Object Recognition and Pose Estimation Based on Principle of Homology-Continuity

Abstract

Based on manifold ways of perception, this paper describes a novel method of object recognition and pose estimation within one integrated work. This method was inspired by bionic pattern recognition and manifold learning. Based on the principle of homology-continuity, we establish shortest neighborhood graph (SNG) for each class and regard it as a covering and triangulation for the hypersurface that the training data distributed on. For object recognition task, we propose a simple but effective classification method, named SNG-KNN. For pose estimation, local linear approximation method is adopted to build a local map between high-dimensional image space and low-dimensional manifold. The projective coordinates on manifold can depict the pose of object. Experiment results suggest that the recognition performance of our approach was similar and sometimes better compared to the SVM method; moreover, the pose of object can be estimated.