]B: 7.1 [1.6]C: 5.9 [1.9]D: 6.7 [1.7]

1 S: 6.9 [1.8]2 S: 6.7 [1.8]

14 C Villa et al.



According to the post hoc Tukey tests,there is no significant difference betweenLanterns A and B. Lantern A producessignificantly less glare than Lanterns C andD. Lanterns B and D produce significantlyless glare than Lantern C. The effect of theLantern is consistent with the results from thestatic protocol (see Section 3.2). However, asshown in Table 5, the de Boer rating after awalk is higher (þ0.4 in average), suggestingthat pedestrians feel less discomfort afterwalking than when standing in front of astatic scene. The same levels of discomfortglare were found whatever the number ofluminaires, except for the lantern C, whichproduced more glare with two luminaires.

3.4. Photometry and discomfort glare

The vertical illuminance at eye-height isnegatively correlated to the mean ratings onthe de Boer scale (R2

¼ 0.711). The correl-ation is even stronger when the data isrestricted to conditions 1 S (R2

¼ 0.776) or2 S (R2

¼ 0.796). Direct vertical illuminancefrom the source, computed from the wholelantern, is also correlated with glare (condi-tion 1 S: R2

¼ 0.717).Various luminance indices can be com-

puted to describe non-uniform sources: meanluminance values LLED and LLantern (seeSection 2.5.3 and Figure 3) as well asmaximum luminance Lmax (this index doesnot depend on the measurement area).Correlations between these luminance indicesand the mean glare ratings have beencomputed: in condition 1 S, the correlationis stronger with LLED (R2

¼ 0.741) than withLLantern (R

2¼ 0.457); the strongest correlation

is found with Lmax (R2¼ 0.771).

We also investigated the influence ofphotometric variations during the walk inthe dynamic protocol. The differences �Ebetween the maximum and minimum illumin-ances, as well as the mean illuminance alongthe path (up to the first pole in condition 1 S,up to the second pole in condition 2 S) were

computed. Because the source luminance wasonly available at the four positions of thestatic protocol, �L and the mean luminancewere estimated from those four values. Meanratings in the dynamic protocol were nega-tively correlated to �Lmax (1 S: R2

¼ 0.584)and to �LLED (1 S: R2

¼ 0.512) and moreweakly to �LLantern (1 S: R2

¼ 0.359) and to�E (R2

¼ 0.427). The mean of the verticalilluminance Etot (R

2¼ 0.773, 1 S: R2

¼ 0.900,2 S: R2

¼ 0.976), the mean of LLED

(R2¼ 0.549), the mean of the maximum

illuminance (R2¼ 0.677, 1 S: R2

¼ 0.876, 2 S:R2¼ 0.946) and the mean of the maximum

luminance (R2¼ 0.612) on the pathway, were

all negatively correlated with the mean globalratings collected from the dynamic protocol.The strongest correlation was found with themean of the vertical illuminance.

4. Comparisons of discomfortglare models

This section explores the relevance of themodels developed in previous work to predictglare ratings on the de Boer scale, for pedes-trian applications. It also investigates thesensitivity of these models when implementedin complex scenes (heterogeneous light sourceand heterogeneous background). Predictionsfrom the models presented in Section 1.2(Sch74, Lin14, Lin15, Bul11) are compared tomean ratings obtained from our psychovisualexperiment in the static protocol (consideredas the reference).

4.1. Model implementation

The four models have been evaluated withseveral variations in their input parameters.The background was taken as the road zoneor as a 308 circular visual field (see Section2.5.2). Lin14 predictions were computed withthe source taken as the whole lantern or asthe LED Module inside the lantern. Whenthe source vertical illuminance was needed




(Sch74, Lin15, Bul11), it was computed fromthe whole lantern area.

In Bullough’s models,9,12,31 the ambientilluminance Ea is the vertical illuminancewhen all the luminaires are switched off(Ea¼Eblack in Table 4). Our understandingis that the surrounding illuminance Es quan-tifies the indirect illuminance due to thelighting installation. For implementation, Es

was either computed following the subtract-ing method described in the original paper9

(Es¼Etot-(ElþEa)), or defined as ERZ orEDZ, computed from the background lumi-nance maps. Because the angular size of thenearest light source was always found higherthan 0.38, whatever the lantern, the sourceluminance was taken into account in modelBul1112 (equation (3)). According to prelim-inary tests, Bul11 performs better when Ls isset to LLantern than Lmax, and that’s why Ls isset to LLantern in the following.

Lin et al.13 proposed to correct their modelpredictions by offset based on the correlatedcolour temperature of the source (but themodel formulation does not take this variableinto account, see equation (4)). An offsetof �1.5 could apply with 4000K LEDs.However, with such an offset, the predictionswere outside the de Boer scale range, andthat’s why we did not take it into account.

