Year: 2007
Type: dissertation
Abstract: Mobile robots are increasingly finding application in fields as diverse as medicine, \nsurveillance and navigation. In order to operate in the real world, robots are primarily \ndependent on sensory information but the ability to accurately sense the real \nworld is still missing. Though visual input in the form of color images from a camera \nis a rich source of information for mobile robots, until recently most people have \nfocussed their attention on other sensors such as laser, sonar and tactile sensors. \nThere are several reasons for this reliance on other relatively low-bandwidth sensors. \nMost sophisticated vision algorithms require substantial computational (and \nmemory) resources and assume a stationary or slow moving camera, while many mobile \nrobot systems and embedded systems are characterized by rapid camera motion \nand real-time operation within constrained computational resources. In addition, \ncolor cameras require time-consuming manual color calibration, which is sensitive to \nillumination changes, while mobile robots typically need to be deployed in a short \nperiod of time and often go into places with changing illumination. \nIt is commonly asserted that in order to achieve autonomous behavior, an \nagent must learn to deal with unexpected environmental conditions. However, for \ntrue extended autonomy, an agent must be able to recognize when to abandon its \ncurrent model in favor of learning a new one, how to learn in its current situation, \nand also what features or representation to learn. This thesis is a fully implemented \nexample of such autonomy in the context of color learning and segmentation, which \nprimarily leverages the fact that many mobile robot applications involve a structured \nenvironment consisting of objects of unique shape(s) and color(s) - information that \ncan be exploited to overcome the challenges mentioned above. The main contributions \nof this thesis are as follows. \nFirst, the thesis presents a hybrid color representation that enables color \nlearning both within constrained lab settings and in un-engineered indoor corridors, \ni.e. it enables the robot to decide what to learn. The second main contribution of the \nthesis is to enable a mobile robot to exploit the known structure of its environment \nto significantly reduce human involvement in the color calibration process. The \nknown positions, shapes and color labels of the objects of interest are used by the \nrobot to autonomously plan an action sequence to facilitate learning, i.e. it decides \nhow to learn. The third main contribution is a novel representation for illumination, \nwhich enables the robot to detect and adapt smoothly to a range of illumination \nchanges, without any prior knowledge of the different illuminations, i.e. the robot \nfigures out when to learn. Fourth, as a means of testing the proposed algorithms, \nthe thesis provides a real-time mobile robot vision system, which performs color \nsegmentation, object recognition and line detection in the presence of rapid camera \nmotion. In addition, a practical comparison is performed of the color spaces for \nrobot vision – YCbCr, RGB and LAB are considered. The baseline system initially \nrequires manual color calibration and constant illumination, but with the proposed \ninnovations, it provides a self-contained mobile robot vision system that enables \na robot to exploit the inherent structure and plan a motion sequence for learning \nthe desired colors, and to detect and adapt to illumination changes, with minimal \nhuman supervision.
Authors: Benjamin Kuipers, Peter Stone, Mohan Sridharan
Institution: The University of Texas at Austin
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W1946011788