基于非线性特征提取的路侧端雷达与摄像头融合的三维目标检测方法、设备、介质
By introducing Kolmogorov-Arnold networks (KANs) to improve image and point cloud encoders, and combining multi-head cross-attention and KAN kernel convolution, the nonlinear feature fusion capability of roadside multimodal 3D perception is enhanced, solving the problems of insufficient robustness and generalization ability in existing technologies, and realizing high-precision detection in complex traffic scenarios.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTHEAST UNIV
- Filing Date
- 2025-08-11
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies lack effective nonlinear feature fusion mechanisms in roadside multimodal 3D perception, resulting in insufficient robustness and generalization ability, especially limiting detection accuracy in complex traffic scenarios.
Kolmogorov-Arnold Networks (KANs) are used to improve image and point cloud encoders. Cross-modal weights are dynamically calculated through multi-head cross-attention, and nonlinear fusion is performed using KAN kernel convolution, thereby improving the feature extraction and fusion capabilities of high-dimensional data.
It significantly improves the 3D detection accuracy in complex traffic scenarios, enhances the ability to perform high-dimensional nonlinear modeling of image texture semantics and point cloud geometric features, optimizes cross-modal fusion strategies, and eliminates fusion degradation phenomena.
Smart Images
Figure CN121074864B_ABST