4.2. Relevance of the glare models for

pedestrian applications

Predictions from the glare models arecompared with the experimental data col-lected under a single luminaire in Section4.2.1. The sensitivity of the results to themodel implementation is discussed in Section4.2.2. Finally, the number of sources isconsidered in Section 4.2.3.

4.2.1. Relevance with a single sourceThe GCM model cannot be implemented

in our experiment because the luminousintensity at 888 of the lanterns tested in thestudy is nil. I88 was deduced from road

lighting, with higher poles than in urbanlighting, for the point of view of drivers.Eighty-eight degree corresponds to a viewingdistance of 115m from a 4m-height lumin-aire, which is far beyond where pedestrianslook.37,47 Therefore, this model seems not tobe relevant for pedestrian lighting.

Figure 4 compares predictions from theother models with the mean glare ratings inthe static protocol, for each stimulus, incondition 1 S. For each model, predictionsusing the different computational methodsare presented (e.g. LDZ and LRZ to computethe background luminance in the Sch74model). Table 6 presents the RMSE, whichis a global index of the errors with respect tothe experimental data. The Spearman correl-ation (R2) is also reported.

The Sch74 and Lin14 models obtained aRMSE above 3.75. The Bul11 and Lin15models performed much better, with a RMSElower than 1 in some conditions.

Lin15 is the only model to overestimate thede Boer score. It is also the best one in termsof RMSE, with a RMSE as low as 0.60 whenthe background illuminance is computedfrom the disc zone (Table 6).

Bul11 underestimates the glare ratings, butthe model performs quite well, with a RMSEof [1.1–1.3]. The lowest RMSE is obtained

Table 6 Prediction comparisons: RMSE and Spearmancorrelation R2 between predicted scores on the de Boerscale and mean ratings in the ‘Static’ experiment, incondition 1 S

Model Background Source RMSE R2

Sch74 LRZ 4.40 0.744Sch74 LDZ 4.57 0.751

Lin14 LRZ LLantern 3.94 0.384Lin14 LDZ LLantern 4.23 0.370Lin14 LRZ LLED 3.76 0.375Lin14 LDZ LLED 4.05 0.368

Lin15 ERZ 0.81 0.781Lin15 EDZ 0.60 0.745

Bul11 Es¼Etot�El�Ea 1.27 0.648Bul11 Es¼ERZ 1.13 0.720Bul11 Es¼EDZ 1.30 0.725

16 C Villa et al.



when Es is taken as the mean illuminance inthe road zone.

As illustrated in Figure 4(a), predictionsfrom Sch74 strongly overestimate the discom-fort glare (i.e. underestimate the rating) withRMSE44, on a scale range of 1–9. This modelwas developed for automotive lighting appli-cations and was based on experimental datawere the source eccentricity was small(between 0.168 and 1.58), which is far fromthe range considered in our study (98 to 628).Also, the illuminances were between 10–3 and101 lx in Schmidt-Clausen and Bindels’ experi-ment.7 Therefore, the illuminances experiencedby the pedestrians were from one to threeorders of magnitude higher than those whichfitted the Sch74 model. Our results suggestthat Sch74 model should not be generalizedoutside the photometric range of the originalmodel. While the absolute level is muchunderestimated, Sch74 predictions are morerelevant for ranking (see Table 6, R2

¼ 0.75).

Similarly, Lin14 predictions stronglyunderestimate the ratings (see Figure 4(b),RMSE43.75). Again, it may be due to thefact that the background luminance is outsidethe range of the data used for model fitting(see Table 1). Moreover, correlations are lowwith this model (down to R2

¼ 0.385).As shown in Figure 4, for the most uncom-

fortable stimuli, the mean ratings vary between4 and 6 on the de Boer scale whereas thepredictions, whatever the model, are nearlyconstant. This is discussed in Section 5.

4.2.2. Sensitivity of model predictionsto the implementation choices

Whatever the model (Sch74, Lin14, Lin15or Bul11), the predictions are always higher inthe de Boer scale with the RZ backgroundthan with the DZ background (þ0.2/0.3),because LRZ4LDZ. Therefore, the RMSE islower with LRZ (or ERZ) for models whichunderestimate the de Boer ratings, and higher

(a) Sch74M

odel

pre

dict

ions

9

4 5 6

LDZ

EDZ ERZ

LRZ LDZ , LLED LDZ , LLantern LRZ , LLED LRZ , LLantern

Es=Etot-E1-Ea Es=EDZ Es=ERZ

7 8 9

876543210

Lin14

Lin15

Mean glare ratings

Mod

el p

redi

ctio

ns

9

4 5 6 7 8 9

87654321

Mod

el p

redi

ctio

ns

987654321

Mean glare ratings4 5 6 7 8 9

Mean glare ratings

Mod

el p

redi

ctio

ns

9

4 5 6 7 8 9

876543210

Mean glare ratings

Bul11

(b)

(c) (d)

Figure 4 Model predictions vs mean glare ratings in condition 1 S: (a) Sch74, (b) Lin14, (c) Lin15 and (d) Bul11. For eachmodel, the predictions vary depending on the implementation choices (e.g. LDZ vs. LRZ in Sch74)




for Lin15 which overestimates the ratings.The RMSE between model predictions com-puted with RZ and DZ backgrounds wascalculated and leads to errors of less than the0.3 level on the de Boer scale (Sch74: 0.18,Lin14: 0.30, Lin15: 0.30, Bul11: 0.18).

