DOI: https://doi.org/10.36719/2663-4619/130/223-245
Multi-UAV Search Strategies for Anomaly Monitoring
Ilham Mirzayev1* , Ismayil Ismayilov2 , Emil Sadigov1 , Farid Guliyev1
Abstract. Teams of unmanned aerial vehicles with onboard vision are used to locate events of interest inside a bounded area, such as disaster victims, wildfire ignition points, and industrial leaks. The choice of search strategy is usually made without a formal basis. This study asks when a detection-driven adaptive strategy outperforms systematic geometric coverage. The spatial clustering of events is treated as a controlled variable generated by a Thomas cluster process. Lawnmower, spiral, and Hilbert-curve sweeps are compared against an adaptive strategy that adds bounded detection-triggered detours to a Hilbert sweep. In a 400 m arena the adaptive strategy found 23 percent more events than the best sweep under tight clustering. The advantage crossed to a deficit near a cluster scale of about 47 m. A frame-level analysis shows that this gain is not simply an artifact of the camera footprint. Co-detectability is nonetheless collinear with the clustering variable, so a second experiment separates the two effects through a minimum event separation of 90 m that exceeds the footprint diagonal. The advantage then moves from event count to detection timing. Clustered events were found 34 s earlier on average, and at the tightest clustering level half of the events were reached after 275 s rather than 483 s. Under uniform placement the detours became a bounded cost and the sweeps performed better.
Keywords: multi-UAV search, coverage path planning, adaptive search, spatial clustering, event detection, co-detection