Machine Vision
SRI has pioneered research in machine-vision techniques for two-and three-dimensional object
location, recognition, and inspection since the 1970s. Our experience spans the spectrum
from basic research that generates new algorithms, through the development of real-time laboratory
systems, to the design and construction of cost-effective preproduction prototypes. Our work
has served as the basis for vision systems marketed by several commercial vendors.
3-D Object Recognition
SRI's 3-D object recognition algorithms have been applied to diverse
problems including the identification
of complex solid objects,
the precise
location of microscopic objects such as
optical fibers, and detecting the pose
of industrial parts
such as the
windshield wiper motors at far right.
Inspection for Defect Location
SRI has developed methods to detect subtle differences in
images. For example, subtle pixel-by-pixel variations indicate errors in printed watermarks,
differences in images taken with different light directions identify flaws in space-shuttle tiles;
and differences in texture measurement identify flaws in composite plies.
Vision for Robot Control
SRI has developed vision techniques for locating 3-D objects
fast enough to control high-speed robots. These techniques are applicable to parcel sorting,
multiarm coordination for ply layup, and visually guided robotic arc welding.
Video of visually guided robotic arc welding.
Robotics for Advanced Factory Applications
We have pioneered machine-vision techniques for detecting, locating, and measuring random objects.
Using these techniques in combination with advanced controllers and robotic actuators, we developed
systems that for the first time automate highly challenging tasks such as parcel handling,
manufacturing assembly, and parts feeding.
|