1
Assessment of a simple obstacle detection device for the visually
impaired
Cheng-Lung Lee1*
, Chih-Yung Chen1, Peng-Cheng Sung
1 and Shih-Yi Lu
2
1Department of Industrial Engineering and Management, Chaoyang University of Technology,
No. 168, Jifeng E. Rd., Wufeng District, Taichung, 41349, Taiwan, R.O.C. 2School of Occupational Safety and Health, Chung Shan Medical University,
No. 110, Sec. 1, Jianguo N. Rd., 40201, Taiwan, R.O.C.
*Corresponding author. Tel: +886 4 23323000 ext 4458; fax: +886 4 23742327.
E-mail address: [email protected]
Postal address: Department of Industrial Engineering and Management, Chaoyang
University of Technology, No. 168, Jifeng E. Rd., Wufeng District, Taichung,
41349, Taiwan, R.O.C.
Abstract
A simple obstacle detection device, based upon an automobile parking sensor, was assessed
as a mobility aid for the visually impaired. A questionnaire survey for mobility needs was
performed at the start of this study. After the detector was developed, five blindfolded sighted
and fifteen visually impaired participants were invited to conduct travel experiments under
three test conditions: 1) using a white cane only, 2) using the obstacle detector only and 3)
using both devices. A post experiment interview regarding the usefulness of the obstacle
detector for the visually impaired participants was performed. The results showed that the
obstacle detector could augment mobility performance with the white cane. The obstacle
detection device should be used in conjunction with the white cane to achieve the best
mobility speed and body protection.
Keywords: Visually impaired; obstacle detection; travel aid.
1. Introduction
Mobility refers to one’s ability to identify the relation between their position and the objects
in the environment and then move independently, safely and efficiently (Kuyk et al. 2010).
2
The most obvious problem faced by blind persons is moving around in their environment
without bumping into unexpected obstacles (Molton et al. 1998). Obstacle detection is thus
one of the major problems to be solved to ensure safe navigation. The white cane is the most
popular and traditional navigation aid for the visually impaired (Molton et al. 1998; Snaith et
al. 1998; Dakopoulos and Bourbakis 2010) in spite of modern technology-based devices
(Clark-Carter et al. 1986a; Schellingerhout et al. 2001). The white cane is the simplest,
cheapest and most reliable device thus far. It can be generally applied to detect static
obstacles on the ground, uneven surfaces, holes and stairs (Cardin et al. 2007). However, the
reach of the cane is limited (Yasumuro et al. 2003) and obstacles not located on the ground
are hardly detected.
A number of electronic travel aids (ETAs) for the visually impaired have been developed
for navigation and obstacle detection/avoidance. Dakopoulos and Bourbakis (2010) presented
a comparative survey among portable/wearable obstacle detection/avoidance systems for the
visually impaired. Cardin et al. (2007) also reviewed several obstacle detection devices
developed in the literature.
The common ETA features based upon new technologies in the literature may conclude
that information was gathered from the environment using sonar, laser scanner or stereo
camera vision. The user was generally informed through auditory and/or tactile sense (Cardin
et al. 2007; Dakopoulos and Bourbakis 2010). Other features and disadvantages of ETAs are
briefly summarized as follows. The signal processing of many novel ETAs usually required
complicated computer algorithms to provide more complete information about nearby
obstacles (Cardin et al. 2007; Sainarayanan et al. 2007). The sensors in some ETAs could
only detect obstacles on the ground just like the white cane. Other ETAs had more functional
body protection capabilities (e.g., shoulder protection, Cardin et al. 2007). Some ETAs were
small, light and handheld, e.g., Miniguide (Phillips, 1998), while other ETAs were bulky, e.g.,
Navbelt (Shoval et al. 1994) and GuideCane (Ulrich and Borenstein 2001), making them
3
difficult to hold or carry when needed. Although some aids were portable, they still required
manual operation (e.g. Ghiani et al. 2009), making co-incident use of conventional aids
difficult or impossible (e.g. white canes). One or more cameras in some ETAs required
mounting on headgear (Sainarayanan et al. 2007) or the frame of eyeglasses for obstacle
capture (Meers and Ward 2005 cited Dakopoulos and Bourbakis 2010). Such devices may be
not accepted by some visually impaired individuals because they may feel cumbersome or
embarrassed to wear in public places. Some ETAs required extensive training periods (Snaith
et al. 1998). The relatively high cost of ETAs available in the market is a discouraging
feature.
To fill the detection gaps in the white cane and overcome the disadvantages of ETAs, a
compact size, lightweight, low-cost and simple detection feedback obstacle detection device
was developed and explored in this study. The prototype of this detector was developed and
evaluated using blindfolded sighted and visually impaired participants in travel experiments.
