A General Framework for Multi-Layer Hypothesis Tracking under Appearance Ambiguity

Reliable multi-target tracking is a critical capability in modern sensing and surveillance systems, supporting applications ranging from electronic monitoring to autonomous platforms. A major challenge arises when targets exhibit highly similar or ambiguous appearances, which can lead to severe ambiguities for appearance-based association strategies. To address this challenge, a multi-layer hypothesis tracking (MLHT) framework is proposed that leverages topological consistency and trajectory hypothesis management. The framework includes a graph matching layer that models spatial relationships to generate robust association hypotheses and a discrete trajectory construction layer that recovers trajectory hypotheses for targets with irregular motion patterns or long-term occlusions. A subsequent fusion and reduction stage resolves conflicts among competing hypotheses and retains a globally consistent set of high-confidence trajectories. Experiments on multiple benchmarks with appearance ambiguity demonstrate that MLHT improves tracking performance and robustness compared with representative baselines, suggesting that the proposed framework can serve as a general framework for multi-target tracking in complex sensing and surveillance scenarios.