Lin14 makes different predictions depend-ing on the source definition (photometry andgeometry). This impacts both the sourceluminance (LLED4LLantern) and the sourcesolid angle (WLED5WLantern) but in oppositeways, so that the result is not predictable.When comparing the Lin14 predictions com-puted with LLED and LLantern, the RMSE is0.21 on the de Boer scale.

Considering all these implementation issuesand Figure 4, it is clear that the glarepredictions are more dependent on themodels than on their implementation.

4.2.3. Relevance of the glare models for two sourcesThe Sch74 model is the only one which

provides a way to compute discomfort glarewith several sources in the visual field. Thereis no consensus in previous work about theeffect of multiple sources on discomfort glare(see Section 1.1). Moreover, published modelsdo not address the number of sources in thesame way. De Boer and Schreuder8 found animprovement of þ0.5 on the de Boer scalewhen the number of visible light sources ishalved, which accounts for the number of

luminaires per km in the GCM model.Schmidt-Clausen and Bindels,7 who devel-oped an automotive lighting model with up tofive sources at low eccentricity, assumeadditivity in the logarithm (see equation (6)):each source illuminance and eccentricity isconsidered separately with the same back-ground. Additivity is also assumed in discom-fort glare models for indoor lighting, such asUGR and DGI.23 For outdoor lighting, theBullough et al.’s model predicts the discom-fort glare from one source. The authors alsoshow that a second source does not impactthe glare if it is at least 98 away from the lineof sight; it only impacts the surroundingilluminance. Lin et al.13,14 did not address themultisource issue at all.

In the following, we have considered twocomputational methods to estimate the dis-comfort glare ratings in the presence of twoglare sources:

� Additive method: the discomfort glare fromthe two sources Lum1 and Lum2 is takeninto account, with the background of con-dition 2S. As proposed by Schmidt-Clausenand Bindels,7 the contributions of bothsources were added. To that purpose, wehave adapted Bul08 Lin14 and Lin15 aspresented in equation (6). These formulascan easily be generalized to N sources.

Sch74 ¼ 5:0� 2:0 log10El1

0:003 ð1þffiffiffiffiffiffiLa

0:04

q� ��0:461

þEl2

0:003 1þffiffiffiffiffiffiLa

0:04

q� ��0:462

0BB@

1CCA

Bul08add ¼ 6:6� 6:4 � log10 log10 10DG1 þ 10DG2� ��

Lin14add ¼ 3:45� log10ðLl1!1Þ

2:21

L1:02b �1:621

þðLl2 !2Þ

2:21

L1:02b �1:622

!

Lin15add ¼ 7:09� log10E2:21l1

E1:02b �1:621

þE2:21l2

E1:02b �1:622

!

8>>>>>>>>>>>>>>>>><>>>>>>>>>>>>>>>>>:

ð6Þ

18 C Villa et al.



� Simplified method: discomfort glare fromthe closest luminaire (Lum1) alone is takeninto account, under the background ofcondition 2 S. Lum2 only contributes tothe background.

The predictions of the additive method and ofthe simplified method were compared. Thedifference always ranges between �0.1 and 0(from �0.085 to �0.015 for Sch74, from�0.02 to �0.003 for Bul08, from �0.006 to 0for Lin15, and from �0.002 to 0 for Lin14).Thus, the contribution of Lum2 as a glaresource in our experimental setting is negli-gible: the second luminaire is ‘far enoughaway’.9 The contribution of Lum2 is toincrease the adaptation level and thereforeto decrease the discomfort glare. Suchfindings agree with Bullough et al.9 forurban lighting. Equation (6) allows quantify-ing the cumulative effect of additional glaresource with the models, to see whether theadditional sources have a significant effect onglare or if they are ‘far enough away’ to beneglected.

Taking into account Lum1 alone as a glaresource while increasing the background lumi-nance due to Lum2 results in higher glarerating predictions, which is consistent withthe experimental findings (see Section 3.2).The RMSE and R2 between model predic-tions and mean glare ratings were alsocomputed, see Table 7. They have the sameorder of magnitude as those obtained incondition 1 S (Table 6), and thus the modelshave similar performances with one and twopoles.

