Airborne Visual Tracking for Cooperative UAV Swarms

This paper presents processing architectures and algorithms for airborne visual tracking in cooperative scenarios, such as the ones relevant to vision-aided swarming of Unmanned Aerial Vehicles. While in the considered applications it is possible to exploit a priori information relevant to the physical configuration of the target UAVs, and coarse knowledge of the relative position among aircraft, significant requirements exist in terms of the tradeoff between false positives and missed detections. The focus is set on medium/large distances among UAVs, and the presented solution combines adaptive template matching and morphological filtering to achieve reliable performance for variable background and illumination conditions. In order to minimize requirements related to computational load and vision system performance (e.g., minimum frame rate), navigation and cooperative information is used within the tracking algorithm. The adopted approach is tested on experimental data acquired in multi-UAV flight tests. First results demonstrate the potential of fusing information derived from cameras, onboard sensors, and target navigation system