A Lane Detection System Based on Sequential Simulated Annealing Algorithm

• Main Authors: WU,MENG-JU, 吳孟儒
• Other Authors: SU,CHIEN-KUN
• Format: Others
• Language: zh-TW
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/99607571902726759459

碩士 === 中華大學 === 電機工程學系 === 104 === A lane detection system that is based on the sequential simulated annealing algorithm is proposed in the thesis. Experimental results show that the correct recognition rate of the sequential simulated-annealing-algorithm-based lane detection system can reach 90%. Compared to the popular Hough transform algorithm, the simulated annealing algorithm uses fewer memories. It consumes about half of the memories used in the Hough transform method. Therefore, the proposed system can alleviate the problem of a large memory demand in Hough-transform-based systems, and it is of much potential in commercial applications. In our research, a dash cam installed in a car was used to record the road images in front of the car. Day-time and night-time still color images on freeways are used as the inputs of the proposed system. The research goal is to effectively detect the lane side lines to be used for improving driving safety. The pre-processing technique in the proposed system includes color to gray-level transformation, binarization, Sobel edge detection, morphology operations (erosion and dilation), and noise removal. After pre-processing, a binary lane image with some noise is obtained, and the sequential simulated annealing pattern detection method is applied for detecting lane position. The sequential simulated annealing pattern detection method can determine a set of parameter vectors with global minimal error. The detected patterns are removed from the binary image, and the remaining binary image is continue to be processed by using the sequential simulated annealing pattern detection method until all patterns are processed completely.