Unsupervised Feature Selection and Category Classification for a Vision-Based Mobile Robot

• DOI: 10.1587/transinf.e94.d.127
• Publication Date: 2011-01-04T06:08:42Z
• Authors: Masahiro TSUKADA; Yuya UTSUMI; Hirokazu MADOKORO; Kazuhito SATO
• Keywords: Scale-invariant feature transform; Feature (linguistics); Contextual image classification; Feature vector

This paper presents an unsupervised learning-based method for selection of feature points and object category classification without previous setting the number categories. Our consists following procedures: 1) detection description features using a Scale-Invariant Feature Transform (SIFT), 2) target One Class-Support Vector Machines (OC-SVMs), 3) generation visual words all SIFT descriptors histograms in each image selected Self-Organizing Maps (SOMs), 4) formation labels Adaptive Resonance Theory-2 (ART-2), 5) creation categories on map Counter Propagation Networks (CPNs) visualizing spatial relations between Classification results static images Caltech-256 dataset dynamic time-series obtained robot according to movements respectively demonstrate that our can visualize while maintaining characteristics. Moreover, we emphasize effectiveness appearance changes objects.