THOR: Holistic Roadside Traffic Monitoring With a Single mmWave Radar

Roadside traffic monitoring offers a scalable way for pervasive sensing by leveraging existing street infrastructure. To support next-generation urban services, roadside units require holistic traffic perception, simultaneously capturing vehicle volume, speed, and 3D geometry. However, existing solutions rely on multi-sensor suites suffering from high maintenance overhead or environmental fragility. While mmWave radar’s versatile and robust sensing capability has potential for holistic and all-weather traffic sensing, it faces fundamental conflicts across all sensing tasks, including vehicle visibility vs. temporal resolution, speed ambiguity vs. range resolution, and imaging fidelity vs. processing latency. Critically, these conflicts are mutually coupled; optimizing for one task degrades the others. Thus, a holistic system cannot be built by simply assembling isolated single-function methods. To address these conflicts, we proposeTHOR, a mutually-coupled pipeline that transforms a single commercial mmWave radar into a multi-function traffic sensor. First, we overcome the conflict between visibility and temporal resolution through a within-frame multi-scale sensing approach to enhance the vehicle visibility under occlusion without sacrificing temporal resolution. Second, we bypass the trade-off between speed ambiguity and range resolution by inverting speed estimation problem into a spatial alignment task. Third, we resolve the fidelity and latency conflict in 3D imaging by introducing an efficient rectified flow imaging model with dedicated vehicle geometry refinement. Extensive field evaluations across diverse traffic conditions show thatTHORachieves 97.8% volume counting accuracy, 3.4 km/h speed error, and outperforms state-of-the-art imaging baselines by up to 9× with an average real-time latency of 0.64 s for holistic monitoring.