Pattern of Life Calculation from Big Graphs

Pattern of Life characterization involves studying how people interact with the world around them on a daily basis: places they visit; people they interact with; actions they perform. Effective tracking of such patterns of life has direct implications to defense intelligence, cyber intelligence and corporate security. To solve this problem, ObjectVideo in collaboration with Dr. Leman Akoglu (Stony Brook), proposes to build a system that mines multiple data sources for relational information between people and places, detects anomalies and provides the tools to track the behaviors of anomalous entities. We will study patterns of life on four data domains: surveillance video, satellite imagery, text streams and non-traditional data sources such as weather feeds. We will use image / video analytics libraries to identify events from surveillance videos and detect change in satellite imagery. We will invoke Natural Language Processing tools to perform named entity recognition, entity resolution and relation estimation. Additionally, we will fit time-series models to weather feeds to study anomaly. We propose a unified graph representation to combine aforementioned data. Finally, we will invoke anomaly detection algorithms that scale up with large dynamically changing graphs and will build a visualization tool that helps analyst track anomalies.