Stereo Computer Vision for Observing People
This project uses stereo computer vision to observe people. The primary goals are (a) to detect people in indoor and outdoor environments, (b) to do a 3D analysis of the position and motion of people e.g. to locate a person's position on a known map of the environment, or to compute statistics about the flow of people through an area (c) to understand the activity of people e.g. to classify standing-walking-running people. This technology is being developed in conjunction with other projects, on face detection and analysis, to create a framework which supports many applications of computer-human observation.
Background & Objective: This work has many application areas including (a) surveillance - detecting intruders in a restricted area, (b) intelligent buildings - detecting people, estimating the number of people in an area, doing adult/child classification (c) intelligent traffic systems - estimating the size of crowds or lines in public transport hubs or at road-crossing places, detecting people in dangerous situations such as close to the edge of a train platform.
Technical Discussion: The major components of the system are feature detection, change detection, and stereo matching. Change-detection is based on features, rather than on pixels, as a superior approach with resilience to changing illumination. The stereo matching is currently based on a trinocular (3-camera) system, but is easily extendible to an arbitrary number of cameras. Cameras are becoming very cheap, so multi-camera stereo devices are likely to become much more common in the future.
Technology Area: Computer Vision
Modification Date: June 13, 2008