Aerial or spaced-based imaging assets cannot continuously monitor a single location or site of interest for prolonged periods of time such as weeks, months, or years without significantly sacrificing surveillance of other locations. Current approaches for modeling patterns of life (PoL) at a location are not capable of incorporating sporadic data and do not gracefully model daily to monthly or yearly PoL cycles. Unsupervised PoL models also result in a large number of false alarms when looking for specific anomaly types. The proposed approach addresses these challenges while also enabling transfer and adaptation of models to new, even dissimilar sites. In particular, our proposed effort builds upon our NGA BIG Topic 4 work to model varying temporal resolution cycles (e.g. weekly, monthly) using a hierarchical temporal compression model. This model incorporates sporadic evidence, fuses Multi-INT evidence (still imagery, video, and optionally real and simulated traffic data) to aid in filling in gaps, and is robust to track fragmentation. We will also incorporate higher order and contextual features and constraints to suppress false alarms and can further characterize the anomalies through hierarchical inferencing techniques. In Phase 1, simulated data using SUMO tool will be used for development and analysis.
