基于毫米波雷达和相机鸟瞰图融合的三维感知方法
By using a feature fusion method based on millimeter-wave radar and camera bird's-eye view, the problems of low coding efficiency and non-alignment are solved, achieving high-precision and robust 3D perception for autonomous driving, improving detection and segmentation accuracy, and making it suitable for a variety of computer vision tasks.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PEKING UNIV
- Filing Date
- 2024-03-11
- Publication Date
- 2026-07-17
AI Technical Summary
Existing multimodal perception methods using millimeter-wave radar and cameras suffer from low coding efficiency, unstable fusion, and an inability to effectively address the misalignment between millimeter-wave radar data and images, thus limiting the perception performance of autonomous driving systems.
Employing an efficient millimeter-wave radar backbone network and a deformable cross-attention multimodal fusion method, feature alignment and fusion of millimeter-wave radar and camera bird's-eye view images are achieved through radar cross-section discretization and cross-view feature projection. This is then combined with convolutional neural networks for 3D perception tasks.
It improves the 3D perception accuracy and robustness of autonomous driving systems, is applicable to a variety of computer vision tasks, enhances detection and segmentation accuracy, and maintains the real-time inference speed of the model.
Smart Images
Figure CN118038396B_ABST