Ground surveillance is a task normally performed by soldiers, but military leaders would like to shift the task to unmanned systems, removing troops from harm’s way.
Humans perform a wide range of visual tasks with ease. They can analyze scenes and make informed decisions as to what actions should be taken based on what they see. But today, current computer-based artificial intelligence systems lack the visual intelligence to perform such tasks.
Now however, the Defense Advanced Research Projects Agency (DARPA) is addressing this problem with Mind’s Eye, a program aimed at developing a visual intelligence capability for unmanned systems.
To do so, DARPA has engaged 12 research teams to develop software suitable for employment on a camera for use on an unmanned ground vehicle which will make use of both computer vision and artificial intelligence techniques. DARPA hopes that smart cameras equipped with such software would be able to report only on activities of interest, enabling a single soldier to monitor multiple observation posts from a safe location.
A key distinction between the new software and current state of the art machine vision software is that while the latter has been used to recognize a range of objects and their properties, the Mind's Eye software would be able to recognize actions it had never seen before, describe observed events using simple text messages, and flag anomalous behavior.
Researchers at Carnegie Mellon University, Co57 Systems, Colorado State University, Jet Propulsion Laboratory, MIT, Purdue University, SRI International, State University of New York at Buffalo, TNO (Netherlands), University of Arizona, University of California Berkeley and the University of Southern California are all taking part in the research program.
More information on the Mind’s Eye program can be found here.
-- by Dave Wilson, Senior Editor, Vision Systems Design