A machine vision system for the recognition and positioning of two-dimensional partially occluded objects

The paper addresses the problem of recognition and positioning of two-dimensional partially occluded objects. A computer vision algorithm that recognizes and locates partially occluded objects is presented. The approach is based upon matching simple descriptions of the scenes and the models by a technique known as HYPER (Hypotheses Predicted and Evaluated Recursively) of hypotheses generation and verification coupled with a recursive estimation of the model to scene transformation. The HYPER technique is modified to be capable of recognizing and locating upside-down reversed objects in the scene description. Models and scene descriptions are constructed using an industry standard computer aided design (CAD) system which represents a very powerful tool to test the effectiveness of the proposed system. Experimental results which illustrate all phases of recognizing and locating the objects under a variety of scene types and conditions are described showing success of the proposed system