The Visionary: Building Computers That See
Professor Dr. Konstantinos Derpanis is utilizing computer technology to recover useful information from images: essentially developing computers that 'see'. With the support of supported by NSERC Discovery and Engage grants, his lab, the Ryerson Vision Lab (RVL), is broadening the scope of how we understand video as it relates to the world around us.
“We have witnessed a deluge of video content due to advances in computing power and networking technologies,” says Derpanis. “YouTube, for instance, reports that 65 hours of video are uploaded to their service every minute. Yet most current solutions rely on humans visually inspecting videos to extract meaning.”
Derpanis is developing two automated solutions to mine videos for information. The first, geometric, involves the recovery of the three-dimensional layout of a surrounding environment. The second, semantic, is about interpreting what people are doing in such an environment. The potential for these solutions can be applied in the practice of video indexing and browsing, intelligent surveillance and human–computer interfaces. “Humans are ubiquitous in many video types and their actions represent a prime source of information for scene understanding,” says Derpanis.
His team has found automated solutions for both spotting and classifying human actions in videos. This research has the potential to develop into a comprehensive online video-searching system – one that doesn't need to rely on painstaking human labor to sift through the information. If computers can see, then we can create better ways of retrieving and classifying documents –affecting everything from filing to getting rid of junk mail.