2D vision detection vehicle target positioning method
By matching the pixel coordinates of the vehicle prediction bounding box, the headlight prediction bounding box, and the grounding key point, combined with the Hungarian algorithm and camera calibration parameters, the problem of vehicle target positioning error caused by fisheye camera imaging distortion was solved, and the positioning accuracy was improved.
CN116862996BActive Publication Date: 2026-07-14BEIJING YINWO AUTOMOBILE TECH CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- BEIJING YINWO AUTOMOBILE TECH CO LTD
- Filing Date
- 2023-07-14
- Publication Date
- 2026-07-14
AI Technical Summary
Technical Problem
In existing technologies, the imaging distortion of fisheye cameras leads to large errors in vehicle target localization, especially when localizing vehicle-type targets.
Method used
The model outputs the pixel coordinates of the vehicle prediction box, the headlight prediction box, and two grounding key points. The vehicle and headlight information are matched with the Hungarian algorithm, and the camera calibration parameters are used to convert them into world coordinates to reduce the impact of distortion.
Benefits of technology
This improved the accuracy of vehicle target positioning, reduced errors caused by distortion, and achieved more accurate vehicle position calculation.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN116862996B_ABST
Abstract
The application discloses a 2D visual detection vehicle target positioning method, comprising the following steps: 1) outputting a vehicle prediction frame, a vehicle light prediction frame and two ground contact key point pixel coordinates through a model; 2) according to the above information output by the model, if there is vehicle light information, matching the corresponding vehicle target for each vehicle light; 3) according to the vehicle prediction frame, the vehicle ground contact point and the vehicle light prediction frame information, calculating the positioning point in the pixel coordinate system of the vehicle; 4) converting the positioning point in the pixel coordinate system into a world coordinate through the internal parameter and the external parameter generated by camera calibration, and taking the world coordinate point as the world coordinate point of the target vehicle. In the 2D visual detection vehicle target positioning method, the vehicle prediction frame, the vehicle light prediction frame and the two ground contact key point information are output through the model, the target position is analyzed, and under the condition that the 2D language diagram is not distorted, the vehicle category target positioning precision is improved.
Need to check novelty before this filing date? Find Prior Art