Target Locating of Robots Based on the Fusion of Binocular Vision and Laser Scanning

Abstract

The accuracy of target locating for vision-based robots would be affect by environment factors such as light. In order to solve the problem, we proposed a target locating method based on binocular vision locating technology, together with laser scanning positioning technology, considering that laser scanning can obtain three-dimensional environmental information with high definition, and is little influenced by light. In this method, the binocular vision pixel image and the laser scanning point cloud image are first jointly calibrated to map the image pixels with the point cloud data; secondly, the visual image and the laser two-dimensional depth map are detected separately using YOLOv3; then, a decision-level fusion method is utilized to fuse point cloud depth image and the camera image; finally, YOLOv3 is used to detect bound box and confidence of the fused map. The results show that the proposed method has the ability to locate object accurately.