Camera Pose Measurement Using Symmetrical Repeated Texture

Camera pose measurement is a fundamental problem in photogrammetry and comupter vision. Its main task is to acquire the orientation and position of a camera. In a man-made environment, symmetrical repeated texture,with a low-rank image matrix,exits on surfaces of many objects. When images of these texture are acquired off the ideal frontal position, symmetry and repetition of the texture in the original direction can no longer be retained due to projective deformation, which leads to the loss of low-rank property of its image matrice. In this paper, a camera pose measurement method of 6 degree of freedom (Dof), whose target objects are planes with low-rank texture, is proposed. Compared to other pose determination methods based on points or lines, the proposed method of camera orientation estimation is based on global features of images, which results in better robustness and higher accuracy. The main contents are as follows:A low-rank camera orientation measurement model is established and solved by iterative methods. Firstly, original low-rank texture image and camera acquired real image are regardes as ideal frontal image and tilted image respectively. Then, relationship model of the ideal frontal image and the tilted image is established based on the projective imaging model of viewing the same points from two different poses. After these steps, camera orientation estimation can be made equivalent to a minimum problem of the rank of texture image, which subjects to constraints of the relationship model. When solved by iterative methods, the tilted image is continuously recovered until it becomes closely to ideal frontal image. When iteration stops, we can acquire camera orietaion parameters from the results of the algorithom.After the orietaion of the camera is acquired, camera position in 3D space is then calculated. Firstly, two points with known position are chosed. Then, coordinates of the origin of the world frame are deduced based on the geometric relationship and imaging model. Lastly, the camera position in 3D space is acquired through coordinate transformation.Simulation and real data experiments under different experimental conditions are implemented to verify the effectiveness of the proposed method. In experiments part, camera orientation estimation results, under different rotaions and noisy conditions, are firstly presented. Then ceiling texture and wall-brick texture are adopted to estimate the pose of the camera. At last, the proposed method are compared to other methods in similar scenes.