closed-loop-inference

TRACER is a training-free framework for traffic accident reconstruction that formulates the problem as closed-loop structured inference, iteratively refining event-anchored motion hypotheses under geometric and kinematic constraints, achieving improved fidelity and consistency over data-driven and physics-based baselines.