基于粒球计算的元宇宙自动驾驶障碍物识别方法与系统
By employing particle-sphere computation and graph neural network models, the problems of target recognition accuracy and virtual-real feature alignment in autonomous driving perception systems under complex environments are solved, achieving efficient obstacle recognition and virtual-real fusion, applicable to both real and metaverse driving scenarios.
Patent Information
- Application Number
- CN202511905628.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-17
- Publication Date
- 2026-07-17
- Estimated Expiration
- 2045-12-17
AI Technical Summary
Existing autonomous driving perception systems lack target recognition accuracy under extreme conditions such as complex road environments, sudden changes in lighting, and rain and snow. They also suffer from unstable cross-scale representation, scarce real data, limited coverage of extreme scenarios, difficulty in aligning virtual and real domain features, and challenges in deploying existing methods on vehicle-mounted devices.
An obstacle recognition method based on particle-sphere computation is adopted. Obstacle features are trained through a graph neural network model, and a graph structure is constructed by particle-sphere partitioning and hierarchical aggregation to achieve adaptive granularity representation of obstacles and efficient information propagation. The virtual and real features are aligned in conjunction with the metaverse simulation platform.
It improves robustness and fine-grained perception accuracy in complex environments, reduces computational costs, and achieves efficient obstacle recognition and virtual-real fusion, making it suitable for both real and virtual driving scenarios.
Smart Images
Figure CN121747072B_ABST
Abstract
Citation Information
Patent Citations
Fast K-nearest neighbor classifier method for large-scale data
CN116363420A
Self-adaptive three-branch K-nearest neighbor classifier based on density pellets
CN117828459A