Recognition and 3D Pose Estimation for Underwater Objects Using Deep Convolutional Neural Network and Point Cloud Registration

摘要

Acquiring high-precision spatial position and posture of the object is essential for operation of underwater vehicle. The traditional methods of pose estimation based on point cloud require the manual extraction of the object point cloud, which have poor autonomy and low efficiency. For this problem, a novel method based on Convolutional Neural Network(CNN) and point cloud registration is proposed. In this paper, we use the center and range recognized by CNN of object to guide clustering segmentation of the point cloud data, and then extract the point cloud of the object. Finally, we complete the precise positioning of the object through coarse point cloud registration and fine point cloud registration. We build underwater experimental environment, and the results show that our method can effectively estimate the position of irregular object with high accuracy.