Abstract

This paper presents a microcomputer-based machine vision system to recognize and locate partially occluded parts in binary or gray level images. The recognition process is restricted to untilted, two-dimensional objects. A new edge-tracking technique in conjunction with a straight-line approximation algorithm is used to identify the local features in an image. Corners and holes serve as local features. The local features identified in an image are matched against all the compatible features stored for the model parts. The algorithm computes, for all image and model features matches, a coordinate transformation that maps a model feature onto an image feature. A new clustering algorithm has been developed to identify consistent coordinate transformation clusters that serve as initial match hypotheses. A hypothesis verification process eliminates the match hypotheses that are not compatible with the image information. The system performance was compared to a vision system restricted to recognize nonoverlapping parts. Both systems require the same hardware configuration and share the basic image processing routines.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call