Exploring the Limits of Machine Vision

Anthony Hoogs, Ph.D., director of computer vision at Kitware, told Automation World in an exclusive interview, “There is lot of progress in cheap 3D sensors."

Aw 6808 1110 Persp Eye

The computer vision cool factor goes way beyond today’s simple facial detection, object recognition and computer generated graphics. Instead of virtual reality, consider augmented reality. Just imagine your cell phone camera being marked up with interesting, useful information about what you’re looking at: who the people are, what a specific building is for, what type of car is in the picture. Or perhaps in robotics applications, where computer vision will be the solution to expanding the role of household robots, enabling them to see the people moving around them, avoid running them over, and even understand what they are doing.

Anthony Hoogs, Ph.D., is director of computer vision at Kitware (www.kitware.com), an open source software company that is receiving funding from multiple U.S. agencies including DARPA, DOE, NIH and NASA. Kitware has a group of 25 scientists and engineers working on advancements in computer vision.

Hoogs told Automation World in an exclusive interview, “There is lot of progress in cheap 3D sensors. We can do laser brain scanning at high rate. We can get this now at 5 Hz. to 10 Hz. and can envision putting this thing on a robot. It is a low power laser, so you can’t see it and it won’t hurt the human eye. Now if you’re in a dynamic environment, you can really know what that 3D environment is like. It is a much better technology than the stereoscopic camera—an idea that will fail.”

There is a whole new range of applications coming from this 3D ranging device, according to Hoogs: “They don’t take an exorbitant amount of power, so you can put them on robots, for example.” Noting the success of the Microsoft Kinect gaming system that uses computer vision, Hoogs thinks this will influence future developments in 3D description of the environment. “We’ll see quite a bit more applications with these breakthroughs,” he said. “We see acceleration of the pace of robots interacting in the human environment. A robot such as the Roomba can work safely with humans. And it will become more robust at detecting people.”

ON THE WEB: Machine Visions Sales Grow. Read how machine vision sales were up 68 percent in late 2010, as reported by the Automated Imaging Association (www.machinevision.org). Visit bit.ly/related_023.

One project Hoogs cites concerns getting a robot to understand what people around them are doing. The robot needs to see and recognize behaviors, so it can predict what it can do and where it can go. Hoog noted, “That’s an ambitious project.”

Hoog added, “There have been breakthroughs in person detection, [and] video rates to run person detection and detect across a range of clothing, heights, distance from camera and body positions. The biggest thing is detecting faces. The next thing is detecting the whole person—that’s a harder problem.”

Another constraint Hoogs noted is that, if you have a moving robot, it’s important to do vision not affected by the motion of the robot. “You can’t do 3D because of moving. That’s one reason moving robots are a problem,” he said.
All of this research into robots and vision will eventually lead not only to better manufacturing, but also to improved ways of life.

Several Automation World columnists have written about the short-sightedness of the U.S. Federal government, and Congress in particular, in supporting these efforts. One research project with advanced technology features a robot folding towels (see YouTube video www.youtube.com/watch?v=gy5g33S0Gzo). Predictably, U.S. Senator Tom Coburn of Oklahoma attacked it as senseless waste of spending. The robotics community rallied the Obama administration to keep it going.

Hoogs concluded, “There hasn’t been a lot of funding for vision for a while. We need a National Science Foundation call for ideas in the area. There’s a lot more funding in Japan than here.”    

Automated Imaging Association (www.machinevision.org)
Kitware (www.kitware.com)

Gary Mintchell, gmintchell@automationworld.com, is Editor In Chief of Automation World.

Subscribe to Automation World's RSS Feeds for Perspectives

More in Data