This study hypothesized that different travel aid devices would alter walking efficiency and
obstacle detection for visually impaired persons. The aims of this study were to compare ETA
walking efficiency and obstacle detection capability using both blindfolded sighted and
visually impaired participants under three test conditions.
2. Method
2.1 Survey of mobility needs
At the start of obstacle detector development a survey of visually impaired students was
conducted to determine their mobility needs in daily life. Participants were interviewed
individually using a constructed questionnaire. The data were then recorded by researchers.
The first part of the questionnaire contained the participant’s personal data which included
name, gender, age, educational level, blind or visually impaired condition, peripheral (side)
vision, awareness of light, colour vision and any other impairment. The second part asked for
travel experiences including daily activities, use of travel aids and obstacle collision
4
experiences in the environment. Several discussions were held with teachers, who were blind
and teaching in the school, to modify the questions before the interview was performed. The
purpose of this study was clearly explained to the participants prior to the interview. Informed
consent to participate was obtained from the participants in this study.
Thirty one participants were junior high and senior high students from a school for the
visually impaired in Central Taiwan. The students had received at least six-month training in
orientation and mobility before joining this study. During the interview one student was
determined to have a learning disability, and thus, was eliminated from the study. A total of
30 valid participants included 15 junior high school students (aged 13-15, 9 males and 6
females) and 15 senior high school students (aged 16-19, 9 males and 6 females). As
self-reported during the interview, 25 students (83.3%) said they were blind and 5 classified
themselves as having low vision, 27 (90.0%) with congenital impairment and 3 with acquired
impairment.
All interviewed participants lived at school during the weekdays and could choose to
return home or stay at school during weekends and holidays. This study (including a survey
and an obstacle detection experiment) was approved by the Institutional Review Board for
Ergonomics Experiment of Chaoyang University of Technology.
2.2 Travel aid development
2.2.1 Obstacle detector
The obstacle detector developed in this study was based upon an automobile parking sensor,
consisting of 3 main modules; a sensing module with transmitting and receiving functions, a
processing module and a warning module. Table 1 shows the dimensions and photos of the
ultrasonic obstacle detector modules. The detector architecture is illustrated in Figure 1. The
detection system works by sending out ultrasonic pulses that are reflected back to the sensor
by obstacles within the sensor detection envelope. The reflected signals are processed by the
processing module, which in turn activates the warning buzzer to alert the user that there is an
5
obstacle in the travel path.
The three detector modules were commercial electronic parts packed as a set used for
obstacle detection while car-parking. The default distance sensitivity setting of the sensor was
adjusted to a lower level for this study. Another detection system modification was made for
the detector power connection. A 12 Volt dry battery was sufficient for detector use by an
individual. A 12 Volt battery with 2Ah capacity was used to supply the experiment with
sufficient operational time during a day. The obstacle detector was small and lightweight, as
shown in Table 1. One sensing module was mounted on the participant’s body with the other
two modules placed in a small bag on a belt worn on the waist or in a pocket. Two sensors
were attached to the participant’s chest and waist, respectively, during the experiment. The
reason for this arrangement was to test if the detector could effectively detect obstacles not
located on the ground.
The sensor detection envelope was investigated in the laboratory before the detector was
assessed in the field. A high flat wooden board (height 245.4 cm, width 32.5 cm) and three
university students were employed to move around as targets for detection envelope
measurements. The mean age of these students was 27.0 yr/SD 2.6 yr; mean body height
165.6 cm/SD 4.6 cm; and mean body weight 60.3 kg/SD 10.8 kg. The test direction was
varied every 5° from the mid-saggital plane to the right and left sides. Each angle for each
target was tested twice. Another test for detection distance was performed at the university
campus after the laboratory test. The obstacles on the campus included the iron railings
outside a building and in an athletic field, the net in a volleyball court, the bonnet of sedans
and the stairways. An individual (stature 163.3 cm and body weight 68.4 kg) was asked to
wear a sensor on his chest (120.4 cm) in the test. A parking lot gate arm (7.5 cm in width)
was also tested in a parking lot by the individual with a sensor worn on his waist.
2.2.2 Evaluation of the Electronic Obstacle Detector
Two experimental stages were performed for obstacle detection evaluation in this study. Five
6
blindfolded sighted participants (4 males and 1 female) who were graduate students at the
university were recruited to detect obstacles in a simulated environment. Fifteen visually
impaired participants (10 males and 5 females) were then invited to travel a real world
mobility route. The visually impaired participants included 5 junior high school students, 6
senior high school students and 4 teachers from a visually impaired school in Central Taiwan.
These participants were totally blind with no residual vision. Thirteen suffered from
congenital visual impairment. All visually impaired participants had received at least
six-month training in orientation and mobility prior to this study. Table 2 shows the
participants’ personal data and anthropometric measurements.
