In this paper, we present a new algorithm to directly calculate the projection matrix between a camera and a LIDAR. Since Line correspondences are used to construct the geometric constraint in our algorithm, the calibration target is flexible and any object with straight line boundary can be an appropriate calibration target. The algorithm has two stages. Firstly, the linear solution is estimated by minimizing the algebraic error. Then it is refined by optimizing the geometric error. Two different kinds of geometric distance are discussed and the one using the distance from a laser point to the plane, which is formed by back-projecting from an image line, is chosen. Because the quality of the measurements impacts on the accuracy of the final result, a sampling strategy is also given to obtain high quality measurements and avoid overfitting. Simulation and experimental results validate the proposed algorithm and sampling strategy.
展开▼
