Study On Camera Pose Determination

Camera pose determination is one of the classic issues in computer vision, that is, to determine a general rigid transformation between the camera and the scene by 3D entities in the space and the corresponding 2D entities in the image. It may affect the following 3D reconstruction, target tracking and recognition directly. So, research on camera pose determination has great important significance of theoretical study and practical value. This paper is organized as follows:Firstly, the existing algorithms of the camera determination are compared which are divided into based on points, lines and quadratic curves respectively. The mature one is PNP(Perspective-N-Point) problem based on feature points. However, the difficulty of PNP problem is the nature of multiple solutions and the points is easily out of field of vision. Usually, it is necessary to select appropriate space points to avoid multiple solutions. However, from a practical point of view, it is often advantageous to use lines. Indeed, line detection is more accurate and more reliable than points and line matching is more robust than point matching owing to partial occlusions. One of the difficulties is that the space line has more complicated mathematical expression than that of the point. The methods on quadratic curves are based on ball and cylinder.Tough this approach is more resist hide compares to feature point and feature linear, that it is difficult to meet required scene to get unique solution hardly to suitable for real-time positioning.A new method for camera pose determination is proposed based on the surface of revolution by using its special symmetry in this paper. The algorithm needs scene which is common in daily life. The basic idea is to determinate camera position by using the symmetry of surface of revolution to find the vanishing point on x axis and z axis of the object reference. This method can avoid the required precise structure of space entities in previous determination approaches. Both synthetic and real data show that the proposed algorithm is feasible and efficient, it can be said that this technique is flexible and easy.