Research On Multi-camera Cooperative Target Tracking

With the advancement of science and technology and the continuous improvement of people's living standards,the incidence of public safety incidents has also increased.In order to effectively prevent crimes and provide clues and permits afterwards,a large number of surveillance cameras have been installed in crowded public places such as shopping malls,subway stations and railway stations.With the increase in the number of surveillance cameras,the traditional way of relying on manual retrieval of required data can no longer meet the actual needs.Therefore,it is necessary to design a multi-camera cooperative target tracking system in crowded scenes,which is of great significance to the prevention of major public safety incidents and commercial analysis.Multi-target tracking is an important part of surveillance video analysis.In recent years,with the continuous development of deep learning technology,researchers have applied deep learning technology to the field of multi-target tracking.It has been recognized because of its high accuracy.The real-time performance of some multi-target tracking method cannot be guaranteed,and because a single camera cannot provide a complete target tracking trajectory due to its limited monitoring range,our studies the target tracking problem of multi-camera cooperation.The main work is as follows: Components:(1)In order to alleviate the real-time problem in the multi-target tracking method and the difficulty in tracking the target due to too high appearance similarity and too many false detections during the tracking process,a multi-target tracking method is proposed,which is based on light Quantitative network is associated with hierarchical data.Because the lightweight network Mobile Net uses deep separable convolution to compress the original network to reduce network parameters,we use Mobile Net to replace the main structure of the YOLOv3 network while retaining the multi-scale prediction part of the YOLOv3 network.Realize to reduce the complexity of the network,make the method meet the real-time requirements.Finally,the experiment verified our proposed method.(2)At present,the development of multi-target tracking methods based on a single camera is relatively mature,but due to its small shooting range,the application range is limited.We use binocular vision theory to find the position of the target under different cameras by using the positional relationship of the imaging points between the cameras,and solve the problems of how the targets under multiple cameras are related,the target and the target occlusion,and the target and the background occlusion.There is no need to perform a global search on the image,which reduces the complexity of the calculation and reduces the calculation time.When communicating with the target,consider solving the multiple sets of feature points in the overlapping part between the cameras to obtain the relationship correspondence matrix of the corresponding pixels of the two images,and realize the tracking through the correspondence of the coordinates between the cameras.This method has high accuracy and simple implementation.,Suitable for transplantation in our design system.(3)According to the method we proposed and used,a multi-camera cooperative tracking and monitoring system was realized,and the system was deployed in the actual scene.The operation of the system proved that the system we designed was practical.