Planar KNN for Multi-camera Interference Mitigation of Point Cloud

Shengtao Li, Jiadun Wang, Daosong Hu, Kai Huang

Abstract: In recent years, Time-of-Flight (TOF) cameras have been extensively employed in fields such as human physiological monitoring. These cameras provide valuable data, but when multiple TOF cameras are used simultaneously, they often interfere with each other, leading to a significant decrease in data accuracy. Previous methods were primarily designed for interference removal in static data, with their effectiveness depending on the use of multiple frames. These methods exhibit limited performance when applied to dynamic datasets, making them unsuitable for interference removal in dynamic scenarios. To tackle these challenges, this paper presents a novel approach to mitigate multi-camera interference (MCI). We simulate the interference caused by multiple TOF cameras and construct a dynamic point cloud dataset of human breathing involving 9 individuals. We propose a planar KNN smoothing method based on point cloud normal information and planar distance. The method calculates the normals of the point cloud to identify interference regions, and uses these normals to determine the affected areas. Neighbors of the points within these regions are identified through 2D distance measurements, which are then used to correct the point cloud. Neighbors of the points within these regions are identified through 2D distance measurements, which are then used to correct the point cloud. Experimental results show that this method significantly improves the accuracy of point clouds affected by multiple cameras, and performance metrics along with ablation analysis further validate its effectiveness.

External IDs:dblp:conf/icic/LiWHH25