5. Discussion

5.1. Findings on pedestrian discomfort glare

A psychovisual experiment was conductedto collect judgements on pedestrian discom-fort glare under various urban lighting instal-lations in static and dynamic conditions.Lanterns with opalescent (Lantern A) and

frosted glass (Lantern B) optics obtained thehighest scores (i.e. produced less glare) in ourexperiment. Moreover, with only one lumin-aire switched on, the vertical illuminance(R2¼ 0.776) and the maximum source lumi-

nance (R2¼ 0.771) were the photometric

characteristics which showed the strongestcorrelation with the subjective data in thestatic protocol. Together, these results con-firm previous findings5,30 which recommendfrosted or opalescent optics for LED lumin-aires to limit maximum luminance andthereby reduce pedestrian discomfort glare.

Discomfort glare was rated as lower whenassessed globally after a walk, compared tostatic ratings. In this case, the strongestcorrelation of the global ratings is obtainedwith the mean vertical illuminance alongthe path (R2

¼ 0.773; 1 S:R2¼ 0.900;

2 S:R2¼ 0.980). A possible interpretation

would be that the discomfort glare reportedafter a walk is based on an average feeling.

In urban lighting, a second luminaire mayproduce glare, but also contributes to anincrease of the adaptation level. With thestatic protocol, we found that glare wassignificantly lower with two luminaires thanwith a single luminaire. With the dynamicprotocol, however, we found that globalratings did not significantly differ between

Table 7 RMSE and R2 between the predictions computedwith the simplified method and the mean glare ratingscollected in condition 2 S

Model Background Source RMSE R2 Spearman

Sch74 LRZ 4.46 0.812Sch74 LDZ 4.61 0.796Lin14 LRZ LLantern 3.87 0.566Lin14 LDZ LLantern 4.09 0.512Lin14 LRZ LLED 3.69 0.543Lin14 LDZ LLED 3.90 0.486Lin15 ERZ 0.87 0.745Lin15 EDZ 0.70 0.733Bul11 Es¼Etot�El�Ea 1.42 0.740Bul11 Es¼ERZ 1.22 0.795Bul11 Es¼EDZ 1.35 0.787




one or two luminaires; the second part of thewalk may have affected the overall judgmentmore than the first part. In terms of visualadaptation, the second part of the walk withtwo luminaires was close to condition 1 S inthe static protocol. The second luminaire wasfound to impact the background, but did notcontribute much to the direct glare. Thisresult is consistent with Bullough et al.9 andwith previous pedestrian studies,5,11 whichfound that pedestrians feel the glare from oneluminaire at a time,5 the closest one.11

However, de Boer and Schreuder8 found lessdiscomfort glare when halving the number ofluminaires, so that further investigations areneeded with a larger number of light sources.

As a conclusion about the protocol, theassessment of discomfort glare due to urbanlighting with a static protocol (which isusually employed) and one luminaire tendsto over-estimate the discomfort glare for apedestrian walking in the street. It is somekind of a ‘worst case’ as far as discomfortglare is concerned.

5.2. Predictions of pedestrian discomfort glare

5.2.1. Performance of the modelsExisting discomfort glare models have been

investigated, and their predictions were com-pared with mean ratings considered as areference. The main goal was to checkwhether these models could be applied in areal-world situation. Three models werefound to overestimate the discomfort fromglare, while the Lin15 model underestimatesit. With RMSE higher than 3.7 on a scale withnine levels, Sch74 and Lin14 models are notable to predict the mean glare ratings of ourexperiment, which suggests that these modelsshould not be generalized too far away fromtheir original data range. The illuminancesemployed to fit the Sch747 model werebetween 10–3 lx and 10–1 lx, much lower thanthe levels expected in pedestrian lighting; thebackground luminance of the case study isone order of magnitude lower than the range

employed to develop the model Lin14.13

Bul11 and Lin15 perform better withRMSEs lower than 1.4 and 0.9, respectively,and could be reasonably employed to predictpedestrian glare ratings. Thus, the two modelsbased on illuminance only performed betterthan the two others. This is consistent withthe strong correlation between the meanratings and the mean vertical illuminance(R2¼ 0.711).

In his review of Bullough et al.’s paper,9

Boyce highlights that the Bul11 model isrestricted to zero eccentricity, and suggeststhat the relevance of such a model for streetlighting applications, where drivers/pedes-trians do not usually look straight at thesources, remains unknown. On the otherhand, Bullough et al.20 found that discomfortglare assessments collected in a fixed gazeexperiment were highly correlated with thosecollected in a free gaze experiment. The lowRMSE (less than 1.4) found on the de Boerscale suggests that Bul11 could be generalizedto other eccentricities.

However, none of the models succeeded inpredicting the discomfort glare variationsamong those conditions which produced thestrongest glare. As these conditions (LanternC: from 8.5m to 4.5m; Lantern D: from 6.5mto 2.5m) also provided the highest sourceluminance values (see Table 4), the poorperformance may result from the underesti-mation of the source luminance due tomeasurement limitations. Therefore, effortsshould be made to improve the measurementof high luminances and contrast beforecriticizing the models.

