Space Partitioning and Classification for Multi-target Search and Tracking by Heterogeneous Unmanned Aerial System Teams
Jared Wood, J. Karl Hedrick · Infotech@Aerospace 2011 · 2011
Search and tracking of an unknown number of targets within large search areas is a task performed in settings such as search and rescue, surveillance, area patrol, and reconnaissance. This task can be accomplished by a team of cooperative, heterogeneous, autonomous agents. In order to search for and track targets by a team of agents, target sensing and position estimation are required, as well as algorithms for agent path planning. The estimation of targets is accomplished by fusing detection/no-detection observations. The fusion of these observations forms a distribution of estimated target density over the target search space. The nature of an unknown number of multiple targets and small sensor fields of view results in several maxima in the target density distribution corresponding to potential locations of targets. The goal of the team of agents is to reach system steady-state consisting of the entire target space searched and all targets discovered and being tracked by agents. This corresponds to the target density distribution reaching its lowest possible uncertainty. Assuming mobile targets, once a target has been detected, frequent repeated detections are required to maintain low local uncertainty of the target location. Consequently, as targets are detected, the region of the detections must be tracked. As tracking regions appear, some agents transition from searching to tracking. Extending this to the entire target search area, based on the target density distribution the area is dynamically partitioned to account for regions of potential targets with varying levels of uncertainty. Features are extracted from these partitioned regions to classify the type of search or track that should be performed for each region. The partitioned regions are viewed as tasks and are assigned to the team of agents. The main focus of this paper is to present an approach for dynamically partitioning the target search area, based on the target density distribution, into evolving regions of potential targets of varying local uncertainty. Classification of the partitioned regions into varying levels of search or tracking tasks is also presented. Additionally, estimation of the target density distribution is discussed. Results are provided of a search task consisting of six Unmanned Aerial System (UAS) agents equipped with visual spectrum camera sensors searching for an unknown number of targets.