Pedestrian Recognition Aigorithm Base On Visual Sensor In Vehicle Pedestrian Collision Avoidance System

In recent years, along with the fast economic growth and the improvement of people’s living standards, the amount of cars are growing rapidly in our country, the traffic accidents caused by pedestrian are arising. On these conditions, The car-pedestrian collision avoidance technology is received widespread attention. As an effective means of reducing traffic accidents and reducing accident losses, the technology is representing the future trend of vehicle development.This dissertation starts from the traffic safety situation in China, using the computer vision to do some research about the key issue that detecting the outside-car pedestrian.The most representative of the pedestrian recognition algorithms is in-depth discussed, from based on the characteristics method, based on the method of the many parts of the body, based on the multi-angle method to based on boosted cascade method are analyzed and compared. The conclusion is made that these methods are starting to solve different problems from different point of views, are complementary. An ideal pedestrian detection system is proposed that it should be these organically combined together in order to achieve the best performance, thus a variety of pedestrian detection algorithm fusion is put forward.The pedestrian detection algorithm based on boosted cascade objects detection is proposed. In this algorithm, introducing a chain structure of the cascade classifier to disseminate the historical information of the cascade classifier to the next cascade of classifiers, and finally get the classifier consists of fewer weak classifiers and has lower error rate than the original cascade method. The experimental results show that this algorithm framework of general is very strong, is not limited to an specific object detection, is applicable to the pedestrian detection.Following, the popular moving body detection algorithem and recognition are study, by comparing the different algorithem we find out the advantages and disadvantages, then we build a model for optical flow algorithem to capture the moving human body base on the Matlab environment. The experiment show that the method can effective capture pedestrian on moving, reciving the wanted result. Althought this step is not the ultimate step to recognize the behave of human, But it can help to extract accurate body contour, and to track the parameters of the pedestrian, and is the basis of the recognition. Directly determines the success or failure of the human body detection.In the final part, in view of the video images collected by the car camera is jitter and the background is complex, etc. we discuss the pedestrian detection method based on the dynamic background of the adjacent odd frames, reducing the time of extracting dynamic background by a half, effectively improving real-time performance of pedestrian detection system. Firstly the video images are processed using mathematical morphology, and then use the Canny operator to extract the pedestrian profile curve. Finally the pedestrian profile curve is optimized by curve fitting techniques, obtaining a comprehensive pedestrian contour curve, eliminating the little island, fitting discontinuous pedestrian contous, the effect is quite satisfactory. laid the foundation for subsequent contour-based pedestrian recognition, and laid a solid foundation for the follow-up to identify the pedestrian.