The Pedestrian Target Detection Research For Intelligent Vehicle

Unlike pedestrian detection with static background, pedestrian detection with a varying background tends to be interfered by the environment, the illumination and the pose of pedestrian, hence is much more difficult. However, most vision applications related to intelligent vehicles belong to this category. The difficulty lies on how to extract moving targets from a changing background. Such research plays an important role on applications such as precision-guided weapons, vehicle patrol for traffic monitoring, intelligent vehicle and mobile robot.In this paper, the pedestrian target detection for intelligent vehicles was studied. Considering the difficulty of dynamic context pedestrian detection technology, this paper used a method that combined the global motion compensation and HOG&SVM pedestrian detector together. Firstly, extract Harris corner points of the adjacent images. Then, use NCC matching algorithm to obtain the feature points of the adjacent images. Thirdly, use the projection model of camera to determine the motion parameters. Then, use the motion parameters to compensate the video frames. Finally, obtain the possible moving target by calculating the difference of frames.We determined the pedestrian target by HOG&SVM classifier on the basis of moving targets detection in this paper. Moving target compensation differential was input to the classifier and pedestrian target was marked. The experimental results show the effectiveness of the method.