Online calibration for monocular vision and odometry fusion

Abstract

A novel calibration approach for mobile robot equipped with encoder and monocular camera has been proposed in this paper. In this work, we divide the calibration into four steps: 1) a coarse calibration step is employed to estimate the transform between odometry and visual system and the scale of visual system; 2) a nonlinear-optimization step is employed to optimize the transform between odometry and visual system; 3) during the process of SLAM of mobile robot, the transform between odometry and visual system is optimized with the poses of mobile robot and positions of mappoints in a local map; 4) the scale of visual system is optimized after a loop detection. This method uses only nature features in visual system without any artificial landmark or any prior knowledge. The experiment and compare have been performed to illustrate the effectiveness.