A corridor 30 meters long outside the laboratory at the university was used as a
simulated route for the first experimental stage. Nine obstacles existed in the corridor,
including 5 purposefully placed obstacles (1 bicycle, 2 chairs, and 2 overhanging cardboard
boxes) arranged in the middle of the corridor and 4 natural obstacles (2 shoe cabinets and 2
shelves) situated against the corridor walls. The positions of the purposefully placed obstacles
were changed for each test and each participant. No arrival time was recorded at each
checkpoint (i.e., the obstacle location) in this experimental stage and only the total time for
each test was recorded.
The route around the school campus travelled by visually impaired participants was used
for the second-stage test. The route shown in Figure 2 was determined based upon the
concern for the participant’s transportation safety when the experiment was proposed and
discussed with the school staff. Eight obstacles were placed in the 138 meter long route. The
purposefully placed obstacles consisted of 2 bicycles/motorcycles and 1 hanging cardboard
box. Their locations were varied for each test condition and each participant. Five natural
obstacles existed in the path, including a plant terrace, a tall tree, a pedestrian overpass, a
side-door and a tilted tree. Ten checkpoints in the path, shown in Figure 2, consisted of 3
locations on the pedestrian overpass and 7 locations at the other obstacles. The arrival time to
7
each of these 10 checkpoints in the path was recorded for each test condition and each
participant.
Three experimental test conditions were completed by both blindfolded sighted and
visually impaired participants: 1) using a white cane only, 2) using the obstacle detector only
and 3) using both devices together. The test conditions were selected at random for each
participant in the experiment.
A collision prevention strategy regarding participant safety during the experiment was
stated as follows. A collision risk assessment was conducted to ensure the safety of
participants before the experiment was performed. A pilot test was performed by an
experimenter with and without a blindfold in order to observe the possible collisions before
and after the collision prevention was discussed and improved. The possible parts of
obstacles that might be collided with by the participant’s body during the experiment were
wrapped with foam instead of asking the participant to wear personal protective equipment
such as helmet and gloves. This was done because observing the participant’s behaviours
after obstacle collision was one of the major study purposes. Several instructions were given
to the participants before the experiment: (a) Participants had the possibility of body collision
with obstacles on the footpath during the experiment. Such obstacles were wrapped with
foam and might exist at ground, waist or head level. (b) Although measures to minimize the
risk for injury had been taken, participants could still suffer light impact with obstacles when
collisions occurred during the experiment. (c) An experimenter would follow the participant
on the path during the experiment to protect the participant from any injury. When an
orientation loss took place or before a harmful collision occurred the experimenter would
guide the participant with verbal warnings or/and hand(s). (d) Participants had the option to
withdraw from the experiment any time.
A clear explanation of the experimental objectives and procedures was given to ensure
no difficulties in operating the experimental aids by the participants A brief training period
8
for both blindfolded sighted and visually impaired participants was provided to ensure
enough practice and familiarity with the tests before actual data were collected. A written
consent from each participant was obtained and the anthropometric data were measured. For
test conditions 2 and 3 (i.e., a detector included), two sensing modules were worn on each
participant's chest and waist to detect possible collisions by the head or waist. Both finish
time and collision frequency of three test conditions for each participant were recorded to
determine whether different test conditions resulted in differences in walking efficiency and
mobility safety. A warning sound was delivered to the participants by a buzzer. An
experimenter followed the participant on the footpath, could hear the warning sound and
record the participant’s behaviour during the experiment. Post experiment interviews
regarding the usefulness of the travel aids were carried out after the participant had completed
all three test conditions.
Finish time (FT) was defined as the time required for one participant to complete one
test condition. Collision frequency (CF) was considered only when participants collided
unintentionally with the obstacles. If the participants walked into and made contact with an
obstacle on the route with their white cane or when an alarm sounded and the participant
recognized that an obstacle lay ahead in test conditions 2 and 3, tentative actions were usually
taken by participants to try to determine the characteristics of the obstacles. Such behaviour
would be considered an intentional collision and was not counted in CF in this study.
2.2.3 Data Analysis
Finish time and collision frequency under three different test conditions from 5 blindfolded
sighted and 15 visually impaired participants were obtained in the experiment. Data were
analyzed using one-way ANOVA using SPSS 10.0 for Windows. When the main effect was
significant, a post hoc analysis using Duncan's new multiple range test was applied to
evaluate differences among test conditions. All significance levels were set at alpha =.05.
9
3. Results
3.1 Mobility needs
The questionnaire results showed that 90.0% of the participants often used a white cane when
going out alone. Ninety percent of the visually impaired did not use white canes in familiar
environments such as school or home. The places visited most frequently included stores
(73%), relatives’/friends’ homes (50%), and restaurants (30%). Figure 3 shows four obstacle
types that the participants often collided with while navigating. Both the
automobile/motorcycle (70%) and piled item/protrusion (63%) obtained the highest obstacle
collision percentage. Figure 4 illustrates the body parts that were most often collided with
obstacles and expected to avoid obstacle-colliding. All participants expressed that they had
never used electronic travel aids before but would like to try if available. The low cost,
convenient aids were of greatest interest for all participants.
