At CMU, researchers are working on software that can not only monitor surveillance video, but also predict what will happen next and prevent crimes before they occur, Phys.org reported. The system would sound an alarm upon detecting behavior considered suspicious.
Details about the project, which is funded by the U.S. Army, were published this week in a paper called Using Ontologies in a Cognitive-Grounded System: Automatic Action Recognition in Video Surveillance [PDF]. Alessandro Oltramari, a postdoctoral researcher, and Christian Lebiere, both from the Department of Psychology at CMU, suggested that the system could be used in both military and civil environments.
The system's cognitive engine identifies actions such as walk, run, carry, follow, pick up and chase, and micro-actions such as bend over, drag and stop.
In addition to identifying what people are doing, the cognitive engine is also able to couple visual signals with background knowledge to draw conclusions. For instance, the algorithm “knows” that cars move, every car needs a driver to move, drivers are people located inside of cars, and if a car moves then a driver inside the car will move along with it. The system can understand things that may sound obvious to people, but can be troublesome for computers.
Researchers said they plan to extend the functionality of the software to include more verbs and run tests on a larger video data set.
Photo courtesy of Shutterstock.com
