基于多模态异质性特征融合的自动驾驶决策方法及系统
By segmenting the features of vehicle driving trajectories and fusing multimodal features, the problem of insufficient modeling of driver behavior mode switching and discontinuous decision-making in existing technologies is solved, realizing a high-precision and highly adaptable autonomous driving decision-making method and improving the performance of intelligent connected vehicles and traffic simulation platforms.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TIANJIN UNIV
- Filing Date
- 2025-07-16
- Publication Date
- 2026-07-17
AI Technical Summary
Existing deep learning algorithms cannot effectively capture the driver's behavior pattern switching process in different driving states such as acceleration, deceleration and steady-state following in autonomous driving decision-making, and lack explicit modeling of driving state transitions and discontinuous decision-making processes, resulting in insufficient prediction accuracy and adaptability of the model in complex driving environments.
A segmentation algorithm is used to segment the vehicle's driving trajectory. The Newell stimulus-response theory and dynamic time warping algorithm are combined to identify the following flow state and the free flow state. An improved LSTM network is used to fuse multimodal features, and a time series prediction model is constructed through GRU to achieve real-time quantification and dynamic modeling of driving state.
It significantly improves the prediction accuracy and adaptability of the model in complex driving environments, enhances the scenario generalization ability of autonomous driving systems, provides a high-precision and highly adaptable car-following behavior modeling tool, and supports the optimization of intelligent connected vehicles and traffic simulation platforms.
Smart Images
Figure CN120735797B_ABST