Fig. 2: Impact of blind sampling strategies on target visibility in a simulated procedural forest environment.
From: Drone swarm strategy for the detection and tracking of occluded targets in complex environments
a Our simulation environment was a 1 ha procedural forest with one hidden avatar. b Blind brute force sequential sampling, as in the case of a single-camera drone that sequentially samples the SA (Fig. 1b)61, d led to a maximum target visibility (MTV) of 51% after a long period of 75 s. c Blind parallel sampling, as with an airborne camera array (Fig. 1c)63, was fast, but d resulted in only 19% MTV after a short period of 3 s. Our objective function O models target visibility by the contour size of the largest connected pixel cluster (blob) computed from anomalies in color (RGB) and thermal channels, as explained in Objective Function. Note that the yellow boxes highlight the target, the white arrows show the movement of the drones between time steps t − 1 and t, the blue lines represent the total sampling paths, and the gray area shows the integrated ground coverage at time t. Simulation parameters: drone’s ground speed = 10 m/s, forest density = 300 trees/ha, b single-camera drone sampling sequentially a 36 × 38 m SA with a 4 × 2 m resolution, c array of 10 cameras sampling at 1 m inter-camera distance with 2 m steps in the flight direction (as shown for the prototype in Fig. 1c). See Supplementary Movie 1.