5.2.2. Direct and indirect lightingBullough et al.9 made a distinction

between surrounding illuminance and ambi-ent illuminance, which are both indirect asopposed to direct illuminance from a glaresource. The amount of light at the observer’seye can be split into two components: thedirect one is due to the light sources and the

20 C Villa et al.



indirect one to reflections on the surfaces inthe scene. Glare happens when the directcomponent is much higher than the indirectcomponent. We have investigated the samemodel formulation (equation (3)) with onlytwo terms: the direct component Edirect¼El,and the indirect component Eindirect¼Es¼Ea

defined as the illuminance of the backgroundwhen the luminaires are switched on, whichis supposed to describe the backgroundadaptation level (Dir/Ind in the following).Table 8 shows the RMSE and Spearmancorrelations of those predictions and themean glare ratings. This model performsslightly better than Bull11 (see Tables 5 and6) although the Spearman correlation R2 islower.

5.2.3. Implementation of the modelsThe models using the source luminance as

input require arbitrary assumptions about thespatial definition of the sources, which is acurrent issue for LED lighting because oftheir strong non-uniformity.38–40,48 We haveconsidered average luminance over the wholelantern and over the LED module only, whichled to a small difference (RMSE¼ 0.21 on deBoer scale) with Lin14 predictions. But forpedestrian applications, according to Section3.4 and previous work,5,30 one of the bestcorrelated photometric descriptions of thesource is the maximum luminance. One maywonder whether the average luminance isthe most suitable variable to characterizediscomfort glare from a non-uniform lumin-aire. Kohko et al.30 compute an ‘efficient

luminance’, first proposed by Tashiro et al.,40

which is a weighted sum of the spatialdistribution of the source luminance, and touse it as input for the source luminance in therelevant glare models. Future work is neededin order to investigate the relevance of thisindex for glare prediction.

Two background zones were tested in themodels’ implementation. The RMSEs betweenmodel predictions obtained with either methodwere lower than 0.30 on the nine-point de Boerscale. Therefore, the adaptation zone does notneed to be precisely defined if the luminanceshave the same order of magnitude over thewhole background. However, our road zonewas large compared to the one tested in otherwork35 and the influence of the adaptationregion definition deserves a complete paperitself, as seen in Section 1.3.1, especially as theregion definition is difficult because of the freepedestrian gaze.37

In the design stage, surrounding lightelements and surfaces are generally notknown and hypotheses are required (e.g.about the material, pavement, etc.). Glaremodels can be employed in a worst-casedesign based on road surface luminance orilluminances recommended by the standard.However, in urban areas, additional ambientlighting provided by billboards and shopwindows may increase the background lumi-nance (they may also be additional glaresources) and therefore impact the discomfortglare level predicted in the design stage.

The models based on illuminance performbetter and seem easier to implement. But inurban lighting applications, both the source(direct illuminance) and the adaptation (indir-ect illuminance) terms need to be measuredwhen the source is switched on. From apractical point of view, it is difficult to splitthe illuminance into a source component andan ambient component. This is, however,possible with complementary luminancemeasurements, or the use of a baffle at theilluminance meter.31

Table 8 RMSE and R2 between the Dir/Ind predictionsand the mean glare ratings collected in condition 1 S andcondition 2 S (simplified method)

Condition Background RMSE Spearman R2

Condition 1 S ERZ 0.76 0.655EDZ 1.05 0.676

Condition 2 S ERZ 0.84 0.655EDZ 0.60 0.688




5.2.4. Measurement issuesAs highlighted above, one of the main

issues in using discomfort glare models isrelated to measurement limitations. Whateverthe measurement method, some bias appears.An ILMD may be employed (as was done inthis paper) to generate luminance maps andcompute the average luminance over variousregions (adaptation regions, light sources) aswell as direct and indirect illuminances. But,the main problem with ILMDs is the under-estimation of the maximum luminance. Thislimit may explain why, in our data, themodel’s predictions were nearly constant forthose stimuli with the highest glare level. Weagree with Slominski49 that it is ‘very difficultif not impossible’ with current ILMD devicesto ‘precisely define the maximum luminancemeasured from a long distance (20–90m)’.Measurement errors increase with the dis-tance because the LED chip is included in toofew photosensors, even sub-pixel; thus, thepixel luminance averages the luminance ofthe LED chip and of its surroundings.Measurements also depend on the lens,sensor size and number of pixels.Slominski49 recommends ‘increasing thefocal length value or decreasing the physicalsize of the area recorded by a single pixel’.According to him, ‘the best solution is todesign an HDR device with lens whose field ofview of a single pixel of the B/W structure isprecisely connected to the resolution of thehuman eye’. A luminance meter may have awider luminance range, but the measurementprocess faces the same problems for choosingthe relevant measurement area; moreover, theviewfinder towards the source is manually setand the angular aperture needs to be adjusteddepending on the distance to the source.Therefore, the measurement region is notfully controlled. On the other hand, illumin-ance measurement does not meet these prob-lems. However, it is difficult to split theilluminance into a source component and anambient component. Bullough et al.9 and

Brons et al.31 proposed using a baffle in theilluminance meter to hide the source from thesensor. With such a trick, it seems difficulthowever to collect reproducible data.

