Detection and Tracking of UAV by Spatial, Temporal and Distance Contour Hybrid Architectures with Fuzzy Information Fusion

In this article, we propose a general triple channel hybrid spatial (images-based), temporal (motion-based) based and depth contour (distance-based) architecture from RGB or RGB-D cameras for UAV detection with fuzzy fusion for accurate positions (bound boxes). We also propose DCF-based tracking algorithm. The goal is for detection and tracking of UAV from a camera mounted on another moving UAV platform for multirange (near, medium and far), for large to small UAV target sizes. Under the cluttered environment, the detection of UAV by image is challenging due to unclear images and changing background, as such, the motion based detection is introduced to compensate by detecting the motion difference from the surroundings. Our triple channels hybrid architecture is based on the the coupled YOLOv8 detection and BotSort tracking for near and immediate target, and another one based on U2U-D&T architecture for small and far target and recognition of geometry based on depth contour and blob detection. The information is finally fused by hybrid meta-heuristics and multi-criteria decision making with linear weighted aggregation operators. The combined overall algorithm is validated using video data collected from multiple fixed-wing UAVs that is manually labelled and is publicly available. and specific video data for clustered environment. Preliminary results indicate that the proposed combined algorithm can achieve highly accurate detection and tracking of UAV even in disturbed or fainted UAVs in both clear background, e.g. in clear sky and over ground, and even in disturbed or fainted UAVs in front of cluttered background.