A Multi-Object Tracking Method Based on Multi-Drone Collaborative Perspectives

For the multi-object tracking problem from multi-drone perspectives, we proposes a method based on measuring trajectory feature similarity. It leverages appearance, position, and projection features from local trajectories observed by individual drone to calculate global trajectory similarity comprehensively. By optimizing the trajectory-target association matrix, we yields interpretable optimal matching results across perspectives. Synergistically utilizing complementary information from multi-drone perspectives, the approach computes global multi-object tracking results. Experimental results on MDMT dataset demonstrate our method’s excellent performance across multiple evaluation metrics, effectively addressing common issues of target occlusion and reappearance in complex scenes.