Point Cloud-Based 3-D Tracking for Asynchronous and Uncalibrated Multicamera Systems

Abstract

Accurate 3-D tracking in heterogeneous, unsynchronized multicamera systems remains challenging because of calibration overhead and temporal drift. In this study, we present a point cloud- based framework that reconstructs the target trajectories without prior calibration or hardware synchronization. A sparse environmental point cloud provides a stable spatial reference; camera poses are estimated using perspective-n-point and refined with bundle adjustment. Moving objects are detected through k-nearest neighbor foreground extraction, and 2-D tracks are compressed into 1-D motion signals. Variational mode decomposition suppresses noise, whereas a two step alignment—subsequence dynamic time warping followed by sliding window fine matching—synchronizes asynchronous video streams. Robust triangulation recovers 3-D path. This method offers a low cost and easily deployable solution for wide area multitarget monitoring.