A hierarchical modular vision system for object recognition

Computer vision constitutes a fundamental component in robotic applications. Object recognition and the determination of object poise are necessary steps for intelligent manipulative tasks.

In this thesis, a hierarchical modular computer vision system for object recognition was developed. It combines statistical and structural techniques, and presents an approach to the problem of associating attributes with a string of pattern primitives.

Results show that the system can recognize objects in the presence of rotation, translation, size variation, sampling size and moderate amounts of noise.