RGB-D Point Cloud Correction using Information of Adjacent Frames

Breathing motion compensation is an important issue to be addressed in robotic thoracic and abdominal surgery. In recent years, with the widespread application of RGB-D depth cameras based on TOF imaging principles, respiratory monitoring has become a trend. However, due to the multi-path interference of TOF principle RGB-D cameras, various measures are often taken to avoid interference when using multiple cameras for respiratory monitoring. In this paper, a deep learning network that fuses previous frame information is proposed for respiratory motion point clouds under multi-path interference. At the same time, a method for generating multi-path interference regions from clean point clouds is designed, and a dataset is created for training. Our experiments prove that point cloud misalignment caused by multi-path interference can be effectively corrected using a neural network that fuses information from previous frame point clouds.