Steel ladle visual alignment method and device based on deep learning semantic segmentation and point cloud registration, and equipment

The invention discloses a steel ladle visual alignment method and device based on deep learning semantic segmentation and point cloud registration, and equipment. Based on a depth picture acquired by an RGB-D camera, RGB steel ladle picture features are extracted by using a coding structure in a SegNet network, a steel ladle contour is restored by decoding the structure, pixel-level accurate segmentation is realized, and finally, point cloud information is extracted from a steel ladle contour segmentation map. For solving the problems that a traditional point cloud initial registration algorithm is low in success rate, unstable in registration, much in point cloud noise, partially overlapped and the like, a semi-supervised rapid point cloud registration method is adopted to carry out registration on point clouds of the steel ladle contour segmentation map, so that the posture of the steel ladle is obtained, and accurate alignment of a trolley and the steel ladle in the casting link is achieved. 本发明公开了一种基于深度学习