A Probabilistic Method of Bearing-only Localization by Using Omnidirectional Vision Signal Processing

Abstract

Self-localization is the most necessary foundation of mobile robot's navigation system. Omnidirectional vision system has been widely used in mobile robot. However, the accuracy of vision-based methods is not enough for many applications. In this paper, based on bearing-only localization, a probabilistic method is presented to accurately estimate mobile robot's position making use of omnidirectional vision signal processing. Firstly, landmarks and bearings are detected by omnidirectional vision signals. Then, a probabilistic mapping composed of coordinates' weight is constructed. The weight indicates the possibility that the robot's location is the coordinate of the point. Finally, Monte Carlo Localization is used to merge this probabilistic mapping with odometer-based prediction model. Experiments in several scenarios show that our method using omnidirectional vision signal processing is robust and accurate for mobile robot localization.