3.2 Experiment
3.2.1 Mobility Performance Data
Figure 5 shows the sensor detection envelopes obtained from two tests using a high flat
wooden board and three persons as targets in the laboratory. The farthest and shortest
detection distances were about 228 cm and 43 cm when the sensor was placed right ahead of
the wooden board (i.e., 0° in the figure) and at ±50° to the mid-sagittal plane of the person.
The sensor did not detect the wooden board or three persons at angle ranges beyond 50°. The
detection distance measured at the university campus after the laboratory test ranged from
72.5 to 152.3 cm for different obstacles, including iron railings, volleyball net, sedan bonnet
and stairways. A parking lot gate arm that protruded about the individual’s waist height (105
cm) could be detected.
Figures 6 and 7 show the finish time (FT) and collision frequency (CF) means obtained
from 5 blindfolded sighted and 15 visually impaired participants under the three test
conditions. The shortest finish time was found using the white cane plus detector test
10
condition for both blindfolded sighted and visually impaired participants. The maximum
differences in FT among the visually impaired participants were 235.4 seconds using the
white cane only, 213.4 seconds using the detector only, and 156.9 seconds using both aids
together. The FT values varied markedly among the visually impaired participants under the
same test condition, however, such results were not found among the blindfolded sighted
participants. Table 3 shows the collision frequency and the obstacles collided with by both 5
blindfolded sighted and 15 visually impaired participants under the three test conditions. The
body parts that collided with the obstacles are also shown in the Table. The white cane only
usually failed to detect obstacles above the ground for both blindfolded sighted and visually
impaired participants, especially obstacles at head level. When the obstacle detector sensors
were placed on the upper parts of the participant’s body the CF values were reduced. No CF
values were obtained under the white cane plus detector condition for both blindfolded
sighted and visually impaired participants. The statistical analysis results revealed that both
finish time and collision frequency for the visually impaired participants showed significant
differences among the three test conditions (F(2, 42)=6.935, p=.002, and F(2, 42)=63.7,
p=.000). The Duncan's groupings for both FT and CF were then obtained by test condition
(α=.05) and present in Figures 6 and 7, respectively. Among the three test conditions, the test
using both devices produced the smallest FT. The test with the white cane only produced the
largest CF among the three conditions.
3.2.2 Post Experiment Interview
All blindfolded sighted participants stated that they did not feel confident in detecting
obstacles while using the detector only. For the other two test conditions they felt more
confident because the white cane was included. However, they all reported positive when the
detector was used in the experiment and agreed that the detector was useful in detecting
obstacles not located on the ground. The post experiment interview results for the visually
impaired participants are listed in Table 4. Eleven participants (73.4%) were satisfied or very
11
satisfied with current detector prototype, and 12 (80.0%) were satisfied or very satisfied with
the prototype weight. Thirteen participants (86.7%) expressed that the travel aid was really
helpful to their mobility and their confidence in mobility could be enhanced (80.0%). In
response to questions about their preference for the devices used in this experiment, all
participants preferred to rely upon the white cane due to its reliability when only one device
could be chosen. However, using both devices would be their best choice if the obstacle
detector was available for use.
4. Discussion
4.1 Mobility needs
The questionnaire inquired into the participant’s vision status (total blindness and low vision),
light perception and colour vision. The visually impaired participants recruited for this study
were totally blind with no residual vision. The participants’ vision status was not reconfirmed
with eye examinations. Their answers to these questions were kept confidential. However,
there might be bias in the study survey results.
The participants were allowed to choose more than one answer (item) in the questionnaire
regarding the obstacle types that they often collided with. Seventy percent of the visually
impaired participants, shown in Figure 3, expressed that automobile/motorcycle was the most
common obstacle that they often collided with. Automobiles or motorcycles are allowed to be
parked on the roadside in Taiwan. Occasionally automobiles or motorcycles are illegally
parked on the sidewalk. Visually impaired persons therefore may easily meet and collide with
them while walking in the street. The participants were allowed to answer with more than one
body part that often collided with obstacles. The result shown in Figure 4 indicated that head
collisions (57%) occurred almost as often as leg collisions (60%). However, there are far
fewer head level obstacles than leg level obstacles in a natural environment. This might be
explained as follows. A head collision is a serious event and the participant was more likely
to remember head collision incidences than other body part collision incidences. The high
12
rate of reported head collisions might simply reflect the long-lasting memory of painful head
collisions instead of the true frequency of head collision incidences. Therefore, the highest
desire was to avoid head collisions, which seemed to be a reasonable assessment, with the
highest percentage (50%) among body parts.
