Recognition of partially occluded flat objects

This paper is focused on the object recognition problem in computer vision under partial occlusion. The approach followed to carry out this goal is the alignment method described exhaustively in the literature. In this approach the recognition process is divided in two stages: in a first stage, the transformation in space between the viewed object and the model object is determined. In a second stage the model that best matches the viewed object is found. Given four points in the image, it is necessary to find the four corresponding points in the model. This problem involving combinatorial search is resolved by means of a genetic algorithm. The occlusion problem has been dealt with special attention, so a new method has been proposed consisting of three processes: identification, grouping and verification. The recognition algorithm proposed here has been tested in several examples obtaining good results.