Adaptive multi-strategy for multi-vehicle with mutual occlusion tracking

Multi-vehicle with mutual occlusion tracking is a great challenging task. In this paper, we present a novel scheme based on adaptive multi-strategy to cope with this problem. First, a rule that determines the relation and degree of occlusion among different vehicles is proposed. The rule will output three cases: the first is severely occluded vehicle, the second is partially occluded vehicle and the third is non-occluded vehicle. And then, for the three cases, different tracking approaches are adopted: for the first case, a trace prediction based method using motion features is employed. For the latter two cases, dictionary and l₂ regularized collaborative representation based method is employed. During tracking, only background dictionary is updated. The experimental results have demonstrated the promising results for multi-vehicle with mutual occlusion tracking.