一种基于深度学习的端到端的多飞行器跟踪方法
By employing an end-to-end deep learning approach, combined with a dynamic multi-scale spatiotemporal network and a global-local extraction module, the problems of occlusion and recognition errors in multi-target tracking models during airport surface surveillance were solved, achieving higher tracking accuracy and stability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- UNIV OF ELECTRONICS SCI & TECH OF CHINA
- Filing Date
- 2023-12-22
- Publication Date
- 2026-07-17
AI Technical Summary
Existing multi-target tracking models cannot effectively utilize the temporal information between video frames in airport surface surveillance, leading to aircraft target occlusion and identification errors. Furthermore, traditional methods cannot extract global and local features simultaneously, resulting in decreased tracking stability.
We adopt an end-to-end approach based on deep learning, extracting the spatiotemporal fusion features of the current frame and keyframes through a dynamic multi-scale spatiotemporal network and a global-local extraction module. Combining the representational capabilities of the Transformer model, we use a deep similarity network for target association and matching, abandoning traditional constraints and directly using detection embeddings and tracking embedding vectors for similarity calculation.
It effectively solves the problem of aircraft target obstruction, improves the robustness and stability of multi-target tracking, and enhances tracking accuracy in airport surface surveillance.
Smart Images
Figure CN117765027B_ABST