Object Recognition with Multi Visual Cue Integration for Shared Knowledge-based Action Recognition System
Kei Okada, Mitsuharu KOJIMA, Masayuki Inaba · Journal of the Robotics Society of Japan · 2008
We have been tackled action recognition system design for a real world robot in a constructive approach. We have proposed the integrated system that shares knowledge-base between action and recognition. In this paper, we focus on a knowledge based object recognition system. Our system enables navigate attention to a visual search area, view prediction based recognition, robust recognition with multi-cue integration and time-series stochastic filter. Developed object recognition system is able to recognize both self location and object location by means of knowledge-base information, so that it is able to apply to humanoid tasks that contain both manipulation and locomotion. Finally, we showed a daily environment tasks consists of tea pouring, cup carrying and wash dishing in order to evaluate an availability of proposed shared knowledge-based action recognition system and object recognition method.