High-Precision Indoor Robot Dynamic Obstacle Detection with Laser and Camera

Abstract

Abstract Indoor robot is a very common robot, and its main navigation method is laser SLAM. In the process of detecting dynamic obstacles with laser SLAM, the clustering effect is not obvious, the accuracy is low, and the real-time performance is poor. To solve this problem, a sensor fusion method of lidar and monocular camera is proposed, which improves the accuracy of detecting the position of moving obstacles. When processing lidar data, first preprocess the data, then use the BDSCAN algorithm to cluster the point cloud, and finally use the Kalman filter to predict and calculate the location of the moving obstacle. When processing the data of the monocular camera, the YOLOv3 algorithm is used to quickly detect the target and return the position of the monocular camera. After obtaining the camera and laser data, according to the installation position of the radar and camera, map the position calculated by the radar to the coordinates of the camera. If the coordinate difference after mapping is small, it can be considered that the two sensors have detected the same target. The result is returned to the display interface. Using this method, the efficiency and accuracy of indoor robots in detecting moving obstacles have been greatly improved.