The Design And Implementation Of Multi Object Tracking System For Intelligent Video Surveillance

Multi object tracking provides basic support for pedestrian affairs analysis in video surveillance and carries the core function in intelligent video surveillance system,which has important research significance.At present,the speed and accuracy of target detection in multi object tracking algorithm still have some room for optimization.In this paper,we focus on the improvement of the speed and accuracy of target detection,and the details of the research are as follows.First,a lightweight target detection model is designed with a lightweight feature extraction network as the starting point.The YOLOv5-based network model has a good performance in terms of detection accuracy and speed,but when it is deployed and used on end devices,it is often limited by the arithmetic power,which hinders the popularity of the model.Therefore,this paper improves the backbone network based on the Rep VGG model and achieves a lightweight design of the network.At the same time,this paper adds a coordinate attention mechanism to enhance the weight of the region of interest while expanding the perceptual field,which improves the robustness of the algorithm in dense crowds and complex environments.After experimental testing,the proposed lightweight network has a significant decrease in the number of parameters and computation,and the detection accuracy and robustness are also higher,which can meet the requirements of engineering applications to a certain extent.Second,the improved detection algorithm is combined with the ByteTrack tracking algorithm to improve the detection accuracy.The detection model used in the original ByteTrack tracking algorithm is YOLOX.In the experiments of this paper,the YOLOX detector has a high accuracy rate,but its relatively large number of parameters leads to the reduction of detection speed.In this paper,the improved lightweight YOLOv5 algorithm is applied to multi object tracking,and the balance between detection speed and accuracy is optimized,and the improved algorithm has better performance on the MOT17 datasets.The implementation and testing of the relevant detection and tracking algorithms in this paper are executed at the command line of the operating system.Considering the good human-computer interaction of the system,the algorithm design of this paper was implemented with a graphical user interface using Py Qt5.After testing,all functional modules run normally under the graphical interface,and the multi object tracking system is implemented,and the user can adjust the parameters conveniently with good user experience.