The Study On Calculating And Interpreting The Optical Flows Of Conics

An image of a scene is a two-dimensional representation of a three-dimensional world. In the processing of projecting the scene onto the two-dimensional image plane, some of the information about the three-dimensional scene is inevitably lost. Given a series of images of a scene, typically taken by a video camera, it is sometimes possible to recover some of this lost three-dimensional information. The extraction of the three-dimensional motion and structure of a moving object from a sequence of images is a termed the structure from motion (SFM) problem. The solution to this problem is a key step in the computer vision tasks of robotic navigation, obstacle avoidance, time to collision, recognition of solid objects, and surveillance.One of the fundamental problems in early computer vision is the measurement of motion in a time-varying image, frequently called optical flow which is an intermediate representation widely used for recovering three-dimensional information in motion analysis. The optical flow represents a set of corresponding features in consecutive images. The problem of estimating the optical flow has received much attention because of its many different applications. Tasks such as passive scene interpretation, image segmentation, structure from motion, inference of egomotion, and active navigation, all use optical flow as input information.In this thesis, we study the computation of the optical flows of conies from the real images and synthesized images, and the determination of motion and structure from the optical flows of conies. A new approach to the problem of visual motion estimation is developed and evaluated.Firstly, we have briefly introduced the current situation about two kinds of methods of visual motion analysis: the feature-based method and the optical flow-based method, and pointed out some of problems that have not been solved.Secondly, We have given a brief description of the fundamental technique: image brightness constancy equation, and several of the implementation specifics. A method of representing a conic in three-dimensional space is proposed, and the concept of the coordinates of the conies in two-dimensional and three-dimensional space is defined, then the definition of the optical flow of a conic is given.Thirdly, we have introduced a series of processing of the edge detection. We have used three kinds of edge detectors to detect edge curve from digital image, and compared the obtained images that are detected by using the three kinds of edge detectors: Sobel's, Roberts's, and Prewitt's, among which the Sobel's edge detector is the best for a synthetic image or a real image with noise. Furthermore, an advanced and effective method of edge detector that combines theSobel's edge detector with edge thinning and contour following has proposed. A new method for representing and extracting edges (conies) from a primitive image, namely the projections of conies in three-dimensional space is proposed. All of the algorithms such as the edge detection, the edge thinning, the contour following, the edge linking, the edge representing have been implemented by using of the C++ Builder. Experiments have been conducted on artificially images and with real world data.Lastly, the mechanism and equations for estimating the three-dimensional motion parameters and structure parameters of a moving object in the three-dimensional scene from the optical flows of conies have been established.This thesis is the project supported by Aeronautical Foundation Science ( No.99F53065) and Research Center of Measuring and Testing Technologies, and Control Engineering in Nanchang Institute of Aeronautical Technology(No.2001-15).