Human Activity Recognition

ObjectVideo currently markets a commercial automated video surveillance product that detects and tracks targets in video streams. The product allows a user to specify simple events of interest via an intuitive graphical user interface. ObjectVideo proposesbasic research with the goal of enhancing product capabilities in the following areas: (1) detecting individual humans and recognizing their poses and gestures; (2) using machine learning techniques to automatically detect abnormal or threatening humanactivities; (3) extending event descriptions to encompass wide ranging activities that can be observed over long periods of time or over a wide spatial extent (via multiple cameras); and (4) integrating the input from multiple sensor modalities (thermal,microwave, acoustic, etc.) to increase the robustness and variety of detectable human events. Particular emphasis will be placed on human interface components that allow easy, intuitive system tasking and feedback to learning algorithms. State-of-the-artsecurity systems consist of a large array of sensors providing a continuous stream of data for human consumption. This raises a major problem - there is too much data for any human to analyze reliably. The data is ignored. Automated approaches torecognition of activities and events are critically needed if this data is going to be used for proactive, real-time threat warnings, and not after-the-fact evidence gathering. This proposal will augment the current state-of-the-art, which includes basicblob tracking and type classification from a single camera, with the more advanced abilities, including human activity recognition from multiple sensors.There are many commercial applications of this technology. Any security installation with remote sensors requires better analysis of that security data. We see two attractive market segments, spanning homeland defense, force protection, and commercialphysical security: (1) truly sensitive installations that will always require the best warning systems, and (2) mid-size installations that often want to reduce personnel costs. ObjectVideo's current product for automated video surveillance has found awarm reception among the country's most security conscious managers, and we believe the new technologies proposed here will be even more warmly received.