A Robot Relocalization Method Based on Laser and Visual Features

Abstract

Relocalization is a well-known problem to regain the robot's pose in an incorrect pose. The robot using Adaptive Monte Carlo Localization(AMCL) algorithm can achieve localization based on laser information, but AMCL is prone to localization drift and challenging to achieve relocalization, especially in environments with simple and repetitive geometric features. Image can obtain more texture information and color information than laser, which determines that the camera is more accessible to relocalization than the lidar. This paper proposes a fast and reliable relocalization method to solve the problem of laser localization drift and kidnapped robot by visual information. The robot can build and save the point cloud map and the grid map and calculate the correspondence. Relocalization is achieved through a relocalization trigger mechanism and Oriented FAST and Rotated BRIEF(ORB) feature points. Finally, Several experiments in the simulation and indoor environment were performed to verify the effectiveness of the proposed approach. The experimental results show that the relocalization recovery time of the proposed method is within two seconds. Compared with the relocalization achieved by lidar alone, the proposed method is three times faster.