Spatial Alignment and Matching Enhancement for Multi-Drone Multi-object Tracking

The goal of Multi-Drone Multi-object Tracking (MDMT) is to detect and track multiple objects from videos captured simultaneously by multiple drones with overlapping fields of view. It helps overcome the challenges of limited view and occlusion in single-drone tracking. However, existing methods often fail to account for the motion characteristics specific to drone platforms, and their cross-view association strategies typically rely solely on positional distance or appearance features. As a result, their performance deteriorates under scenarios involving rapid camera motion or small object tracking. To this end, we design a Drone Motion Compensation (DMC) module and a Position-IoU Fusion (PIF) module, which constitute a Spatial Alignment and Matching Enhancement method for Multi-Drone Multi-object Tracking (SAME-MDMT). We utilize drone motion compensation for tracking correction to enhance spatial feature alignment, and perform object association based on the fusion of position features and IoU to improve the accuracy of cross-view object matching. We conduct experiments on the MDMT dataset. The results validate the effectiveness of our method in enhancing identity consistency and cross-view association capability in MDMT tasks.