Selection of SIFT feature points for scene description in robot vision

This paper presents a method for selection of SIFT(Scale-Invariant Feature Transform) feature points using OC-SVM (One Class-Support Vector Machines). We proposed the method for automatic category classification using a network system that combine incremented learning function of ART-2(Adaptive Resonance Theory-2) networks and self-presentation characteristic of CPN (Counter Propagation Networks). In our former methord, the feature points of nontarget region make misclassification. In our method, OC-SVM selects feature points of target region. Experiment results that used Caltedh-256 object category dataset and the time-series image dataset caught from a mobile robot show that classification accuracy of our method is better than that of the former method. Key words SIFT OC-SVM ART-2 CPN

• Type: Journal
• Year: 2010
• Where: SMC
• Authors: Yuya Utsumi, Masahiro Tsukada, Hirokazu Madokoro, Kazuhito Sato
• Keywords: Automatic Category Classification, Combine Incremented Learning, Control Systems, Counter Propagation Networks, SMC 2010