Design And Implementation Of Video Target Tracking System Based On Siamese Neural Network

With the development of our society,people have higher requirements for security capabilities.The video surveillance system is an important means to ensure social public safety,enterprise production safety,and family anti-theft safety.Target tracking,as one of the most basic and important functions in the video surveillance system,provides powerful assistance for various needs such as vehicle driving violations and suspicious person investigation.At the same time,target tracking is also the basis for analyzing video content.Accurate target tracking and positioning combined with other related algorithms can realize the intelligent analysis and understanding of video target behavior.Therefore,the research on video target tracking technology is of great significance.In this article,through analysis of the Fully Convolutional Siamese Networks for Object Tracking algorithm(Siam FC),it is found that Siam FC is susceptible to interference from complex backgrounds or similar objects during the tracking process,resulting in target loss.The main reason is Siam FC only uses appearance information of the target as the tracking basis,and the appearance information is easily changed due to the influence of the target itself or the external environment.To this end,a dual-branch siamese neural network tracking algorithm based on dynamic weights is presented.On the basis of Siam FC,a tracking branch using semantic information is added as an effective supplement to the tracking branch of appearance information.At the same time,the effect of the attention mechanism on feature extraction is analyzed,and a dual attention mechanism is added to the semantic branch,which strengthens the semantic feature extraction from the two dimensions of spatial location and feature channel.And through the visual analysis of the tracking results of the two branches,a branch combination method based on dynamic weights is given,which effectively combines the advantages of the two branches for tracking.The algorithm has been experimented on numbers of standard target tracking datasets,and the effectiveness of each module is verified by ablation experiments.Proposed algorithm also compared with other excellent tracking algorithms through comparative experiments.The results show that the proposed algorithm has good results and performance for video target tracking tasks.In addition,according to the needs of practical application scenarios,a video target tracking system is built by using the proposed tracking algorithm,Django back-end framework and front-end development technology.The system is based on the B/S structure,video surveillance module,target tracking module,user management module,and camera management module are designed and implemented which improves the efficiency of video analysis.