System Integration and Optimization for 6D-Vision Vehicle Collision-Avoidance

• Main Authors: Yu-Chih Lu, 盧昱志
• Other Authors: Yeong-kang Lai
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/46987557460840313991

碩士 === 國立中興大學 === 電機工程學系所 === 102 === Our main purpose is to identify the object''s distance and direction by using the dual-lens camera which can handle the stereo environment. The images are presented in 1920 × 1080 resolution in the paper, so the computing time is large. The computing time to perform disparity generation is the most critical by using ARM DS-5 analysis. Therefore, the resolution reduction and the Sobel edge detection are used to reduce computation time. Then, the value of the disparity will be obtained by our proposed "diamond census adaptive support weight". Later, the value of disparity can be converted to depth information of the object''s distance. Also, with the depth information, the distance of moving object can be detected by experimenting an object''s self-moving. Our proposed methods are implemented and optimized on embedded system platform, so we use a lot of system optimization techniques such as parallel processing, simplified loop, induction variable elimination, memory allocation management and so on. With these achievements above, the function requirements can be achieved. In order to make use of the embedded platform device in the car, an embedded platform is needed to display the image handled by collision avoidance system. The embedded platform is built with a Linux system called Qt creator. The result of the collision avoidance system can be displayed based on the connecting signals and slots of objects. The driver can easily recognize the obstacles on the road due to the dynamic update of the image display system.