One may speculate that computationsbased on the background luminance and theoverall illuminance may lead to the lowestmeasurement uncertainty: it would first needto measure the overall illuminance, then tocapture a luminance map with an ILMD. Thebackground photometry would then be com-puted from the luminance map (luminanceand then illuminance, based on a Lambertianreflection hypothesis). Finally, the source-related illuminance (and the source lumi-nance) would be computed by subtractingthe background illuminance from the overallilluminance. However, this is very speculativeand would deserve a complete study.

5.2.5. Conclusions on our model investigationsEven though our results depend on the

measurement and calculation choices anduncertainties, our findings do not. Two ofthe tested models (Bul11 and Lin15) give fairresults to predict the discomfort glare on thede Boer scale in urban pedestrian settings.A simplified model based on Bul11 also leadsto good results (Dir/Ind). The main issue topredict pedestrian discomfort glare is thechoice of the model; predictions are not verysensitive to implementation choices.

5.3. Limits and future work

Our experiment was designed in an out-door but controlled luminous environment,which would obviously be different in real-world situations (because of vehicles’ head-lamps, shop windows, etc.). Pilot studieswould be needed to investigate the potentialbias between glare ratings in controlled andfield environments. Our ‘walking condition’ isa step in this direction.

According to Miller et al.,4,5 the pedestriancould feel ‘overhead glare’,50 i.e. discomfortglare under the luminaires. Eccentricities

22 C Villa et al.



above 628 were not assessed in this work inthe static protocol and it would be interestingin future work to quantify this overhead glareand study if the glare models are able to takeit into account.

This paper focused on discomfort glaremodels using the de Boer scale, butother models could be tested, either developedfor outdoor51 or indoor23,52 applications.Especially, Kohko et al.30 investigated thepedestrian glare under seven light sourcesfocusing on the luminance uniformity issue.They compared the subjective data collectedon the de Boer scale with indoor glare models(e.g. DGI, UGR) and recommended to usethe mDGI index (DGI index where the sourceluminance is replaced by the efficient lumi-nance) for both uniform and non-uniformlight sources.

Finally, the dynamic protocol was a firststep to include temporal variations in dis-comfort glare assessment. All current modelswere fitted to ratings in static experimentalconditions. In future work, we plan to teststatic models in scenarios where temporalchanges occur, due to body and gaze motion.

6. General conclusion

This work investigates the discomfort glareexperienced by pedestrians under variousurban lighting installations. Psychovisualtests were conducted on a closed outdoorarea with four urban lighting installations tocollect discomfort glare judgments. Theresults confirm previous work5,30 which rec-ommended optics to limit maximum lumi-nance and vertical illuminance. Fourpredictive models have been tested withour data. The models from Lin et al.14 andBullough et al.9,12 provided the best predic-tions of the experimental data. Measurementand calculation choices were needed in orderto implement these models in complex visualscenes. Little impact of the background andsource definition was found on the model

predictions. However, measurement uncer-tainties (e.g. underestimation of the sourceluminance) may impact the accuracy of themodels, especially for high glare levels (i.e.low ratings). Such limitations deserve to beinvestigated in future work to ensure anaccurate implementation of discomfort glaremodels in urban environments. In addition,current models do not take into account thetemporal changes in the pedestrian’s glareexperience. Therefore, in future work, we planto investigate the glare rating during a walk.

Declaration of conflicting interests

The authors declared no potential conflicts ofinterest with respect to the research, author-ship, and/or publication of this article.

Funding

The authors disclosed receipt of the followingfinancial support for the research, authorship,and/or publication of this article: This workwas supported by Thorn Lighting.

Acknowledgements

The authors would like to thank Manuel LePape for his help, Alexandre Taron for hisavailability, his help and the facilities, EricDumont for his contributions and helpfuladvice, and all the participants who took partin the experiment.

References

1 Raynham P. An examination of the fundamen-tals of road lighting for pedestrians and drivers.Lighting Research and Technology 2004; 36:307–316.

2 Kostic A, Djokic L. Subjective impressionsunder LED and metal halide lighting. LightingResearch and Technology 2014; 46: 293–307.

3 Luo W, Puolakka M, Zhang Q, Yang C,Halonen L. Pedestrian way lighting: User




preferences and eye fixation measurements.Ingeniria Illuminatului 2013; 15: 19–34.

4 Miller NJ, Koltai RN, McGowan TK.Pedestrian Friendly Outdoor Lighting.Gateway Report PNNL-23085. Portland,OR: Pacific Northwest National Laboratory,2015.

