The problem with machine vision is that the cameras are two-dimensional recorders of three-dimensional scenes. Objects in an image may be obscured by lighting, occluded by obstacles or camouflaged by similar colors in the background. Sophisticated software can sometimes piece together ob- jects from subtle cues, such as when two camera positions are used to reveal parallax. But such algorithms take time to run, making them inappropriate for real-time applications, like automobile collision avoidance. Now one company claims to have solved the problem with a real-time 3-D camera that uses pixel-level hardware to reveal the distance to any object in any scene, regardless of lighting, occlusion or blending. By integrating 3-D hardware into its SunShield CMOS 3-D time-of-flight imaging sensors, Canesta Inc. (Sunnyvale, Calif.) says it enables machine vision cameras that perceive objects, rather than just sense images, in a scene.
Text: http://www.eetimes.com/showArticle.jhtml?articleID=196513738