Beyond the FoV: Identity-Preserving Cross-Region Multi-Target Tracking with Collaborative mmWave Radar

Recent advancements in millimeter-wave (mmWave) radar have transitioned from static sensing to dynamic multi-target tracking. However, maintaining continuous target trajectories across disjoint sensing regions remains a significant challenge. Most existing solutions are confined to a single radar's Field of View (FoV) and lack mechanisms to preserve target identities (IDs) during cross-region transitions. In this paper, we present MCTrack , a multi-target cross-region tracking framework designed for identity preservation across disparate radar's FoVs. MCTrack utilizes a distributed architecture where each sensing unit, comprising an mmWave radar and an edge device, performs synchronized data acquisition and bidirectional inter-unit communication. To achieve robust cross-region re-identification (Re-ID), we propose a hybrid-domain feature descriptor that fuses the coarse-grained point cloud domain with the fine-grained Doppler domain, capturing unique behavioral signatures for each individual. Furthermore, we develop a neural network-based Re-ID matching mechanism specifically optimized to handle irregular or incomplete feature patterns inherent in real-time deployments. We implement a full-scale prototype of MCTrack and evaluate its performance across diverse experimental scenarios. Our results demonstrate high localization precision with an average tracking error of 0.17m for up to 10 targets. In identity-matching tasks involving 20 targets, MCTrack achieves an average Top-1 Re-ID accuracy of 83.7% in three representative scenarios, demonstrating its robustness and scalability for large-scale, ubiquitous sensing applications.