Computational and Information Sciences Directorate Research Areas

Distributed Visual Surveillance

Advisor: P. David

Key words: image processing, computer vision, visual surveillance, distributed algorithms

A large number of surveillance cameras are being deployed in many types of environments. It would be nearly impossible for human operators to continually monitor the video feeds produced by these cameras. To help reduce the burden on the human operator, we are developing algorithms and systems for automatic persistent surveillance from multiple visual and infrared cameras.

The goal is to continually monitor (24/7) an environment in order to detect events deemed by a human to be significant or unusual. Indoor and outdoor urban environments are our current focus. Objects, including people and vehicles, are detected and tracked from both stationary and moving cameras.

Cameras must self-calibrate so that locally detected events can be related to events observed by other cameras in the network as well as to a global world model. Analysis of human and vehicular activity is performed in a coordinated and distributed fashion over the network of cameras using local processors. Events detected by cameras in one part of the network should cue other cameras to investigate the event in order to obtain more detailed or complete information about that event. Events of interest to the human are identified and all relevant information is retrieved for the operators review.

We perform research and development of image processing and computer vision algorithms that support visual surveillance applications.

Research is required in the following areas:

• image enhancement
• object detection, recognition, and tracking
• behavior analysis
• content-based image retrieval
• 3D scene reconstruction
• motion analysis
• multi-sensor coordination and fusion
• real-time algorithms