The Applied Research For Snake Models And Stereo Vision In Pedestrian Detection

With the rapid development of China’s economy, the incidence of traffic accidents is rising as the vehicle quantity increases. Therefore, it is a very urgent matter to improve the degree of intelligence for vehicle. As a key technology of intelligent vehicle and Advanced Drive Assistance Systems, vision-based pedestrian detection has been paid great attention. At present, it has become the front research topic in the field of intelligent vehicle, computer vision and pattern recognition, etc.-because of its important role in vehicle safe driving and pedestrian protection.The process of pedestrian detection mainly contains two states of work: region of interest segmentation and objects recognition. In this paper, stereo vision technology will be used to ROI segmentation and contour curve is used to object recognition as an important feature. In the process of pedestrian detection implementation, the main work is as follows:1、An ROI segmentation method based on stereo vision be established: firstly, the disparity map be extracted for stereo images, then it be projected to the XZ plane according to the disparity of every points, meanwhile the initial segmentation for targets be completed in this plane; finally the segmented targets be re-projected to original image or disparity map, i.e. the process of ROI segmentation be finished.2、In the process of extracting disparity, a stereo matching method based on edge index be used and which has better effect on ROI segmentation compared with region-based stereo matching method.3、During initial segmentation, in order to eliminate the influence of road noise(such as road markings, shadows, etc.) and high altitude targets(such as trees, etc.), a height-based target segmentation method on depth map be proposed.4、Presented here a new method for contour curve extraction, i.e. an edge disparitybased segmentation region in the stage of ROI segmentation is acquired firstly, then the contour curve is extracted for targets using Snake Models.5、Analyzed the advantages and disadvantages of several Snake Models in MATLAB environment about the ability of contour extraction. As the represents of Snake Models, GVF Snake model have some drawbacks such as noise-sensitive, high computational complexity, etc.-which seriously affect the accuracy and timeliness for contour extraction. Accordingly, presented an anti-noisy Snake Models based on GVF models which can effectively overcome these drawbacks.6、Finally, the methods proposed in this paper are tested in MATLAB environment through many stereo images for typical traffic scenes and get ideal results.