A new algorithm for the extrinsic calibration of a 2D LIDAR and a camera

摘要

In this paper, we propose a new extrinsic calibration algorithm for a camera and a 2D Light Detection And Ranging sensor (LIDAR). N plane-line correspondences are used to establish geometric constraints for computing the extrinsic parameters. The existing algorithms that use the same constraints either require N ≥ 5 or focus on the minimal solution for N = 3. In contrast, the proposed algorithm is effective for all N ≥ 3. This is made possible by decoupling the initial estimation of rotation from translation and imposing the orthonormal constraints on the computation of rotation. When the rotation matrix is obtained, the estimation of translation is reduced to solving a linear least-squares problem. Then an iterative procedure is followed to refine the initial estimation. This algorithm gives a new minimal solution, which can be used as the hypothesis generator in the RANdom SAmple Consensus (RANSAC) algorithm. Simulation and real experimental results show that the proposed algorithm outperforms the state-of-the-art ones.