A direct method for the extrinsic calibration of a camera and a line scan LIDAR

This paper presents a direct method for the extrinsic calibration of a perspective projection camera and a line-scan LIDAR. We freely move a checkerboard pattern to obtain different plane poses in the camera coordinate system and depth readings in the line-scan LIDAR reference frame. From 3 plane-line correspondences, we first present a closed form solution for the relative orientation between two sensors, then the translation can be determined linearly. Unlike the method proposed in [12] to reformulate the registration of 3 plane-line correspondences as a standard P3P problem, we find novel and direct constraints on rotation. From these constraints, a closed form solution for rotation is found by multivariate polynomial elimination using a resultant from the Jacobian method of Salmon. Some degenerate cases in the P3P based method [12] inherited from the standard P3P problem do not occur in our method, which makes our method a little more stable and accurate in some cases. Experimental results demonstrate the effectiveness of the proposed approach.