Catadioptric vision based localization and mapping for indoor mobile robot

Abstract

Localization is not only a fundamental problem in mobile robot but also key prerequisite for successful task accomplishment. However, there have been so many problems caused by the variety of environment and their dynamic characteristics. In this dissertation, I propose a novel Stratified Localization and Mapping (S-LAM) based on a single catadioptric vision sensor and demonstrate the robustness and practicality in applying the algorithm to the navigation of mobile robot. First, a novel SLAM algorithm, named PR-SLAM, is proposed. It solves the complexity problem arising in adopting a particle filter in SLAM. By partitioning the state and alternating the turns for the state update, the computational capacity required to process SLAM is reduced to scale linearly with the number of landmarks in the map. Second, the Colored Occupancy Map (COM) generation method is proposed, which compensates LRF with a single camera. Third, a robust global localization algorithm based on particle filter is proposed. These algorithms are unified and stratified in a single framework for the localization and mapping. In addition, I proposed a novel single camera catadioptric stereo vision system which expands the problem of localization and mapping of the proposed framework from two to three dimension. Various experiments conducted on natural corridor environment which span 150m demonstrate the feasibility of the proposed algorithm. The experiments on the proposed single camera catadioptric stereo vision sensor show the accuracy of the 3D reconstruction and real-time performance which is very important for robotic applications.