Visual Odometry of a Low-Profile Pallet Robot Based on Ortho-Rectified Ground Plane Image from Fisheye Camera

This study presents a visual-only odometry technique of low-profile pallet robot using image feature tracking in ground plane images generated from fisheye camera. The camera is commonly used many vision applications because it provides larger field view (FoV) around robot. However, the large radial distortion, generally converted to pinhole for visual or matching. Although distortion can be eliminated via undistortion with lens calibration parameters, causes several side effects, such as degraded resolution and significant reduction FoV. In this paper, instead model, we propose generate (GPI) image. GPI virtual top-view that only contains at front First, original projected cubemap. Second, bottom faces cubemap are GPI. Third, homographically transformed again further reduce distortion. As result, an accurate ortho-rectified obtained For GPI, number 2D motion vectors extraction between previous current frames By calculating scaled vector, which measurement wheel encoder mobile robot, estimate velocity steering angle vector. Finally, pose by applying kinematic model