towards auto-extracting car park structures: image processing approach on low powered devices

21
Towards Auto- Extracting Car Park Structures: Image Processing Approach on Low Powered Devices Ian K.T. Tan, Poo Kuan Hoong and Yap Chin Hong Multimedia University

Upload: poo-kuan-hoong

Post on 14-Apr-2017

247 views

Category:

Technology


2 download

TRANSCRIPT

Page 1: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Towards Auto-Extracting Car Park Structures: Image

Processing Approach on Low Powered

DevicesIan K.T. Tan, Poo Kuan Hoong and

Yap Chin Hong

Multimedia University

Page 2: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Introduction

Page 3: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Problem

• Time wasted for motorist to search for available car park space.– Leads to traffic congestion, wastage of resources

in the process.– Inefficient use of available car parks.

• Implementation of digital signages coupled with sensors are expensive and complicated– No effective delivery of this information from car

park management to the motorists.– Localized signage, only viewable in the vicinity.

• Integrated car park management system for open car park such as recreational parks or stadiums would be unjustifiable

Page 4: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Research Area

• There have been numerous interests in the area of detecting availability of car park bay using image processing techniques.

• An area that has been neglected in doing so is the initial calibration of the image capturing device. – On determining the car park structures

• We propose a technique that attempts to address this issue– using the limited processing capabilities of

embedded systems.

References 1 – 13.

Page 5: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Extracting Car Park Structures

• Considerable works have been conducted on extracting car park structures. However, many of them (Cheng et al., 2014) (Seo & Urmson, 2009) (Tong et al., 2014) (Tschentscher et al, 2013) are from images captured from an aerial view.– This will incur significant cost, and– Unrealistic for commercial implementations

• Another approach would be to conduct scene determination using object identification (Felzenszwalb et al., 2010) (Schneiderman & Kanade, 2004).– The computational resource requirements to

conduct object identification will not be easily achieved on low powered devices.

References 18 - 22.

References 23 - 24.

Page 6: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

MethodActual Image from the System

• Low cost (initial and maintenance) solution for open space car park using low powered devices,– Remove the need of manual mapping of the

car park structures for each installation of the camera.

– Reduce the need to re-calibrate the device due to movement.

• Two main process involved– Pre-processing to produce a binary image,

and– Car park structure extraction.

Page 7: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Hardware• Raspberry Pi Model B+• Raspberry Pi Camera• 8GB SD CardSoftware• Raspbian Wheezy• GNU C++ Compiler• OpenCV 2.4.4

Tools

Page 8: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Image Pro-processing

Original RGB Image

RGB Grayscale Conversion

Background subtraction

Denoise (smoother filtering,

thresholding, dilation filtering)

Binary image output

Page 9: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Extraction Method

HoughlineP Algorithm will detect numerous lines, from which they are used to determine the corners

A FitLine algorithm is then used to “connect the dots (corners)”. This is done for two axes.

From the lines, the FindSquare which is technically find the “trapeziums” in the images and finally a MergeSquare algorithm is applied which is heuristics based on adjacent “squares” that may be a single square.

Page 10: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Results

• Applied to images downloaded from the Internet for testing– Different image sizes– From approximately “lamp post” angle– The car park structures of interest are the

nearest rows only

• Results vary from 25% to greater than 90%

Page 11: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Current Limitations and Conclusion

• Robust solution but is still work in progress to achieve acceptable consistent accuracy rates.

• Current limitations of the solution– The image has to be taken at approximately right angles in order for the lines to

be drawn appropriately.– The angle of the image has to be taken from approximately between 30º to 60º.– The physical car park structure lines demarcations are required.

• Future Plans– Multi-camera aggregation method where results from multiple cameras are

taken into consideration in order to determine the car park structures.• Initial results shows improvement.

– In funding proposal stage for alternative approach.

Page 12: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Sample ProcessA more detailed output graphically of each stage of the process

Page 13: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Original Image

Page 14: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Binary Image

Page 15: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Pixel Segmentation Technique

Page 16: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Line Detection

Page 17: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Corner Detection

Page 18: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Linking the Corners

Page 19: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Square Box Mapping

Page 20: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Merged (from multiple cameras) Boxes Mapped to Original Image

Page 21: Towards Auto-Extracting Car Park Structures: Image Processing Approach on Low Powered Devices

Thank You