3d detection method, device, medium and vehicle based on fisheye camera
By establishing a horizontal coordinate system in the fisheye camera and predicting only the yaw angle, the problems of high prediction complexity and large error in fisheye camera 3D detection are solved, and efficient and accurate 3D target detection is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- 安徽蔚来智驾科技有限公司
- Filing Date
- 2022-12-20
- Publication Date
- 2026-06-12
AI Technical Summary
The mounting method of fisheye cameras results in the 3D bounding box of objects in their camera coordinate system containing pitch and roll angles. Existing methods that directly predict yaw angles will produce large deviations, and predicting three Euler angles at the same time increases the complexity of the model and makes it difficult to correlate with the appearance of the image.
Establish a horizontal coordinate system that coincides with the origin of the fisheye camera's coordinate system. Predict only the yaw angle. Transform the yaw angle in the radar coordinate system to the horizontal coordinate system using a rotation matrix. Train a 3D detection model to obtain the target's 3D bounding box.
While ensuring prediction accuracy, it reduces model complexity and makes 3D detection results easy to correlate with image appearance, eliminating errors caused by ignoring pitch and roll angles.
Smart Images

Figure 1 
Figure 2