Pedestrian Target Tracking Based On Cross-camera And Feature Online Learning

At present,cross-camera pedestrian target tracking has become a research hotspot in the field of computer vision,and has received extensive attention from researchers.Most existing cross-camera pedestrian target tracking methods are proposed under a specific camera network structure,which is not conducive to promotion to other scenarios.In response to this problem,this paper proposes a cross-camera pedestrian target tracking method.The main work of this paper is as follows:(1)Aiming at the problem that pedestrian recognition rate is not high,this paper proposes a target recognition method combining face recognition and pedestrian recognition.In the case where a human face can be detected,face recognition is used instead of pedestrian recognition,which compensates for the defect that the pedestrian recognition rate is low,and improves the recognition rate of the target to the system.(2)Aiming at the problem that the kernel correlation filter tracking algorithm can not scale adaptive tracking,this paper proposes an improved kernel correlation filter tracking algorithm.The algorithm not only satisfies the real-time requirements,but also makes the tracking more accurate.(3)In an actual camera network,there may be cases where part of the camera's field of view overlaps and some of the cameras' fields of view do not overlap.Aiming at this problem,this paper proposes a cross-camera pedestrian target tracking method for linear processing of video streams,which can simultaneously deal with crosscamera pedestrian target tracking problems in overlapping and non-overlapping fields of view.(4)Aiming at the cross-camera target tracking method proposed in this paper,this paper designs a software interface integrating recognition and tracking,and verifies the real-time and effectiveness of the algorithm through experiments.