Research On Multi-view Single Target Tracking Algorithm Based On Siamese Network

This paper focuses on the Siamese network based multi-view single target tracking algorithm.In the current target tracking field,Siamese networks have become a very effective tracking method.However,there are still some challenges in cross-field target tracking,such as target deformation,scale variation and illumination variation between different viewpoints.Therefore,in order to improve the accuracy of cross-view target tracking,this paper proposes a tracking algorithm based on the Siamese network for recognition and localization based on target detection and re-identification methods.Firstly,in terms of the target tracking algorithm in a single field of view,this paper improves the tracking algorithm based on Siamese FC algorithm by introducing deformable convolution for its weak feature extraction capability,which improves the accuracy of the tracking algorithm.In order to ensure the tracking effect,a residual join is used to fuse the features of different convolutional layers to ensure its robustness in the tracking process.Secondly,a single Siamese network cannot accomplish better tracking when performing cross-view target tracking.To improve the accuracy of cross-view target tracking,this paper proposes a tracking algorithm based on target detection and target re-identification on the basis of Siamese network.In terms of target detection,this paper improves the yolov5 n algorithm by introducing a lightweight structure to reduce the number of model parameters,and introduces channel attention and spatial attention to improve the accuracy of cross-field target detection.In terms of re-identification,this paper adopts the reliable re-identification baseline algorithm Re ID Strong Baseline,which is used to improve the re-identification accuracy of targets.The target frame based on the detection algorithm is used as a priori information for the re-identification network when the target is switched in the field of view in order to find the target location in different fields of view.Finally,this paper integrates the above algorithms and develops a visual interface to implement a pyqt5-based multi-view domain single-target tracking system.