Neither rain, nor fog, nor sleet thwarts this searchlight

Engineers at San Antonio's Southwest Research Institute are combining real-time image processing and machine vision with traditional surveillance cameras to scan through reams of video data and spot suspicious events.

Amy Higgins

July 24, 2003

Machine vision gives security cameras an edge

The top image shows the underside of a car, but there's something extra. In the bottom image, enhanced by machine vision and image processing from Southwest Research Institute, the "something extra" is automatically highlighted, making detection much easier.

The system also gives security officers instant replay and other options that simplify monitoring large, complex areas. Previous systems relied on simplistic approaches to detect incidents and were prone to false triggers generated by moving foliage, passing headlights, and cloud shadows.

"Our system uses algorithms with temporal processing and model-based analysis to recognize motions common to particular scenarios, such as moving people and vehicles," says Brent Nowak, manager of automation engineering at the Institute's Automation and Data Systems Div. "It also disregards or overcomes false triggers caused by moving foliage and shadows and can detect stationary objects left at the scene, such as packages and suitcases."