Parallelepiped-indoor navigation for mobile robots using monocular vision-based positioning

Abstract

In this dissertation, we concentrate on development of mobile robot navigation system using COTS monocular vision system without a map. We present navigation system with path planner, localizer and map builder. Also, we develop positioning algorithm using mathematical formulation under geometric constraints that is derived in indoor environment. At last, we verify their validity by various navigation experiments. Chapter 2 gives basic backgrounds needed for developments in the following chapters. First, a coordinate system model and a definition of positioning is given. Next, camera calibration method is given and followed by bucketing, the clustering method. And, dynamic programming, line matching algorithm used in the dissertation is given. Last, least square fitting is described, which is widely used in deriving a straight-line equation among a set of data points. In Chapter 3, we describe an efficient positioning algorithm for autonomous navigation of mobile robots referring prerecorded images without any environment maps. The algorithm estimates the robot`s relative position and orientation with respect to the reference robot, imaginary robot corresponding to the reference image. We use vanishing points to measure the robot`s orientation. To estimate robot`s position from reference and current images, dominant vertical and oblique lines are used. In the processing, we use as much as information as possible for accuracy. This method results in low-cost, fast positioning in the restricted environments. Chapter 4 presents a byproduct of Chapter 3. In this chapter, we present another formulation of position determination method for autonomous indoor navigation of mobile robots using monocular vision. We utilize vanishing points to determine the robot orientation. For accuracy, we use multiple vanishing points with weighted-sum method. We also use oblique lines corresponding to floor-wall edges and ceiling-wall edges to determine the lateral position and ver...