Research On Technology Of Road Recognition Based On DSP

The technology of Vision-based Road Recognition is the basic of automatic to realize the anti-collision function which is used on Intelligent Car; it is also a hot field of automatic control in and abroad. In this paper, according to TI's DSP processor TMS320DM6437, we built the hardware platform for Automotive Vision. Through theoretical analysis, simulation experiments and practical application we made research on reliability and real-time about the Road Recognition algorithms.Firstly, the paper briefly introduced the development status of Intelligent Car in and abroad, and emphasized on the domestic research institutions and their results on the Road Recognition. Against the problem hadn't be done, we raised the purpose and significance of this paper.Our research was focusing on the road lane detecting, and the main improvement was on the real-time capacity. While in the image preprocessing stage, with analysis on the characteristics of road, we used the Gray Transform and Image Effective Region Segmentation methods to reduce the amount of video information. Then we made improvement on the Median Filter and Sobel edge enhancement algorithms to enhance the algorithm's efficiency, and we also selected to make video contrast enhancement to improve the differences between characteristics of the target object and background. According to the characteristics of road and preprocessing algorithms we used, while in the road lane fitting stage, we used a histogram-based threshold method to obtain the best threshold to split the video, which can reduce complexity of the algorithms. Then we combined Hough transform and least squares method together to fit road lane which can meet the requirement of reliability and real-time, the same time we used lane tracking algorithm to predict the Region of Interest, it can reduce the amount of data processing about fitting algorithms.During our research, the simulation to algorithms was against the road simulate video sequences, PET2001, and it was done on VC++platform with the help of Open source Computer Vision (Open CV) library. It was to verify the correctness of our theoretical analysis. Finally, according to the practical platform-TMS320DM6437 hardware system, we program the algorithms which had been verified into C language, and according to the characteristics of DSP processor's architecture and compiler's performance, we do optimizing on the algorithms program to improve the implementation and compiling real-time capacity of the algorithms.