An overview of Homogeneous Multi-UAV Air Tracking Framework (HOMATracker), a system designed for efficient tracking of homogeneous swarm UAVs in air-to-air scenario. (IMAGE)
Caption
Our proposed HOMATracker comprises the following steps: (A) Object detection: A detector is used to locate the objects in each frame; (B) Instance-level appearance feature extraction: We involve a multi-frame UAV pose-attention-based appearance component (MPA-Net) that captures the pose features of objects across consecutive frames. A transformer-based pose attention module then calculates the similarity of pose-appearance features between the high-confidence objects and existing tracklets. (C) Spatial and motion feature similarity calculation: The multi-frame motion difference accumulation (M2DA) strategy calculates the spatial and motion feature similarity between high-confidence objects and tracklets using the Wasserstein Distance. (D) Multi-frame homogeneous object association: Detected objects are categorized into high-confidence and low-confidence groups. The multi-frame homogeneous object association (MHA) framework is used to link high-confidence detections with existing tracklets. Any high-confidence detections that remain unmatched are then initialized as new tracklets.
Credit
Chinese Journal of Aeronautics
Usage Restrictions
News organizations may use or redistribute this image, with proper attribution, as part of news coverage of this paper only.
License
Original content