Hoosiers join Japanese in subway sensor system
Distributed sensor network is capable of learning from human instruction; “a large-scale practical system that incorporates learning”
More evidence in favor of the value of behavorial recognition technology. Researchers at the Indiana University School of Informatics, joined by Japanese scientists at Wakayama University and Osaka University, have designed a distributed sensor network capable of learning from human instruction. In their Kyoto study, they blanketed a subway station with twenty-eight wide-view cameras, each one one of them a combination of flat and hyperbolic mirrors designed by professor Hiroshi Ishiguro. The key element, however, was the software, which permits human station operators to teach the cameras to recognize particuiar situations.
“This is a large-scale practical system that incorporates learning,” said professor Karl MacDorman. “Normally, the system designer develops an explicit model of human beings and specific modules for recognized different kinds of behavior. Our system allows for the system to be trained by non-technical station operators in an hour or two.” In one example, the system was taught to recognize overcrowding, and it is imagined that it could also recognize public misbehaviors such as graffiti and public urination.
