Research On Catadioptric Omnidirectional Vision System And Its Optimal Design

The omnidirectional system is a visual system which can cover a 360°horizontal field-of-view. It has been widely put into various applications such as mobile robot/tank navigation, scene surveillance, video conference, remote education and outer space exploration. Among all omnidirectional systems, the catadioptric system has been more and more popular due to its unique merits, which includes compact structure, strong reliability, low cost of building, great flexibility of designing.The main advances of this dissertation include:The systematic fundamental knowledge of omnidirectional system is summarized, and the research situation of the field is stated.Some important concepts, which are composed of the effective perspective projection model and the single viewpoint constraint, etc., are discussed in detail.Based on the single viewpoint constraint, all standard mirror profiles of single-camera-single-mirror catadioptric systems are discussed and compared.As a special case, a optimal design framework of spherical mirror catadioptric systems is proposed. In the framework, a series of principles, algorithms and pratical issues are summarized in order to build a spherical system. An overall analysis of the system parameters is done, and the relationship of a point in the scene and the correspondent image point is connected. An algorithm based on Hough Transformation is developed to extract the center and radius of the omnidirectional image. MATLAB programs are used to unwarp the image as well.An experimental system is built in the laboratory, which is used to verify the theory and framework that this dissertation puts forward. Pratical issues of component selection, installation and calibration are carefully discussed. The results of the experiments are analyzed, which prove the spherical mirror system is an ideal solution of wide-field-of-view low-cost real-time scene surveillance.