Research And Implementation Of Pedestrian Detection Technology Based On Edge Feature

With the development of economy, the number of vehicle is increasing. The traffic problem becomes more obvious though the road is expanding, and the traffic accidents happen frequently. In the traffic accident, the pedestrians hardly dodge the vehicle when they encounter danger. So the pedestrian detection technology has been gotten the attention of automobile manufacturers and consumers. Pedestrian detection can accurately detect and predict the position and motion direction of the pedestrian. Based on this information it can judgment the threat to pedestrian, the warning system will warn the driver and even make the emergency treatment. So the accidents are avoided and the pedestrian detection can effectively improve the urban traffic safety.After analyzing the traditional algorithms, this paper researched and implemented the pedestrian detection ahead of vehicle algorithm based on vision sensor. First, in the state of pedestrian candidate regions segmentation, this paper designed a method of segmenting by the symmetry of pedestrian’s vertical edge. This method measured the symmetry of the image to extract the candidate symmetry axis, and used the symmetry axis to locate the pedestrian, and the pedestrian candidate regions were gotten by edge and width to height ratio of the pedestrian. Then in the state of pedestrian recognition, this paper used SVM and HOG feature to recognize pedestrian. This method extracted the HOG feature of pedestrian and trained the feature vectors which were input to SVM classifier to get a pedestrian classifier. At last, this paper researched the traditional Mean-Shift target tracking algorithm, and improved this algorithm. This paper joined the relocation process in the tracking, first the algorithm judged the target whether lost by the moving of centroid of the prediction region and motility of the target judged by frame difference method. And then the algorithm judged whether the target was lost, if it lost then relocated the lost target. The improved algorithm can decrease the loss rate and to removal the non-pedestrian which is error recognized and to increase the rate of the recognition.Experiments show, the algorithm which based on the edge symmetry of pedestrian can fine segment the pedestrian candidate region before the vehicle, and the improved tracking algorithm can make the tracking and recognition more accurate. At the same time, this algorithm has better robustness and efficiency.