基于光流引导及双流网络的动作识别方法、系统及装置
By constructing a pet dog action video dataset and improving the dual-stream network, introducing optical flow-guided features, and optimizing feature extraction and fusion, the problems of high computational cost and low accuracy in pet dog action recognition were solved, achieving a more efficient action recognition effect.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HANGZHOU YUNXIANG NETWORK TECH
- Filing Date
- 2022-10-31
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies struggle to effectively handle pet dog motion recognition tasks, especially in open environments with multi-scale, multi-target, and perspective-shifting issues. They also require significant computation and lack pet dog motion video datasets, resulting in poor performance of existing algorithms for pet dog motion recognition.
A dataset of pet dog action videos was constructed, the dual-stream network was improved, optical flow guided features were introduced, and temporal feature extraction was optimized. Feature extraction and fusion were performed through feature generation subnetwork, OFF subnetwork, spatiotemporal information fusion network and LK optical flow neural network, and the network architecture was optimized to improve recognition accuracy.
By improving the dual-stream network and optical flow-guided features, the accuracy of pet dog motion recognition was improved, the extraction of temporal features and the fusion of spatiotemporal features were optimized, and the recognition performance was enhanced.
Smart Images
Figure CN115631535B_ABST