5 Miller NJ, McGowan TK. Correspondence:Glare in pedestrian-friendly outdoor lighting.Lighting Research and Technology 2015; 47:760–762.

6 Behar-Cohen F, Martinsons C, Vienot F,Zissis G, Barlier-Salso A, Cesarini JP, EnoufO, Garcia M, Picaud S, Attia D. Light-emitting diodes (LED) for domestic lighting:Any risks for the eye?. Progress in Retinal andEye Research 2011; 30: 239–257.

7 Schmidt-Clausen H, Bindels J. Assessment ofdiscomfort glare in motor vehicle lighting.Lighting Research and Technology 1974; 6:79–88.

8 de Boer JB, Schreuder DA. Glare as a criterionfor quality in street lighting. Transactions ofthe Illuminating Engineering Society 1967; 32:117–135.

9 Bullough JD, Brons JA, Qi R, Rea MS.Predicting discomfort glare from outdoorlighting installations. Lighting Research andTechnology 2008; 40: 225–242.

10 Zhu X, Demirdes H, Gong X, Lai J,Heynderickx I. The luminaire beam-shapeinfluence on discomfort glare for led roadlighting: Proceedings on MidTerm CIE 2013,Paris, 15–16 April 2013, pp. 331–339.

11 Viliunas V, Vaitkevicius H, Stanikunas R,Vitta P, Bliumas R, Auskalnyte A, Tuzikas A,Petrulis A, Dabasinskas L, Zukauskas A.Subjective evaluation of luminance distri-bution for intelligent outdoor lighting.Lighting Research and Technology 2014; 46:421–433.

12 Bullough JD, Sweater-Hickocox K, NarendranN. A method for estimating discomfort glarefrom exterior lighting system. Alliance forSolid-State Illumination Systems andTechnologies 2011; 9: 1–7.

13 Lin Y, Liu Y, Sun Y, Zhu X, Lai J,Heynderickx I. Model predicting discomfortglare caused by LED road lights. OpticsExpress 2014; 22: 18056–18071.

14 Lin Y, Fotios S, Wei M, Liu Y, Guo W, SunY. Eye movement and pupil size constrictionunder discomfort glare. InvestigativeOphthalmology and Visual Science 2015; 56:1649–1656.

15 Mayeur A, Bremond R, Bastien C. Effect oftask and eccentricity of the target on detectionthresholds in mesopic vision: Implications forroad lighting. Human Factors 2008; 50:712–721.

16 Sivak M, Flannagan M, Ensing M, SimmonsCJ. Discomfort Glare is Task Dependent.Report No. UMTRI-89-27. Ann Arbor, MI:University of Michigan TransportationResearch Institute, 1989.

17 Land MF, Lee DN. Where do we look whenwe steer. Nature 1994; 369: 742–744.

18 Lappi O. Future path and tangent pointmodels in the visual control of locomotion incurve driving. Journal of Vision 2014; 14: 1–22.

19 Tom A, Granie MA. Gender differences inpedestrian rule compliance and visual search atsignalized and unsignalized crossroads.Accident Analysis and Prevention 2011; 43:1794–1801.

20 Bullough J, Van Derlofske J, Fay C, Dee P.Discomfort glare from headlamps: Interactionsamong spectrum, control of gaze and back-ground light level. SAE Technical Paper2003-01-0296. Warrendale, PA: Society ofAutomotive Engineers, 2003.

21 Hickcox KS, Narendran N, Bullough JD,Freyssinier JP. Effect of different colouredluminous surrounds on LED discomfort glareperception. Lighting Research and Technology2013; 45: 464–475.

22 Bullough JD, Fu Z, Derlofske JV. Discomfortand disability glare from halogen and HIDheadlamp systems. Advanced LightingTechnology for Vehicles 2002: 1–5, SP-1668.

23 Clear RD. Discomfort glare: What do actuallyknow? Lighting Research and Technology 2013;45: 141–158.

24 Niedling M, Kierdorf D, Volker S. Influence ofa glare sources spectrum on discomfort anddisability glare under mesopic conditions:Proceedings on MidTerm CIE 2013, Paris,France, 15–16 April 2013, pp. 340–347.

25 Sivak M, Flannagan MJ, Traube EC, KojimaS. The influence of stimulus duration on

24 C Villa et al.



discomfort glare for persons with and withoutvisual correction. Transportation HumanFactors 1999; 1: 147–158.

26 Theeuwes J, Alferdinck JWAM, Perel M.Relation between glare and driving perform-ance. Human Factors 2002; 44: 95–107.

27 Gibbons RB, Edwards C. A review of disabilityand discomfort glare research and future direc-tion: 18th TRB Biennial Visibility Symposium,College Station, TX, USA, 17–19 April 2007.

