Automated Detection and Prevention of Disorderly and Criminal Activities

• NCJ Number: 235579
• Author(s): Nils Krahnstoever
• Date Published: August 2011
• Length: 125 pages
• Grant Number(s): 2007-RG-CX-K015
• Agencies: NIJ-Sponsored
• Publication Type: Report (Study/Research)
• Award: Automated Detection and Prevention of Disorderly and Criminal Activities

Annotation

This report describes the development of a wide range of intelligent video capabilities relevant to law enforcement and corrections that can detect activities and alert operators allowing for early detection and prevention.

Abstract

The technology is intended to go beyond simple motion-based behavioral features to a semantically high level of understanding human activities by estimating meaningful social relationships between people and groups who are interacting in crowded environments. One of the main technical challenges was to detect events as well as motion and behavior patterns from tracks of people in crowded environments. The technology to be developed was required to recognize common group and crowd motion patterns, along with parameters such as crowd size, crowd velocity, agitation level, and events such as group formation and dispersion. The detection and recognition must occur on noisy data obtained by a video tracking system. A second technological challenge was to establish identity records of individuals based on their facial images in non-cooperative environments. Once identity records are established, associations (interactions) between subjects can be recorded; over time, association graphs can be built that represent the social connections between individuals. An overview is provided of the various motion and behavior pattern recognitions as well as the facial capture and social network estimation algorithms developed under this program. A performance evaluation was conducted based on the data collected at the 2009 Mock Prison Riot. The analysis indicates that the current system has an approximately 70-percent chance of detecting the occurrence of disorderly or aggressive events in the observed prison environment and currently has a 20-percent chance of predicting the event before it occurs. Examples are provided of how the technology developed can impact law enforcement operations. 62 figures and 41 references