一种RGB-E多模态目标跟踪方法及系统
By using multi-stage feature extraction and adaptive pooling networks, the problems of ignoring channel information differences and fixed pooling kernels in RGB-E multimodal target tracking are solved, achieving higher accuracy target tracking, especially improving feature extraction and tracking accuracy in sparse scenes.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JIANGNAN UNIV
- Filing Date
- 2025-09-26
- Publication Date
- 2026-07-17
AI Technical Summary
Existing RGB-E multimodal target tracking methods ignore the information differences between channels in their event feature extraction networks and use fixed pooling kernels to extract event features, which lacks adaptive perception capabilities and results in poor target tracking accuracy.
An event feature extraction network with a multi-stage feature extraction mechanism is designed. By combining 1×1 convolution with average pooling and max pooling operations, and combining deformable average pooling and max pooling, it adaptively captures different channel information of event images. It uses bilinear interpolation to calculate the offset feature values and combines Shannon entropy theory to adjust the event frame aggregation strategy to improve feature extraction accuracy.
By employing multi-stage feature extraction and adaptive pooling, the spatiotemporal dynamics of target motion are accurately characterized, noise event interference is reduced, and target tracking accuracy is improved. In particular, it effectively alleviates the feature dispersion problem in sparse scenarios.
Smart Images
Figure CN121190519B_ABST