All participants agreed that the white cane could be used to detect static obstacles on the
ground and protect the lower body parts but protection gaps did exist. The white cane was
insufficient to protect the head, hands and waist while walking. Head collision was the
participant’s strongest desire to avoid. Additionally, the cost and ease-of-use of travel aids
were also a great concern. This motivated the development of a low cost electronic detector
with a warning for obstacles near the head and waist of the visually impaired.
The limitations of this survey were that the sample size was small, the participants were
young and their travel environment was restricted mostly to a school campus or several
limited sites that they frequently visited. Future surveys must collect larger samples with a
wide range of ages and visual impairment for further statistical analysis.
4.2 Evaluation of aids
The longest finish time (FT) was found in the test condition using only a detector for both
blindfolded sighted and visually impaired. All blindfolded sighted and visually impaired
participants had no previous experience using a detector and were observed to walk forward
slowly and carefully when using only the detector during the experiment. It was also found
that the mobility speed varied markedly among participants. An individual’s walking habit
partly accounted for the speed difference.
The shortest FT appeared in the condition where both devices were used together. The
body protection (for head, waist/hands, and legs) was also enhanced and the collision
frequency (CF) decreased. The mobility performance and body protection using both devices
used together were the highest in all test conditions. Using a detector only protected the
user’s body parts better than the white cane but was worse in mobility speed. The
13
experimental results verified the study hypotheses. This is consistent with the results of
Snaith et al. (1998) and Cardin et al. (2007) that the white cane is a major mobility aid and
other aids can assist with the cane’s deficiency. The developed detector did augment the white
cane. Participants walking with both devices together would obtain more information about
obstacles ahead and their confidence in mobility might be increased. This finding could be
observed from the post experiment interview results, in which all participants expressed
positive comments for the presence of the detector. It should be noted that the benefit of using
the cane and the detector together only occurred in finish time, not in obstacle collision.
There was no significant difference in obstacle collision between the detector alone condition
and the white cane plus detector condition.
Visually impaired participants received mobility and orientation training prior to this study.
They could more easily adopt coping strategies and obtain more confidence with their
mobility needs. It was observed that they showed good navigation behaviour during the
experiment. However, the study results showed that visually impaired participants gained
greater finish times. The difference in finish time between visually impaired and blindfolded
sighted participants might result from the travel routes selected for the experiment. The route
around the school campus navigated by visually impaired participants was to some extent
complicated. There were 3 turning points, 1 pedestrian overpass, 1 plant terrace and 1
side-door on the route. This route was longer than that travelled by the blindfolded sighted
participants. Visually impaired participants sometimes deviated from the planned path during
the experiment and the actual travel distance became larger when an orientation loss took
place. However, path deviation and orientation loss were found in only a few instances for
blindfolded sighted participants because the test route was straight, flat and shorter.
Additionally, the blindfolded sighted participants usually waited at laboratory for the
experiment and basically knew the route and the obstacles before they were blindfolded.
Therefore, blindfolded sighted participants would have much shorter finish times. The
14
preliminary test for the travel aid detection function (i.e., white cane and detector), observed
from blindfolded sighted participants, was the purpose for the first experimental stage.
Observing visually impaired participants’ behaviour during the experiment was the purpose
of the second stage. The navigation performance comparison between these two groups was
not the original experiment design in this study.
The detector developed in this study could not differentiate the characteristics of obstacles.
The warning sound simply raised the alertness level of the participant and prompted the
participant to take cautious actions. It was observed that the participants always moved with
their hand(s) raised forward and upward when the buzzer sounded under test conditions 2 and
3 (i.e., a detector included). They also swung their white cane vigorously when an obstacle
was detected by the white cane. They adopted these actions to determine what and where the
obstacles were. They then lowered their head or stooped down to avoid head collisions,
side-stepped away from an obstacle on the ground with hands touching or/and with the white
cane swinging. A loss of orientation usually occurred for most participants after they collided
with an obstacle during the experiment. Visually impaired persons should decrease their
walking speed to avoid an accidental collision. Additionally, they should be trained how to
solve the orientation loss problem while collisions happen during navigation.
The warning signal for the obstacles ahead was the only information provided by the
detector. This has been generally sufficient in most situations for the visually impaired to take
action and avoid possible collisions. The developed device was small in size and lightweight.
One or more sensing modules can be easily worn or attached to user body sites such as the
head, chest, waist, hand or other body part to detect obstacles. One advantage of the current
detection system is that it is hands free.
At the travel aid development stage pilot trials were performed to test the detection system
capabilities. The warning signal frequently sounded in trials with the original detection
system (i.e., without any sensor modification). It was observed that objects located a little
15
farther on the side of the road and within the detection envelope might be detected by the
sensor. However, such objects did not obstruct the users’ movement. The sensor default
distance sensitivity setting was adjusted to a lower level. The detection system then worked
reasonably. This sensor adjustment avoided false alarms while walking and successfully met
the obstacle detection demands of this study.
