用于行人意图识别的多特征融合算法
By using a multi-feature fusion algorithm to extract spatiotemporal features from optical flow frames and image frame streams, and by utilizing LSTM and a robot learning module, the problem of insufficient generalization ability in pedestrian intent recognition is solved, achieving high accuracy and high interpretability in pedestrian intent prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NANJING UNIV
- Filing Date
- 2024-05-09
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies for pedestrian intent recognition suffer from insufficient generalization ability due to the use of single features or a single network structure, making it unable to effectively capture dynamic environmental changes, and also exhibiting weak interpretability and real-time performance.
A multi-feature fusion algorithm is adopted to extract spatiotemporal features by combining optical flow frames and image frame streams. The temporal and spatial information of video frames is extracted by LSTM and robot learning modules. Pedestrian intention prediction is performed by fusing traditional machine learning with shallow networks.
It improves the accuracy of pedestrian intent prediction, enhances computation speed and interpretability, and has strong transferability and computational resource optimization capabilities, making it suitable for pedestrian recognition in road scenarios.
Smart Images
Figure CN118608906B_ABST