A Fully Automatic Calibration Algorithm for a Camera Odometry System

This paper focuses on the calibration of a mobile robot system with odometry and a monocular camera. Most current approaches are based on either batch optimization or Bayesian filter, which require a good initial guess to obtain the optimal calibration results. In this paper, a two-step calibration algorithm is proposed for a fully automatic calibration of the camera-odometry system, consisting of: 1) a non-iterative auto initialization step for spatial extrinsic calibration based on the landmark measurements and 2) a joint optimization step to estimate both the spatio-temporal extrinsic parameters and the odometric parameters. By exploiting the planar constraints of landmark measurements, our auto initialization method can achieve better accuracy compared with the current solutions. Experiments are conducted based on both the simulation data sets and also the real world data sets collected from an autonomous guided vehicle (AGV) system, which validate our algorithm compared with a VO-based non-iterative method.