| CPC G06T 7/80 (2017.01) [G06T 7/70 (2017.01); G06T 7/73 (2017.01); G06T 2207/20221 (2013.01); G06T 2207/30204 (2013.01); G06T 2207/30261 (2013.01)] | 20 Claims |
|
1. A computer-implemented method, comprising:
capturing a first frame and a second frame for an environment of an autonomous driving vehicle (ADV) from at least a first and a second cameras mounted on the ADV, the first and second cameras being two calibrated cameras with designated positioning angles;
determining at least two points in the first frame having corresponding points in the second frame, wherein the at least two points in the first frame have corresponding points in the second frame and are used for a calibration refinement process;
determining a geometric relationship between the at least two points in the first frame and their corresponding points in the second frame based on epipolar constraints, the relationship comprising a rotational discrepancy for actual positioning angles of the first and second cameras from the designated positioning angles;
determining distance and angle measurement information from the first camera to the at least two points and from the second camera to the corresponding points by applying the relationship to the at least two points in the first frame and the corresponding points in the second frame; and
refining a calibration of the first and second cameras by determining the actual positioning angles of the first and second cameras with respect to an orientation of the ADV based on the distance and angle measurement information and pixel information of the at least two points in the first frame and the corresponding points in the second frame, wherein the actual positioning angles are used to compensate misalignments in positioning angles for the first and second cameras.
|