Depth Estimation of the Potholes With Stereo Machine Vision

Lim, Chu Hang (2018) Depth Estimation of the Potholes With Stereo Machine Vision. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text LIM CHU HANG - FULL THESIS.pdf Restricted to Registered users only Download (7MB)

Abstract

This project dissertation had proposed a method to estimate the depth of the potholes. A pothole is a pit or a hole with bowl shaped and the diameter with more than 150 mm on a road surface. It is caused when an asphalt pavement is experiencing the road surface distress. A pothole could be formed by the erosion of rock, it is usually caused by the action of water and climate changes. As according to MIROS, the road defects is about 10% of overall crash contributing factor throughout the year 2007 to 2013. Which the pothole is the most common road defects. Moreover, there are too much time and labour cost on the measurement of the pothole. Furthermore, the most updated road condition cannot be known without the pothole detection technology. Consequently, the depth estimation of the potholes is very important to solve all of these problems. In order to estimate the depth of the potholes, the stereo vision cameras were used. Two black-box cameras were used to record the stereo videos of the potholes with the power supply from the car charger. After obtaining the stereo videos of the potholes, the videos were converted into the image frames. Then, the stereo images of the pothole were processed to estimate the depth of the pothole. After estimating the depth of the potholes, the error percentage on the depth estimation of the potholes was calculated. Although there was the depth estimation on the real pothole in this project, the error on the depth estimation of the real pothole is rather higher. The error percentage on measured height of the stereo vision cameras from the pothole is around 70%. Since the height of the stereo vision cameras was having such high error percentage, the depth estimation of the pothole is also inaccurate. The error percentage on the depth estimation of the pothole is about 82.25%. Therefore, instead of using the real potholes, the other objects were used for the depth estimation. These objects are the trays, a bowl and a bathtub. The error percentages of estimating the depth of these objects are much lower than that of the real potholes. The object having the lowest error percentage is the white tray, the error percentage is about 11.19%. The object having the highest error percentage is the bathtub, the error percentage is about 25.27%. By averaging the error percentage, the mean error percentage of the depth estimation is about 18.94%. In other words, the accuracy of the system is about 81.06%.