<p>Accurate data association methods are needed in Multi-Object Tracking (MOT) to reduce mismatches. However, existing association methods cannot effectively match ambiguous trajectories and detections in scenarios with severe occlusion. This because they fail to clearly describe the occlusion conditions of the targets and cannot finely differentiate the statuses of detections and trajectories, resulting in mutual interference between ambiguous trajectories and detections during the matching process. To overcome these concerns, this work presents a robust association method based on occlusion awareness (ROA). Specifically, occlusion relationships are determined based on the spatial relationships among targets, after which active trajectories and high-confidence detections are sequentially categorized into clear, occluded, and severely occluded states according to the degree of occlusion. This categorisation is intended to provide a more complete description of the occlusion state for both trajectories and detections. Subsequently, a new cascade matching method is introduced to breakdown the global matching problem into a series of local matches. By finding successive local optimal matches, it ultimately converges to a globally optimal solution. Finally, a multi-object tracker, ROA-SORT, is developed based on the proposed ROA method, enabling robust performance in occlusion scenarios. Our experiments demonstrate that the ROA-SORT algorithm has achieved scores of 76.91 for IDF1, 76.92 for MOTA, 62.81 for HOTA, and 62.5 for AssA on the challenging MOT20 dataset. In addition, applying ROA to four different trackers still enhances their tracking performance. This demonstrates ROA’s potential as a novel solution for addressing tracking failures in occluded scenarios.</p>