Study On Fisheye Binocular Vision Imaging And Positioning Technology

With the development of computer vision technology, the angle range limitations of conventional camera system cannot meet the demand on many occasions. Fisheye camera has an advantage of wider field of view and more information could be obtained, thus attract more and more attention in military and civil field. Fisheye binocular vision technology is an important research direction of computer vision. It has great prospects in the field of robot navigation and 3D reconstruction. This thesis conducts systematic researches on fisheye camera modeling, camera calibration and the space points positioning, which has a certain guiding significance for the application of fisheye binocular vision. The main contribution as follows:(1) The thesis starts with camera projection model and introduces transitions matrix between coordinates. Second, general imaging model and fisheye imaging model is introduced respectively. Then, the commonly used nonlinear distortion models are summarized. Finally, fisheye binocular vision positioning model is discussed and derived based on general lens binocular vision positioning model.(2) Due to the low calibration accuracy for fisheye lenses by using circular control points, this thesis proposes a fisheye camera calibration algorithm optimized by centroid method. Firstly, initial values of intrinsic and extrinsic parameters of imaging model are obtained. Secondly, centroid of ellipse detected from image and ellipse projected by imaging model are solved respectively. Then distance between those two centroid is minimized and cost function is established in order to optimizing the model's parameters. Finally, parameters of imaging model are calibrated precisely. Experiment results show that our method can improve the accuracy of calibration for fisheye lens.(3) Due to the big position error and strong limitations by using high odd polynomial method and optical refraction model, this thesis proposes an optimization of fisheye lens imaging position algorithm. Firstly, decentering distortion model is derived and optimized imaging model is established based on the distortion model and optical refraction model. Second, after using Taylor expansion approximation method, physical image points are derived. Finally, space targets are precisely positioned. Experimental results show that the proposed algorithm has far less relative positioning error than previous method when the measured object respectively in fisheye image center and edge areas. Compared with previous method, the image distortion of proposed algorithm obviously has less influence on measurement precision and it effectively improves the positioning accuracy.