Research On HOG-based Pedestrian Detection System

Pedestrian detection is an important research area in the computer vision, which has an important significance to improve the quality of people's life. In the field of security, especially in building intelligent monitoring system, automotive driver assistance system and advanced human-machine interface, it plays an important role. Due to the diversity of background, the changing weather and light, pedestrian movement and posture's uncertainty, clothing diversity and other factors, it is difficult to implement pedestrian detection with respect to the other target detection. Besides, researchers also need to consider the real-time requirement.The paper mainly studies the classification and recognition of pedestrian combined with the pedestrian movement information in the scene of fixed camera, and the track to the detected pedestrian is also considered. The main work is as follows.First, background subtraction method is proposed to detect moving targets. Vibe is a moving target detection algorithm based on background subtraction idea. In the paper, the algorithm used in conjunction with Vibe and OSTU algorithm can quickly eliminate the area of Ghost, and remove the shadows moving target that may arise to get more realistic picture of prospect moving targets.Second, the paper extracts the HOG features of the detected moving objects as the feature descriptors. Due to the higher dimensions resulting in slower speeds insufficient, a PCA algorithm that an reduce dimension is used to get HOG-PCA feature descriptor,then classify training is used with SVM classifier. Experimental results show that this algorithm can improve the detection speed without reducing the recognition rate.In order to improve the real-time of video pedestrian detection, pedestrian tracking module is included. In the paper, a new algorithm combined CamShift algorithm with Kalman algorithm is able to solve the case of the occurrence of transient occlusion of the pedestrian and achieve real-time pedestrian stable tracking.Finally, the integration of all modules forms a video sequence pedestrian detection and tracking system, the experiments show that the proposed pedestrian detection system can achieve higher recognition rate and faster detection speed.