Component-Based Object Recognition Algorithm

Abstract

The paper presents an approach to object recognition based on the hypothesis of representing objects using a set of geometric primitives and relations between them. The goal of the paper is to develop a method for object recognition in the environment, which allows to recognize objects based on their description. For this purpose, the following tasks are solved: the recognition of a set of geometrical objects (primitives), the estimation of relations between primitives and the search of correspondences between the found primitives and relations and the defined templates (descriptions objects). The set of geometric primitives is selected taking into account the nature of the subject area of the objects to be recognized. The paper presents object recognition examples through the use of the method proposed. As a result, the operability of the proposed object recognition method is confirmed. An object description method has been developed. For experiments, the images of primitives were used generated in the Blender 3D, as well as photos of primitives from the kid’s toy constructor. The primitive detection model was trained on a training sample consisting of 1000 artificial images and 50 real images. The research results can be applied in algorithms for recognizing traffic participants as well as traffic signaling objects