Security cameras are going to get a lot smarter, and a lot more automated, if a pair of Carnegie Mellon University researchers have anything to say about it.
Imagine a security camera that could detect when a bag is left unattended at the airport with no human help. No more need for monitor banks of security feeds with an fallible person behind the desk.
It sounds like science fiction, but that is exactly what a paper presented this week by researchers Alessandro Oltramari and Christian Lebiere proposes.
The U.S. government-funded research has developed what it terms a "Cognitive Engine" for artificial intelligence to automatically detect and interpret a person's actions through a surveillance feed.
Making cameras think
While computer vision technology can detect objects, understanding what those objects are doing requires the "Cognitive Engine."
For example, when a person picks up a ball they might bend over and outstretch their arms or simply kneel down and stand up again.
While video detection on its own would interpret those as two very different actions, the Cognitive Engine understands they are the same.
It would make connections between the person and ball, along with all possible interactions between the two, in order to replicate an almost human-like process of interpreting actions.
Predictive cameras for future tech
Oltramari and Lebiere believe that the research can have wide ranging applications, and that the Cognitive Engine will be able to "eventually predict" outcomes based on the actions it observes.
Bringing back the ball example, the Cognitive Engine could predict that the man will throw the ball by anticipating the most likely actions for what a person would do with a ball.
This predictive analysis is the ultimate goal of Oltramari and Lebiere's research, which is conducted as part of the DARPA Mind's Eye program to develop smart cameras for automated video surveillance.
The technology could even come home to consumer products eventually, possibly bringing Google's self-driving car, and our inevitable robot overlords, another step closer to a reality.
Via CNET, Alessandro Oltramari and Christian Lebiere research paper
