Recognition and positioning of three-dimensional objects by combining matchings of primitive local patterns

Abstract

A model-based 3D object recognition and positioning method is proposed. The proposed recognition method is based on the idea of combining many local matchings into a comprehensive and consistent interpretation of the scene: primitive local patterns in the scene are matched with those in the model by using the edge and surface information together, and the compatible local matchings are accumulated and combined to give several interpretations of the scene. The most appropriate interpretation is selected as the recognized object. In this paper a pair of adjacent facets is adopted as the primitive local pattern. The proposed method is robust to occlusion and works with only a partial view of the object. This method is also suitable for those objects which have very similar shapes when viewed from a particular direction. The method was tested on images of realistic 3D object scenes obtained with a laser range finder.