Vision Based Robot Localization Using Artificial And Natural Landmarks PDF Download

"Mobile robotics is an active and exciting sub-field of Computer Science. Its importance is easily witnessed in a variety of undertakings from DARPA's Grand Challenge to NASA's Mars exploration program. The field is relatively young, and still many challenges face roboticists across the boards. One important area of research is localization, which concerns itself with granting a robot the ability to discover and continually update an internal representation of its position. Vision based sensor systems have been investigated, but to much lesser extent than other popular techniques. A custom mobile platform has been constructed on top of which a monocular vision based localization system has been implemented. The rigorous gathering of empirical data across a large group of parameters germane to the problem has led to various findings about monocular vision based localization and the fitness of the custom robot platform. The localization component is based on a probabilistic technique called Monte-Carlo Localization (MCL) that tolerates a variety of different sensors and effectors, and has further proven to be adept at localization in diverse circumstances. Both a motion model and sensor model that drive the particle filter at the algorithm's core have been carefully derived. The sensor model employs a simple correlation process that leverages color histograms and edge detection to filter robot pose estimations via the on board vision. This algorithm relies on image matching to tune position estimates based on a priori knowledge of its environment in the form of a feature library. It is believed that leveraging different computationally inexpensive features can lead to efficient and robust localization with MCL. The central goal of this thesis is to implement and arrive at such a conclusion through the gathering of empirical data. Section 1 presents a brief introduction to mobile robot localization and robot architectures, while section 2 covers MCL itself in more depth. Section 3 elaborates on the localization strategy, modeling and implementation that forms the basis of the trials that are presented toward the end of that section. Section 4 presents a revised implementation that attempts to address shortcomings identified during localization trials. Finally in section 5, conclusions are drawn about the effectiveness of the localization implementation and a path to improved localization with monocular vision is posited"--Abstract.