Combining Visual Features and Correlation Filters for UAV-Based Multi-target Association in Consecutive Frames

In multi-target perception tasks involving UAVs, linking multiple targets across successive frames is crucial, as precise target association enables the estimation of the number of targets present in the current scene. To address this issue, this paper proposes a continuous frame multi-target association algorithm that integrates visual features with correlation filter. DETR is first used to detect targets and extract their visual features. Correlation filters then predict the target’s position in the next frame. Finally, a weighted calculation combining IoU and feature similarity is performed to obtain the target association score, determining the final association result. The algorithm’s effectiveness was validated in the AirSim simulation environment.