28 Commission Internationale de l’Eclairage.Glare and Uniformity in Road LightingInstallations. CIE Publication CIE-031-1976.Vienna: CIE, 1976.

29 Sammarco JJ, Mayton AG, Lutz T, GallagherS. Discomfort glare comparison for variousLED cap lamps. IEEE Transactions onIndustry Applications 2010; 47: 1168–1174.

30 Kohko S, Ayama M, Iwata M, Kyoto N,Toyota T. Study on evaluation of LED light-ing glare in pedestrian zones. Journal of Lightand Visual Environment 2015; 39: 15–25. doi:10.2150/jlve.IEIJ150000566.

31 Brons JA, Bullough JD, Rea MS. Outdoorsite-lighting performance: A comprehensiveand quantitative framework for assessing lightpollution. Lighting Research and Technology2008; 40: 201–224.

32 Van Bommel W. Road Lighting, Fundamentals,Technology and Application. London: Springer,2015.

33 Davoudian N, Raynham P. What do pedes-trians look at night? Lighting Research andTechnology 2012; 44: 438–448.

34 Cengiz C, Kotkanen H, Puolakka M, Lappi O,Lehtonen E, Halonen L, Summala H.Combined eye-tracking and luminance meas-urements while driving on a rural road:Towards determining mesopic adaptationluminance. Lighting Research and Technology2014; 46: 676–694.

35 Uchida T, Ayama M, Akashi Y, Hara N,Kitano T, Kodaira Y, Sakai K. Adaptationluminance simulation for CIE mesopic pho-tometry system implementation. LightingResearch and Technology 2016; 48: 14–25.

36 Maksimainen M, Puolakka M, Tetri E,Halonen L. Veiling luminance and visualadaptation field in mesopic photometry.Lighting Research and Technology 2016. First

published 22 March 2016, doi: 10.1177/1477153516637400.

37 Foulsham T, Walker E, Kingstone A. Thewhere, what and when of gaze allocation in thelab and the natural environment. VisionResearch 2011; 51: 1920–1931.

38 Ayama M, Tashiroi T, Kawanobe S, Kimura-Minoda T, Kohko S, Ishikawa T. Discomfortglare of white led sources of different spatialarrangements: Proceedings on MidTerm CIE2013, Paris, 15–16 April 2013, pp. 119–122.

39 Higashi H, Koga S, Kotani T. The developmentof evaluation for discomfort glare in led lightingof indoor work place: The effect of the lumi-nance distribution of luminous parts on subject-ive evaluation: Proceedings on MidTerm CIE2013, Paris, 15–16 April 2013, pp. 648–656.

40 Tashiro T, Kawanobe S, Kimura-Minoda T,Kohko S, Ishikawa T, Ayama M. Discomfortglare for white LED light sources with differ-ent spatial arrangements. Lighting Researchand Technology 2014; 47: 316–337.

41 de Boer JB. Lighting of traffic routes. PhilipsTechnical Review 1961–62; 23: 258–272.

42 Commission Internationale de l’Eclairage.Road Surfaces and Lighting. CIE Publication066:1984. Vienna: CIE, 1984.

43 European Committee for Standardisation(CEN). Road Lighting – Part 2: PerformanceRequirements. European Standard EN 13201-2. Brussels: CEN, 2003.

44 Howell DC. Statistical Methods forPsychology. 6th Edition, Belmont, CA:Duxbury Press, 2008.

45 Mauchly JW. Significance test for sphericity ofa normal n-variate distribution. The Annals ofMathematical Statistics 1940; 11: 204–209.

46 Bardy B, Warren WH, Kay BA. The role ofcentral and peripheral vision in postural con-trol during walking. Perception andPsychophysics 1999; 61: 1356–1368.

47 Fotios S, Uttley J, Cheal C, Hara N. Usingeye-tracking to identify pedestrians’ criticalvisual tasks. Part 1. Dual task approach.Lighting Research and Technology 2015; 47:133–148.

48 Kasahara T, Aizawa D, Irikura T, MoriyamaT, Toda M, Iwamoto M. Discomfort glarecaused by white LED light sources. Journal ofLight and Visual Environment 2006; 30: 95–103.




49 Slominski S. Identifying problems withluminaire luminance measurements for dis-comfort glare analysis. Lighting Research andTechnology 2016; 48: 573–588.

50 Boyce PR, Hunter CM, Inclan C. Overheadglare and visual discomfort. Journal of theIlluminating Engineering Society 2003; 32:73–88.

51 Bennett CA. Characterizing discomfort glarefrom roadway lighting. Transaction ResearchRecord 1149. In Visibility for HighwayGuidance and Hazard Detection, 1987.

52 Luckiesh M, Guth SK. Brightnesses in visualfield at borderline between comfort and dis-comfort (BCD). Illuminating Engineering 1949;44: 650–670.

26 C Villa et al.



assessment of pedestrian discomfort glare from urban led

Documents