The distance envelope obtained with the test using a wooden board as the target was larger
than that used by the sample individuals (Figure 5). This result was fair because more
reflected signals were received from the flat board by the ultrasonic sensor due to its smooth
surface. A certain amount of reflected signals occurred from an individual, however, these
signals became scattered and could not be received by the detector. This decreased the
effective detection distance. Furthermore, the detection distance varied for different obstacles
tested at the university campus after the laboratory test. It was found that the sensor distance
envelope could be affected by several factors such as the body movement while walking, the
individual’s clothing, the body sites where sensors were attached and the obstacle features
(e.g., dimensions, appearance, and surface roughness). Users should be aware that the sensor
detection distance is target and direction dependent.
The visually impaired participants were not told or did not ask about the nature and
number of obstacles appearing on the route as well as whether the same route was used for all
three tests during the experiment. Three test conditions were completed in random order by
each participant and the locations of the purposefully placed obstacles were varied in each
condition. The experimental results could be used to reasonably explore the usefulness and
make a comparison analysis of the tested devices. However, a familiar test environment
might lead to insufficient generalizability of the FT and CF results.
Eight obstacles were scattered over a 138-meter path with no two obstacles within the
detection envelop at the same time in this study. However, the participants were not told that
there was only one obstacle when the warning sounded. They always behaved as described
16
above when an obstacle was detected. In a realistic environment the warning sound might be
continuous. If this occurs it is suggested that the user turn the detector off and use a white
cane or other suitable travel aid instead. This might be the limitation of using this detector.
The experimental results presented positive responses to the detector. However, detector
improvement is still needed. To enhance user convenience the detection device system should
be designed with a user-friendly interface and wireless function. A vibro-tactile feedback and
earphone function should also be provided to ensure the warning signals are received by users
who are deaf/blind or in a noisy environment. Furthermore, a global positioning system (GPS)
may be taken into account in detection system development to provide location information
for the visually impaired in outdoor navigation.
The current developed detector provides only information that a detected obstacle is in
front and nearby. If the actual position of the obstacle (i.e., appearing on the left side, right
side, upper level, or lower level of the user) can be provided the user may more easily take
correct action to avoid obstacle collision. Future work should be done regarding this
suggestion: 1) the feasibility of manufacturing such a detector prototype/product, 2)
experiments conducted for detection performance and 3) a training course for the user.
5. Conclusions
A compact, lightweight, simple detection feedback, easy-to-use, low-cost obstacle detection
device for the visually impaired was developed in this study. The results showed that the
developed detector could augment the user’s performance with a white cane. To achieve the
best mobility performance and body protection effect, the obstacle detection device should be
used in conjunction with the white cane.
Acknowledgements
The authors wish to thank the National Taichung Special Education School for the Visually
Impaired for their support and the participants for their time and patience in completing the
questionnaire survey and the experiment.
17
References
Cardin, S., Thalmann, D. and Vexo, F., 2007. A wearable system for mobility improvement of
visually impaired people. Visual Comput. 23, 109-118.
Clark-Carter, B.D., Heyes, A.D. and Howarth, C.I., 1986a. The efficiency and walking speed
of visually impaired people. Ergonomics 29, 779-789.
Dakopoulos, D. and Bourbakis, N.G., 2010. Wearable obstacle avoidance electronic travel
aids from blind. IEEE Trans. Syst., Man Cybern. C, Appl. Rev. 40 (1), 25-35.
Ghiani, G., Leporini, B. and Paterno, F., 2009. Vibrotactile feedback to aid blind users of
mobile guides. J. Visual Lang. Comput. 20, 305-317.
Kuyk, T., Liu, L., Elliott, J. and Fuhr, P., 2010. Visual search training and obstacle avoidance
in adults with visual impairments. J. Visual Impair. Blind. 104 (4), 215-227.
Molton, N., Se, S., Brady, J.M., Lee, D. and Probert, P., 1998. A stereo vision-based aid for
the visually impaired. Image Vision Comput. 16, 251-263.
Phillips, G., 1998. The Miniguide ultrasonic mobility aid. GDP Research, South Australia.
online at http://www.gdp-research.com.au/minig_1.htm
Sainarayanan, G., Nagarajan, R. and Yaacob, S., 2007. Fuzzy image processing scheme for
autonomous navigation of human blind. Appl. Soft Comput. 7, 257-264.
Schellingerhout, R., Bongers, R.M.,Van Grinsven, R., Smitsman, A.W., and Van Galen, G.P.,
2001. Improving obstacle detection by redesign of walking canes for blind persons.
Ergonomics 44 (5), 513-526.
