Title: Precision Mobility Issues to Mobile Robots in Outdoor Industrial Environments

Authors: Sukhan Lee

Speaker: Sukhan Lee

Affiliation: Intelligent Systems Research Center, SKKU, South Korea

Abstract:
Due to a large variation of radiometric and visual contexts, robust real- time recognition and understanding of traffics, signs, and pedestrians by cameras mounted on-board moving vehicles are quite challenging. However, overcoming this challenge is essential should we achieve a true advancement of intelligent vehicles, especially, for the next generation of cyber transportation. In this talk, an approach to meet this challenge is presented. The proposed approach is based on the notion that human vision, or human perception in general, is dependable not because of the perfection of our eyes as a sensor but because of the perceptual behavior associated with visual processing that proactively seeks for and focuses on an optimal set of evidences for decision. More specifically, an interaction between data-driven bottom-up and model-driven top-down visual processing is defined under the framework of spatio-temporal image sequence based multiple evidence and model matching paradigm implemented by particle filtering, where a process of selecting and collecting an optimal set of evidences based on perceptual behavior is integrated into. Experimental results demonstrate the dependability of visual recognition of traffics and signs achieved by the proposed approach.