Wednesday, December 14, 2016

Learning Models for Predictive Behavioral Intent and Activity Analysis in Wide Area Video Surveillance

Author: Shishir K. Shah

Providing automatic monitoring of surveillance video to help security officers detect and predict suspicious activities is critical so that they have enough time to take specific actions to prevent a crime or mitigate a security threat.

The ultimate goal of this research is to develop an intelligent, non-obtrusive, real-time, continuous monitoring system for assessing activity and predicting suspicious and criminal behavior across a network of distributed cameras.

To that end, the purpose of this project was to develop new recognition algorithms that can track and identify human activities for wide area video surveillance across large distributed camera networks. The developed prototype has two main modular components: human detection and tracking, and re-identification of body parts.

It is envisioned that the algorithms would support a non-obtrusive system that could continuously track all objects across a network of distributed cameras, analyze movements, and detect and measure descriptive information continuously; and serve as a decision system that can correlate each object’s patterns with others, generate models of suspicious/criminal activity, and generate activity alerts for security personnel who monitor and make critical decisions.

The findings indicate that the new tracking algorithms can better understand the motion of humans dependent on their interactions with both animate and inanimate objects. The research has successfully developed a model for collecting, testing, and tracking these algorithms for use in a distributed camera network.

Future work realized from the research is related to person re-identification and the understanding of human behavior that can be modeled to impact all aspects of wide area video surveillance.

No comments: