A multiple UAV system for vision-based search and localization, Report no. LiTH-ISY-R-2865

The contribution of this paper is an experimentally veri ed real-time algorithm for combined probabilistic search and track using multiple unmanned aerial vehicles (UAVs). Distributed data fusion provides a framework for multiple sensors to search for a target and accurately estimate its position. Vision based sensing is employed, using xed downward-looking cameras. These sensors are modeled to include vehicle state uncertainty and produce an estimate update regardless of whether the target is detected in the frame or not. This allows for a single framework for searching or tracking, and requires non-linear representations of the target position probability density function (PDF) and the sensor model. While a grid-based system for Bayesian estimation was used for the ight demonstrations, the use of a particle lter solution has also been examined. Multi-aircraft ight experiments demonstrate vision-based localization of a stationary target with estimated error covariance on the order of meters. This capability for real-time distributed estimation will be a necessary component for future research in information-theoretic control.