- B.S., control and computer engineering, Istanbul Technical University, 1997
- M.S., computer science, University of Southern California, 2000
- Ph.D., robotics, Carnegie Mellon University, 2007
Digital SLRs capable of shooting Hollywood-quality footage are available for only a few hundred dollars. High-resolution cameras are part of nearly every smartphone. And practically every tablet or notebook computer can stream a movie or TV show.
But although digital video seems to be everywhere these days, Goksel Dedeoglu believes we still have a way to go before computers will be able to watch video and routinely understand it, without human help.
“We have built great multimedia processors for human media consumption, but we haven’t quite figured out the right hardware or software for computer vision applications,” he says. “When a computer is just trying to extract information (from video), it doesn’t necessarily need a full 30 frames per second, and if it’s just for algorithmic consumption, it doesn’t need to store the video for later, either.”
“Computers will eventually be able to autonomously read and interpret video in real-time,” Dedeoglu says, “but we have a few fundamental gaps that need to be closed.”
Dedeoglu’s new company, PercepTonic LLC, is all about closing those gaps. PercepTonic consults with software and hardware companies to help them incorporate computer vision into their products. “Computer vision is all about extracting information from images,” says Dedeoglu, “and it’s starting to cross the chasm from being an R&D topic to a product feature,” in applications such as collision avoidance and self-parking cars.
The improved quality of digital image sensors is one factor making computer vision affordable in consumer applications. Another factor, according to Dedeoglu, is powerful and inexpensive embedded processors. “People have being doing research for years into understanding human facial expressions, for instance,” Dedeoglu says. “We now have enough computing power on our portable devices to make using these algorithms a reality.”
One of the leading pioneers in computer vision is Takeo Kanade, CMU’s U.A. and Helen Whitaker University Professor of Computer Science and Robotics. When Dedeoglu was working on autonomous robots as an undergraduate and master’s student, he naturally became interested in computer vision. Going to CMU to work on his doctorate under Kanade “was a no-brainer.”
“I learned a lot from Takeo,” Dedeoglu says. “Takeo always says, ‘Think like an amateur, but execute like an expert.’ In other words, let yourself be creative and not limited by what you know is possible, but when it comes to doing something, do it with the best tools and the best algorithms.”
It was good advice that served Dedeoglu well during his seven years in the Embedded Vision R&D lab at Texas Instruments. “Some of the popular things people are doing in academia don’t map to consumer products,” he says. “If you’re going to do something, you don’t do it for the sake of having a more elegant mathematical model, you do it to solve a real problem and ship a product.”
PercepTonic is based in Texas, where Dedeoglu and his wife, Susan Rossbach, make their home. It’s also where they’re raising their son, who was born in Pittsburgh. “I was a robo-grad (robotics grad student) and she was a member of the RI technical staff, so we used to joke that our son was a robo-baby,” Dedeoglu says.
Jason Togyer | 412-268-8721 | jt3y [atsymbol] cs.cmu.edu