Video Analytics for Criminal Justice Uses
Surveillance cameras give security personnel an overwhelming amount of video data to sift through. Video analytics technology
can automate and accelerate the review of this data to identify threats. Intelligent software can analyze movement in live
and recorded video to enhance officers' situational awareness and alert them to suspicious behaviors. At its best, this smart
technology can interpret the intent and predict the behavior of monitored subjects.
NIJ is supporting, through awards to GE, the development of technology for the automated detection of disorderly and criminal
activities. This technology not only assesses simple motion-based behavior, but estimates meaningful social relationships
between people and groups. This intelligent video system can:
- Operate in crowded environments such as prisons, public parks and schools where many people are interacting with each other
at the same time.
- Estimate crowd and group size, crowd density, and group speed and direction.
- Analyze patterns of following, chasing, fast movement, group formation and dispersion.
- Predict behaviors such as fighting and agitation.
- Automatically control pan-tilt-zoom cameras for face recognition.
- Automatically assess social group network structure, number of groups and leadership structure in small communities.
This system has been demonstrated at the NIJ-supported Mock Prison Riot. Further development and operational demonstration
in a law enforcement environment in Schenectady, New York is underway. In addition, the NIJ Sensors, Surveillance, and Biometric
Technologies Center of Excellence is evaluating the technology.
With NIJ funding, the University of Houston is developing a smart video surveillance system to predict suspicious behaviors
and analyze large-area activities in an effort to counteract aggravated assault, vandalism and theft in public spaces. The
system uses computational tools to detect and track human subjects, analyze their motion patterns across a distributed camera
network, and extract motion trajectories across non-overlapping camera views. This system is in early development.
Date Created: August 29, 2012