A Method of Performing Loop Closing Using Mask R-CNN Model in SLAM System

The loop closing in the visual SLAM system can eliminate accumulated errors and perform relocation, and has a positive effect on improving the accuracy of construction and system robustness. In the traditional method, only the geometric features of the image are analyzed, and there is no semantic information, so the anti-interference ability of the system is insufficient. Mask R-CNN is an object instance segmentation algorithm that gives accurate semantic labels and object masks. We propose a method of loop closing using the Mask R-CNN model. The feasibility of the proposed algorithm is verified by a series of experiments performed on the Turtlebot2 mobile robot.