Research On Application Of Omni-directional Computer Vision Based On The Fisheye Lens

Vision based mobile robot navigation is one of the hot topics in robot field recently. Just like humans, vision is also the most important source for robot to know the world around. it can supply the information of the environment, such as target identification,real-time tracking and navigation. Omni-directional vision is a new computer vision technology, which can acquire 180-degree spherical field of view of an environment. With the property of large field of view, Omni-directional vision system can meet requirements that machine vision acquires teal-time environmental information. It is quite suitable for a wide range of applications. Omni-directional images generated from omni-direction vision system have problem of serious distortion because of the unique imaging processes of omni-direction system, so the information of the scene taken by the omni-directional vision needs to be recovered when applying the system to practices. This paper studies the distortion correction problem of omni-direction vision image taken by fisheye lens.This project is based on the international science and technology collaboration project supported by the Ministry of Science and Technology of the People's Republic of China. Firstly, the technology of omni-directional vision is introduced, and the background of the research is described, including the theory and the applications. Secondly, the difference between the conventional cameras and omni-directional vision imaging system is deeply studied, and the structure of fisheye camera and principles of optical imaging are detailed. Based on the fisheye camera imaging model, a mathematical model is established for the fisheye camera. For calibration problem, this paper proposes a method to get the internal parameters of fisheye camera. Compared with the traditional method of calibration, this method is easy to apply and avoids complex calculation. The result is high precise. Thirdly, according to the characteristics of the fisheye image distortion, two distortion correction algorithms are derived in this paper. One method is based on the isometric image model of fisheye lens. The function of the relationship between distortion image and ideal image is established. Another is regression algorithm based on Support Vector Machine (SVM) which is novel. With strong ability of classification and regression, SVM can provide a mapping between the fisheye image and the standard image for human eyes, which involves a coordinate conversion between fisheye image and real world view. A specific training target is needed for SVM to regress. Two methods are adopted by the VC program. The results of experiments verify the correctness of the algorithm.Lastly, using high-speed Digital Signal Processor (DSP) as the core hardware, this paper transplant distortion correction algorithm to DSP, completing a large number of image processing in the limited time and meeting the requirements of real-time. This work lay the foundation for vision navigation and appears extensive application.