Mobile Robot Self-LocalizationUsing Visual Odometry Based on Ceiling Vision

Abstract

Self-localization is an important topic for mobile robot. Although some ceiling-based visual odometry have been proposed, the localization process is still cumbersome. In this paper, we propose a novel monocular visual odometry based on ceiling vision that is fast and easy to implement. By pointing the camera to the ceiling, visual odometry reduces the computation and eliminates interference from dynamic environments. We apply Oriented FAST and Rotated BRIEF (ORB) to match the features between two frames of the ceiling view and use sliding window filter to remove incorrect matches, and then get the orientation and position of the robot through coordinate transformation. Compared to traditional visual odometry, our algorithm performs better in speed, and only needs one correct match to get the displacement and rotation angle of robot.