基于双重更新策略与互指导损失的视频目标跟踪方法及系统
By employing a dual-update strategy and a cross-guided loss method, a video target tracking model is constructed, which solves the problems of poor template update robustness and inaccurate bounding box prediction, and achieves efficient tracking in complex scenarios.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XIDIAN UNIV
- Filing Date
- 2025-08-11
- Publication Date
- 2026-07-17
AI Technical Summary
Existing visual target tracking technologies suffer from poor robustness of template update strategies when faced with complex scenes. Single-stage methods struggle to adapt to target changes, while two-stage methods achieve high target confidence scores but low bounding box accuracy, resulting in trackers failing to output accurate bounding boxes.
By employing a dual update strategy and mutual guidance loss, a video target tracking model is constructed. This model utilizes a feature extraction network, a joint feature modeling module, a target query module, and a classification and regression module. By combining the dual update strategy and mutual guidance loss, the model continuously adapts to target changes and improves the accuracy of bounding box prediction.
Without reducing the tracking speed, it significantly improves the accuracy and robustness of target tracking, especially maintaining stable tracking performance in long-term tracking and complex scenarios.
Smart Images
Figure CN121033726B_ABST