Development Of Algorithms For Vehicle Classification And Speed Estimation From Dynamic Scenes By On-Board Camera Using Image Processing Techniques

• Main Author: Aminuddin, Nur Shazwani
• Format: Thesis
• Language: English; English
• Published: 2018
• Subjects: T Technology (General); TA Engineering (General). Civil engineering (General)
• Online Access: http://eprints.utem.edu.my/id/eprint/23257/1/Development%20Of%20Algorithms%20For%20Vehicle%20Classification%20And%20Speed%20Estimation%20From%20Dynamic%20Scenes%20By%20On-Board%20Camera%20Using%20Image%20Processing%20Techniques.pdf; http://eprints.utem.edu.my/id/eprint/23257/2/Development%20Of%20Algorithms%20For%20Vehicle%20Classification%20And%20Speed%20Estimation%20From%20Dynamic%20Scenes%20By%20On-Board%20Camera%20Using%20Image%20Processing%20Techniques.pdf; http://eprints.utem.edu.my/id/eprint/23257/; http://plh.utem.edu.my/cgi-bin/koha/opac-detail.pl?biblionumber=112237

Vehicle assistance system applications benefit the drivers and passengers to promote better and safer driving situations. In terms of usability of dash camera, most vehicle owners pre­ installed the camera as a personal safety purpose to record the path they went through. The wide availability of various models of the dash cameras on the market, however, lacks in intelligence to process the information that can be obtained from the camera technology system itself. Moreover, in most studies for Intelligence Transport System (ITS), the implementation of static camera, for example CCTV, is popular thus, it is an encouragement for improvement to develop a vehicle assistance system using dynamic camera scenes. The main purpose of this research was to develop a vehicle detection, vehicle classification, and vehicle speed estimation system in dynamic scenes fully by image processing technique. The scope of this research covered Malaysia highway in Skudai, Johor; Ayer Keroh, Melaka and Kajang, Selangor. Video database of these highway areas was recorded by the on-board camera unit placed on the front dashboard area of the host vehicle. Image dataset was collected with positive image sets containing four vehicle classes namely car, lorry, bus, and motorcycle. It was decided that the technique for vehicle detection were Haar-Like and Cascade Classifier while vehicle classification was based on the ratio characteristics of the vehicle detected. The use of ratio value was an added advantage for the classification process since the prepared image dataset were based on each vehicle class dimension and the ratio value are the uniqueness property for each vehicle class. Speed estimation of the vehicle started with host vehicle speed estimation by lane detection technique since the road lane was the most consistence moving object inside the video region. The Host vehicle distance measurement used the broken lane detection and for a scale factor calculation, the width of the highway lanes was calculated by measuring the lane width inside the image and calibrated with real value in meter of the lanes stated by (Jabatan Kerja Raya, 1997). Detected vehicle speed measurements were based on its centroid tracking measurements. Result analysis on accuracy measurement in vehicle detection system obtained 0.93 true positive rates from 300 vehicles presented in the video data. Further analysis in vehicle classification was proved to obtain true positive rate of 0.98 for car class, 0.89 for lorry class, 0.89 for bus class, and 0.75 for motorcycle class. For analysis of speed estimation achieved with the average percentage 6.42% for speed error of host vehicle tested on 10 different videos. In detected vehicle, it speed estimations were based on the host vehicle speed calculation by observation its position and motion behavior in comparison with the host vehicle speed value. Overall the e development indicated that image processing has the ability to visualize the surrounding area for drivers and passengers that was near to real human visions a contribution to human-machine interactions that can be beneficial.