Multi-Extended Target Tracking for Roadside Radar Based on a Short-Term GRU Association Network

To address the tracking problem of extended targets in traffic scenarios, where roadside millimeter-wave radar measurements are characterized by multiple scattering points, dense target interactions, clutter, and missed detections, this paper proposes a multi-extended-target tracking algorithm for roadside radar based on a short-term Gated Recurrent Unit (GRU) data association network. Measurements and track states are first mapped into a common state space, and residual sequences between each current measurement and the short-term historical states of candidate tracks are constructed. The GRU then extracts short-term dynamic-consistency features. A dummy-target class and an effective masking mechanism are introduced to identify clutter and support multi-measurement assignment and continuous tracking with time-varying numbers of measurements and tracks. Simulation results show that the proposed algorithm achieves lower OSPA distances, higher F1 scores, and shorter association times under different detection probabilities and a comprehensive complex traffic scenario, demonstrating good association accuracy, robustness, and real-time performance.