Point Cloud Fusion of Human Respiratory Motion Under Multi-View Time-of-Flight Camera System: Voxelization Method Using 2D Voxel Block Index

Abstract

Time-of-flight (ToF) 3D cameras can obtain a real-time point cloud of human respiratory motion in medical robot scenes. Through this point cloud, real-time displacement information can be provided for the medical robot to avoid the robot injuring the human body during the operation due to the positioning deviation. However, multi-camera deployments face a conflict between spatial coverage and measurement accuracy due to the limitations of different types of ToF modulation. To address this, we design a multi-camera acquisition system incorporating different modulation schemes and propose a multi-view voxelized point cloud fusion algorithm utilizing a two-dimensional voxel block index table. Our algorithm first constructs a voxelized scene from multi-view depth maps. Then, the two-dimensional voxel block index table estimates and reconstructs overlapping regions across views. Experimental results demonstrate that fusing multi-view point clouds from low-precision 3D cameras achieves accuracy comparable to high-precision systems while maintaining the extensive spatial coverage of multi-view configurations.