Selective and Hierarchical Allocation of Sensing Resources for Anomalous Target Identification in Exploratory Missions

We present an approach for selective, hierarchical allocation of sensing resources that aims to maximize information gain in exploratory missions such as search and rescue (SAR) or surveillance in an efficient manner. Specifically, we propose a methodology for perception-enabled SAR or crowd surveillance driven by anomaly detection based on low-level statistical assessment of a region. The characterizations of previously-observed regions are used to populate a window of observations that serves as 'short-term memory,' providing a contextually-appropriate characterization of proximate regions in the scene. Currently-observed regions are compared with this short-term memory window, and if sufficiently dissimilar, can be considered as candidates for the presence of a SAR target or unexpected event. We adaptively allocate additional sensing resources for subsequent exploration of anomalous regions through a novel utility function that balances varied mission objectives and constraints including exploratory sensing actions, maintaining situational awareness, or ensuring some degree of confidence in self-localization. Simulation results validate the proposed approach and demonstrate its benefits with regards to efficiency in exploration while maximizing potential information gain and balancing other mission requirements and objectives.