Shoval, S., Borenstein, J. and Koren, Y., 1994. Mobile Robot Obstacle Avoidance in a
Computerized Travel Aid for the Blind”, in Proceeding of the IEEE International
Conference on Robotics and Automation, San Diego, CA, 2029-2032.
Snaith, M., Lee, D. and Probert, P., 1998. A low-cost system using sparse vision for
navigation in the urban environment. Image Vision Comput. 16, 225-233.
Ulrich, I. and Borenstein, J., 2001.The GuideCane – applying mobile robot technologies to
18
assist the visually impaired people. IEEE Trans. Syst., Man Cybern. A, Syst., Humans 31
(2), 131-136.
Yasumuro, Y., Murakami, M., Imura, M., Kuroda, T., Manabe, Y. and Chihara, K., 2003.
E-cane with situation presumption for the visually impaired. In: N. Carbonell and C.
Stephanidis, ed. User Interfaces for All, LNCS 2615, Berlin Heidelberg: Springer-Verlag,
409-421.
19
Table 1. Dimensions and pictures of the electronic obstacle detector modules.
Table 2. Personal data and anthropometric measurements of 5 blindfolded sighted (BS) and
15 visually impaired (VI) participants in the experiment.
Table 3. Collision frequency (CF) and obstacles collided with by both 5 blindfolded sighted
and 15 visually impaired participants under three test conditions.
Table 4. Detector dimension, weight, mobility assistance and confidence satisfaction survey.
Values in parenthesis are the number of visually impaired participants.
20
Figure 1. The obstacle detector architecture which included three modules: a sensing module,
a processing module, and a warning module.
Figure 2. The route sketch for the visually impaired participants to travel in the second-stage
test.
Figure 3. Four obstacle types that participants mostly often collided with while walking.
Figure 4. The percentage of body parts that often collided with obstacles and most expected
to avoid collisions. Hands/waist is expressed together because the hands and waist are
usually at the same vertical level while walking.
Figure 5. Sensor detection envelopes obtained from two tests using a high flat wooden board
and three persons as targets.
Figure 6. Means and standard deviations of finish time (FT) and the Duncan’s new multiple
range test by test condition for both the visually impaired and blindfolded sighted
participants. Means with the same letter are not significantly different (α=.05).
Figure 7. Means and standard deviations of collision frequency (CF) and the Duncan’s new
multiple range test by test condition for both the visually impaired and blindfolded sighted
participants. Means with the same letter are not significantly different (α=.05).
21
Table 1. Dimensions and pictures of the electronic obstacle detector modules.
Module Dimension (mm) Weight (g) Photo
Processing module
(Control host)
Length×Width×Height
=95.2×57.0×25.8
74.8
Sensing module
(transmitting and
receiving unit)
Length×Width×Height
(Max.) =37.6.×24.2×15.4
16.6×2a
Warning module
(buzzer)
Diameter×Height=41.5×15.9 10.5
a: 2 sets of the sensing module were worn on the participant's chest and waist during the experiment.
22
Table 2. Personal data and anthropometric measurements of 5 blindfolded sighted (BS) and
15 visually impaired (VI) participants in the experiment.
Variable Age (yr) Stature
(cm)
Body weight
(kg)
Limb length
(cm)
Waist height
(cm)
Shoulder
height (cm)
Mean (SD)a
BS 31.2(2.8) 165.3(5.0) 62.1(6.5) 71.6(4.5) 92.5(3.7) 129.1(4.3)
VI 23.8(13.5) 161.3(6.9) 57.3(11.1) 68.6(6.4) 93.4(6.8) 128.4(5.8)
Range
BS 27.1-34.2 160.5-173.7 54.7-68.4 67.1-76.2 86.4-96.4 125.6-140.1
VI 13.7-51.4 149.8-172.3 45.0-74.6 56.4-79.3 80.4-109.4 119.5-139.4 a: SD indicates standard deviation
23
Table 3. Collision frequency (CF) and obstacles collided with by both 5 blindfolded sighted
and 15 visually impaired participants under three test conditions.
