Real-Time Multi-Target Tracking via Signal Variation Clustering Without Prior Target Count

Abstract

Accurate and real-time tracking of multiple moving targets remains a fundamental challenge in radar-based indoor sensing, particularly when the number of targets is unknown or varies over time. This paper presents a real-time motion tracking system RTCtrack that operates directly on signal variations induced by human movement. At each time step, the method extracts significant variation points relative to a static baseline and incrementally groups them into coherent trajectories without requiring prior knowledge of the number of targets. A tail-based spatial association strategy enables robust path formation, while weak or inconsistent clusters are automatically removed based on activity level. To maintain tracking stability in evolving environments, an adaptive baseline update mechanism replaces the reference signal when persistent global deviations are detected. RTCtrack is evaluated using a public indoor radar dataset involving multiple individuals moving independently. Simulation results show that the system reliably identifies and tracks multiple motion paths in real time without knowing the exact number of targets. Comparisons with video-based ground truth confirm strong spatial and temporal alignment, demonstrating its effectiveness for radar sensing and passive human motion tracking.