Methods and systems to compensate for vehicle calibration errors
Methods and system for compensating for vehicle system errors. A virtual camera is added to vehicle sensor configurations and a coordinate transformation process which attempts to match multiple 3D points associated with a landmark to a detected landmark. The virtual camera is associated with the detected landmark. The 3D world coordinate points may be transformed to a real 3D camera coordinate system and then to a virtual 3D camera coordinate system. The 3D points in real and virtual camera coordinate frames are projected onto the corresponding 2D image pixel coordinates, respectively. Inclusion of the virtual camera in the coordinate transformation process presents a 3D to 2D point corresponding problem which may be resolved using camera pose estimation algorithms. An offset compensation transformation matrix may be determined which accounts for errors contributed by mis-calibrated vehicle sensors or systems and applied to all data prior to use by the vehicle control systems.
1. A method for compensating for autonomous vehicle (AV) system errors, the method comprising:
obtaining multiple 3D points for an object;
transforming the multiple 3D points to a camera frame;
obtaining actual 2D points for the object;
establishing a virtual camera corresponding to the actual 2D points;
determining an offset compensation matrix between a camera associated with the camera frame and the virtual camera;
applying the offset compensation matrix to data points prior to use by vehicle control systems; and
controlling operation of the AV with offset compensated data points.
2. The method of claim 1 , wherein the determining the offset compensation matrix further comprising:
executing a camera pose algorithm using the multiple 3D points in the camera frame and the actual multiple 2D points in a virtual camera frame associated with the virtual camera to determine the offset compensation matrix.
3. The method of claim 1 , wherein the offset compensation matrix includes translational matrices and rotation matrices.
4. The method of claim 1 , wherein the multiple 3D points are in a world coordinate frame.
5. The method of claim 4 , wherein the multiple 3D points in the world coordinate frame are transformed to an inertial measurement unit (IMU) frame, from the IMU frame to a Light Detection and Ranging (LiDAR) frame, and from the LiDAR frame to the camera frame.
6. The method of claim 1 , wherein the actual 2D points are determined using at least one of image detection, object detection, and corner detection.
7. The method of claim 1 , wherein physical properties of the virtual camera and the camera are the same.
8. The method of claim 7 , wherein intrinsic parameters of the virtual camera and the camera are the same.
9. The method of claim 1 , wherein the offset compensation matrix includes yaw, pitch and roll matrices.
10. A vehicle system comprising:
a processor configured to:
obtain multiple 3D points for an object;
transform the multiple 3D points to a camera frame;
obtain actual 2D points for the object;
establish a virtual camera corresponding to the actual 2D points;
determine an offset compensation matrix between a camera associated with the camera frame and the virtual camera; and
apply the offset compensation matrix to data points prior to use by vehicle control systems; and
a controller configured to control operation of the AV with offset compensated data points.
11. The vehicle system of claim 10 , wherein the processor further configured to:
execute a camera pose algorithm using the multiple 3D points in the camera frame and the actual multiple 2D points in a virtual camera frame associated with the virtual camera to determine the offset compensation matrix.
12. The vehicle system of claim 10 , wherein the offset compensation matrix includes translational matrices and rotation matrices.
13. The vehicle system of claim 10 , wherein the multiple 3D points are in a world coordinate frame.
14. The vehicle system of claim 13 , wherein the multiple 3D points in the world coordinate frame are transformed to an inertial measurement unit (IMU) frame, from the IMU frame to a Light Detection and Ranging (LiDAR) frame, and from the LiDAR frame to the camera frame.
15. The vehicle system of claim 10 , wherein the actual 2D points are determined using at least one of image detection, object detection, and corner detection.
16. The vehicle system of claim 10 , wherein physical properties and intrinsic parameters of the virtual camera and the camera are the same.
17. The vehicle system of claim 10 , wherein the offset compensation matrix includes yaw, pitch and roll matrices.
18. A method for compensating for autonomous vehicle (AV) system errors, the method comprising:
obtaining projected 2D points from multiple 3D points for an object;
obtaining actual 2D points for the object;
calculating an error between the projected 2D points and the actual 2D points;
executing a compensation algorithm when a calculated error is equal to or greater than a defined threshold, wherein the compensation algorithm comprises:
transforming the multiple 3D points to a camera frame;
establishing a virtual camera corresponding to the actual 2D points;
determining an offset compensation matrix between a camera associated with the camera frame and the virtual camera; and
applying the offset compensation matrix to data points prior to use by vehicle control systems; and
controlling operation of the AV with offset compensated data points.
19. The method of claim 18 , wherein the determining the offset compensation matrix further comprising:
executing a camera pose algorithm using the multiple 3D points in the camera frame and the actual multiple 2D points in a virtual camera frame associated with the virtual camera to determine the offset compensation matrix.
20. The method of claim 19 , wherein the offset compensation matrix includes translational matrices and rotation matrices.