Pedestrian Recognition Under The Conditions Of The Single-camera Tracking Method

The detection and tracking of moving pedestrian is a key techn ique in pedestrian safety state recognition system under mixed traffi c condition. It has great significance in improving the safety of urb an pedestrian traffic. In this paper, aiming at the problem of pedest rian detection in pedestrian safety state recognition under complex c onditions, a method for detection and tracking of moving pedestrian with monocular camera is proposed, and the feasibility of the prop osed method is validated with experiment. The main contributions o f this dissertation are:1. Using the combination of the adaptive background model m ethod and the motion segmentation method, the two value image of pedestrian is extracted. The Otsu threshold value segmentation is t hen performed based on the extracted two value image and the seg mented image is morphological processed in order to detecting the moving pedestrian. This algorithm is insensitive to noise and has re lative strong adaptability; it can realize efficient segmentation of mo ving pedestrian.2. According to the location characteristic of the pedestrian in the video, using the Blob merging method, the Blob of the same p erson in the segmented image is merged and the location characteri stic of the pedestrian is extracted. The location characteristic after merging is then detected. This method reduces the false detection p robability of the pedestrian efficiently. A Kalman filtering prediction and tracking method is performed with the extracted location chara cteristic as the eigenvector. With the Kalman filtering the pedestrian prediction variable is simplified. The prediction value and the dete cted pedestrian location characteristic are them matched to archive p edestrian tracking. This method demonstrated superior detection and tracking results in the experiment.3. Through collecting two video with complex background, the proposed method is used to perform real time analysis and experi ment, the results shows the proposed method can detect and track s ingle or multiple moving pedestrian in the video. It demonstrates th e proposed method is feasible.