N Statusa Gender
b White cane Detector Both
d
CF Obstacle (body part)c CF Obstacle(body part) CF Obstacle(body part)
The blindfolded sighted participant
1 G M 2 box 1 (H), box 2 (H) 1 chair 1 (W) 0 0
2 G F 2 box 1 (H), box 2 (H) 0 0 0 0
3 G M 2 box 1 (H), box 2 (H) 0 0 0 0
4 G M 2 box 1 (H), box 2 (H) 1 chair 2 (W) 0 0
5 G M 2 box 1 (H), box 2 (H) 0 0 0 0
Mean 2 0.4 0
SD 0 0.5 0
Range 2 0-1 0
The visually impaired participant
1 T M 2 Box (H), tilted tree (H) 0 0 0 0
2 T F 2 Box (H), tilted tree (H) 0 0 0 0
3 T M 1 Tilted tree (H) 1 bike/motor 2 (W) 0 0
4 T M 0 0 0 0 0 0
5 S M 3 Bike/motor 1 (W), box
(H), tilted tree (H)
0 0 0 0
6 S M 2 Box (H), tilted tree (H) 1 bike/motor 2 (W) 0 0
7 S F 2 Box (H), tilted tree (H) 0 0 0 0
8 S F 2 Box (H), tilted tree (H) 1 bike/motor 2 (W) 0 0
9 S F 2 Box (H), tilted tree (H) 0 0 0 0
10 S F 2 Box (H), tilted tree (H) 0 0 0 0
11 J M 2 Box (H), tilted tree (H) 0 0 0 0
12 J M 2 Box (H), tilted tree (H) 1 bike/motor 1 (W) 0 0
13 J M 2 Bike/motor 2 (W),
tilted tree (H)
0 0 0 0
14 J M 2 Box (H), tilted tree (H) 0 0 0 0
15 J M 1 Box (H) 0 0 0 0
Mean 1.8 0.27 0
SD 0.7 0.5 0
Range 0-3 0-1 0
a G: graduate blindfolded sighted student; T: visually impaired teacher, S: senior high visually
impaired student, J: junior high visually impaired student; b M: male, F: female;
c The body site
collided by obstacles, H indicates head collision and W indicates waist collision. d
using both
white cane and detector in the experiment.
24
Table 4. Detector dimension, weight, mobility assistance and confidence satisfaction survey.
Values in parenthesis are the number of visually impaired participants.
Item Detector
dimension
Detector weight Mobility
assistance
Mobility
Confidence
Very satisfied 26.7% (4) 40.0% (6) 26.7% (4) 33.3% (5)
Satisfied 46.7% (7) 40.0% (6) 60.0% (9) 46.7% (7)
Acceptable 26.7% (4) 20.0% (3) 13.3% (2) 20.0% (3)
25
Buzzer
Warning Module
Control
host
Processing ModuleSensing Module
Obstacle
Transmitting
unit
Receiving
unit
Figure 1. The obstacle detector architecture which included three modules: a sensing module,
a processing module, and a warning module.
26
7.0 m 26.8 m (walking distance 33.4 m) 9.0 m
9.3
m4
1.3
m
11.2 m 20.4 m
4.5
m
S
F
12
3
4 5 6 7
8
9
10
2.2 m
Checkpoints:
Bike/motorcycle
Cardboard box
Plant terrace
Tall tree
Pedestrian overpass 1
Side-door
Bike/motorcycle
Tilted tree
1
F Finish pointS Start point
2
3
4
5
6
7
8
9
10
Turning point Check point
Pedestrian overpass 2
Pedestrian overpass 3
Figure 2. The route sketch for the visually impaired participants to travel in the second-stage
test
27
70%
63%
20%17%
0%
10%
20%
30%
40%
50%
60%
70%
80%
Automobile/
motorcycle
piled item/
protrusion
step Table/chair
Obstacle type
Subject
Figure 3. Four obstacle types that participants mostly often collided with while walking.
28
57%
40%
60%
50%
17%
33%
0%
10%
20%
30%
40%
50%
60%
70%
Head Hand/waist Leg
Body part
Subject
often collide with obstacles
mostly expect to avoid colliding
Figure 4. The percentage of body parts that often collided with obstacles and most expected
to avoid collisions. Hands/waist is expressed together because the hands and waist are usually
at the same vertical level while walking.
29
0
50
100
150
200
250
-60∘ -50∘ -40∘ -30∘ -20∘ -10∘ 0∘ 10∘ 20∘ 30∘ 40∘ 50∘ 60∘
Angle
Detection distance
Board
Man
Figure 5. Sensor detection envelopes obtained from two tests using a high flat wooden board
and three persons as targets.
30
A
301.0
A
52.1
A
333.0
A
79.4
B
257.7
B
44.1
0
50
100
150
200
250
300
350
400
Visually impaired Blindfolded normal
Participant
(s)
Cane
Detector
Both
Mean finish time
Figure 6. Means and standard deviations of finish time (FT) and the Duncan’s new multiple
range test by test condition for both the visually impaired and blindfolded sighted participants.
Means with the same letter are not significantly different (α=.05).
31
A
1.8
A
2.0
B
0.4B
0.3
0
0.5
1
1.5
2
2.5
3
Visually impaired Blindfolded normal
Participant
CaneDetectorBoth
Mean collision frequency
B
0.0
B
0.0
Figure 7. Means and standard deviations of collision frequency (CF) and the Duncan’s new
multiple range test by test condition for both the visually impaired and blindfolded sighted
participants. Means with the same letter are not significantly different (α=.05).