Vehicle Tracking Based On Siamese Network In Intelligent Video Surveillance System

In recent years,the importance of intelligent video surveillance system in urban security work has become increasingly significant and has important research value.In particular,tracking vehicles with drunk driving,causing accidents,and speeding abnormal behaviors in the traffic monitoring system,and reducing the loss of personnel and property caused by the escape of illegal and disciplinary vehicles,has always been one of the important functions of the intelligent monitoring system.However,it is affected by the similarity of the target and background caused by complex shooting scenes such as bad weather(cloudy,rainy,snowy),air pollution(smoggy),and dark night,which makes the vehicle tracking system more difficult.Traditional vehicle tracking algorithms cannot meet the needs of bad weather,and they are still inadequate in dealing with the scaling problems of moving vehicles.Therefore,this paper is based on the Siam FC tracking algorithm of the Siamese network to study and improve the target and background similarity caused by bad weather,the lack of scale estimation module and the missing model update module in the Siam FC algorithm,And add a YOLO detector to the front end of the tracker,Realize fast,automatic and efficient vehicle tracking algorithm.The main content of this article is as follows:1.Aiming at the problem of similar target and background caused by environmental impact in vehicle tracking,this paper proposes an efficient feature fusion method based on twin network.Use the VGG-16 network to extract the apparent and semantic features of the target vehicle,assign weights to the apparent and semantic features,and obtain a relatively complete feature map,which can distinguish between the target and the background and the same category For the target vehicle,a large number of redundant features are removed to facilitate subsequent similarity comparison.At the same time,in view of the lack of a model update module in the Siam FC algorithm framework,which leads to the problem of reduced algorithm stability,an improvement plan is proposed to continuously improve the stability of the algorithm by training the update threshold.Experiments prove that the algorithm proposed in this paper can effectively improve the success rate and accuracy of the tracking algorithm,and the overall performance of the algorithm has been improved.2.Aiming at the problem of many scale transformations in vehicle tracking,this paper proposes a tree scale estimation method based on correlation filtering.Use one-dimensional correlation filtering to first determine whether the current picture is zoomed in or zoomed out.After judgment,if zoomed in,the picture will be assigned to the zoom factor for factor matching(scale matching).Otherwise,it will be assigned to the shrinking party for factor matching(scale matching).This method of using correlation filters can not only efficiently solve the impact of scale scaling transformation on vehicle tracking,but also reduce the computational complexity and improve the real-time performance of the algorithm.Finally,in order to reduce human intervention and reduce the complicated operations caused by manually marking the initial frame of the tracking video,this paper adds a YOLO detector to the first part of the tracking algorithm to complete the automatic frame operation of the tracking target,which meets the requirements of vehicle tracking in the real market.Competitiveness also improves the real-time performance of the algorithm.The verification of the above methods through multiple sets of experiments confirms that the vehicle tracking algorithm proposed in this paper effectively solves the above problems,has good performance in terms of efficiency and accuracy,and meets the needs of practical applications.