Research And Application Of Video-based Wandering Detection And Tracking Algorithm

To detect whether a pedestrian is wandering,we first need to identify the pedestrian and then track the detected pedestrian,thereby determining whether the target is wandering through trajectory analysis.However,the current algorithm is difficult to meet the needs of high-accuracy and real-time detection.There are three main problems:(1)The existing detection algorithm has limited ability to extract small target features,and there are missed detections when the target is occluded,which is not conducive to the multi-feature fusion of the tracker,and affect the performance of tracking.(2)When the target is occluded in traditional tracking algorithms,the tracking interruption will cause the ID to jump,which will lead to the misjudgment of the wandering detection algorithm.(3)The existing loitering detection algorithms have a large amount of computation and are difficult to detect in real time,and have poor ability to discriminate loitering behaviors under complex paths,which is prone to misjudgment.At present,the loitering detection algorithm performs well in the field of security,and the existing fire detection technology can only detect after the event,but cannot predict in advance and give early warning of malicious arson behavior.Therefore,it is crucial to design an arson early warning and monitoring system based on the loitering detection and tracking algorithm.In order to solve the above problems,this paper studies from three aspects:improving the accuracy of the detector and tracker,improving the accuracy of the loitering detection algorithm,and building an arson warning and monitoring system.The main contents are as follows:(1)Aiming at the poor feature extraction ability of small targets and the ID jump problem in traditional tracking algorithms under the condition of occlusion,a pedestrian tracking model based on multi-feature fusion is proposed.On the basis of the YOLOv5 algorithm,the feature enhancement module and DIo U＿nms algorithm are added to the backbone network,and the apparent feature extraction branch is added to the prediction layer.After the improvement,the m AP value of the algorithm reaches 92.82%,and the detection accuracy of the detector is improved.The Deep SORT algorithm adopts the secondary training strategy of the apparent feature extraction branch,which improves the apparent feature extraction ability,adds a trajectory correction scheme,and solves the ID jump problem.After the improvement,the MOTA value is 69.3%,and the ID＿SW value is 461.(2)Aiming at the slow detection speed of traditional loitering detection algorithms and easy misjudgment for complex paths and stationary targets,a loitering detection algorithm based on path displacement time is proposed.When the total movement distance of the target is greater than the displacement and the movement time is greater than the time threshold,it is determined that the target is wandering.The algorithm in this paper realizes the requirement of real-time judgment and improves the accuracy of wandering detection.(3)Aiming at the problem that the existing fire detection technology can only detect after the fact and cannot predict beforehand,an arson early warning and monitoring system based on the wandering detection and tracking algorithm is constructed.The system analyzes the video stream and issues an arson alarm when a lingering pedestrian with a fire source is detected.The administrator can view the alarm information through a graphical interface,and make additions,deletions and modifications to the information.