SHAPE RECOGNITION OF 3-D OBJECTS FROM STEREOVISION DATA BY USE OF OBJECT-BASED MRF MODEL

摘要

In this paper, we propose a shape recognition method of three-dimensional (3-D) block-like objects from stereovision data. Flat planes and ridge lines of the objects are represented by 3-D primitive models, and their interrelations are formulated by use of a 3-D Markov Random Field (MRF) model that is extended so as to have the object models as its elements. The object shapes are recognized by minimizing the energy function of the MRF model. Experimental results are shown for synthetic and actual stereo data.