Determining Three-Dimensional Motion Of Rigid Object From Image Sequence Using Correspondences Of Conic Curves

This master thesis is the project supported by Aeronautical Foundation Science ( No. 03I53062 ).Visual motion analysis is a major field in computer vision. Image sequence analysis is a very important problem in visual motion analysis. It's task is to detect motion object and reconstruct three-dimensional motion and structure parameters from image sequences. Once the motive object have been detected, computer vision system can recognize it and estimate the current and future object location and pose with the motion and three-dimensional structure information, it can be used to resolve the problem of object recognition and object tracking. Visual motion analysis involved the knowledge of image processing, computer vision and artificial intelligence. It is widely used in industry, military, medical field and so on.In order to detect the edge contours of the object in images, several edge detectors are firstly introduced and we make a comparison of them in performance. A method for fitting a two-dimensional curve based on slopes of the curve is also shown. There are two kinds of methods for determining three-dimensional motion and structure of a rigid object which are those based on features and based on optical flow in visual motion analysis. Recovering three-dimensional motion and structure by using of correspondence of straight lines and the computational theory of optical flow for conic curve are expounded in article. The representation of a conic curve in three-dimensional space and optical flow based on conic curves in two-dimensional plane are introduced.Three-dimensional motion of a rigid object can be decompounded as a translation and a rotation around a straight line that passes through origin of the coordinates system. Based on the model of perspective projection where the camera is fixed whereas the object is motive, the problem of recovering the three-dimensional motion and structure of a rigid object from image sequences by using of correspondence of conic curves is researched. The motion models and the corresponding motion equations for pure rotation and pure translation are established respectively. Several algorithms are proposed for calculating the pure rotational motion parameters and the pure translation motion ones using correspondence of conic curves respectively. It is shown that a system of linear equations is resolved toobtain the rotational parameters by using of seven couple of conic curves in image sequences when the motion of a rigid object is pure rotational. When the motion of a rigid object is pure translation, a system of non-linear equations is resolved using iterative method, or using the method for resolving quartic equation. The translation parameters with a scale factor may be determined and the translation direction can be obtained. There is only one couple conic curve in image sequences needed in order to determine translation direction. How to reconstruct the three-dimensional structure of conic curve in three-dimensional space from translation parameters is discussed.