Vision-based bin-picking is increasingly more dicult as the complexity of target objects increases. We propose an e cient solution where c omplex objects are su ciently represented b y s i m p l e f e atures/cues thus invariance to object complexity is established. The region extraction algorithm utilized in our approach is capable of providing the focus of attention to the simple cues as a trigger toward r ecognition and pose estimation. Successful bin-picking experiments of industrial objects using stereo vision tools are p r esented.

No abstract

No abstract

Our work addresses the problem of using robot vision to recognize and localize objects in outdoor environments. Successful applications of robot vision are mostly found in indoor environments where illumination and backgrounds can be carefully controlled. For outdoor applications, a vision system must be robust with respect to the changing ambient illumination, not to mention the difficulties caused by complex and varied backgrounds against which an object must be recognized. In this paper we describe an outdoor stereo-vision brick recognition system for a construction robot. Our algorithms can accommodate large variations in the reflectivity properties of the background and shadowing effects caused by adjoining buildings and other features of the landscape. The performance of the algorithms remains unaffected when the background is changed from, say, dirt to grass. We also have achieved tolerance with respect to the changing sun angle and some shadowing caused by clouds. ,

A new general-purpose color image segmentation editor (CIE)