SLAM Method of Mine Inspection Robot based on Stereo Vision and IMU

In view of the poor robustness And poor accuracy of single sensor SLAM technology in the complex surroundings of coal mine, A global localization method based on multi-sensor fusion of IMU and stereo vision is proposed. Firstly, the ORB feature were extracted and matched, and the FAST keypoints were extracted by adaptive threshold to improve the quantity and quality of feature points. The quadtree method was used to realize the uniformity of point feature extraction and overcome the problem of feature stacking redundancy. Then, the direction consistency check and RANSAC algorithm are used to remove the mismatching of feature points. Finally, the tight coupling optimization mechanism was used to fuse the sensor data, and the loop closure detection module is configured to increase the overall positioning accuracy. According to the experimental results in the EuRoc public data set, compared with ORBSLAM2, a single stereo vision sensor, the absolute positioning accuracy increased by 36.49% on average, and compared with VinsFusion based on visual-inertial fusion, the absolute positioning accuracy is improved by 87.21% on average. It effectively improves the robustness and positioning accuracy of the coal mine inspection robot.