Automatic Extrinsic Calibration of Monocular Camera and LIDAR in Natural Scenes

Abstract

It has practical significance to automatically calibrate the multiple sensors in autonomous driving vehicles. This paper proposes an automatic calibration method to address the extrinsic calibration for a perception system comprised a LIDAR and a monocular camera, which can be used in natural scenes without calibration boards. The calibration problem is formulated as minimizing the structure feature errors via nonlinear least squares optimization. In order to solve the nonlinear optimization problem, the continuous frames information are used to initialize parameters and extract features. Simultaneously, Levenberg-Marquardt algorithm and a global optimization algorithm are combined to obtain the best optimal parameters. Extensive experiments tested on KITTI benchmark demonstrate that our method reaches a valid and competitive result.