Global Image Orientation Method for PTZ Camera with Pure Rotation

Objectives: Pan-tilt-zoom(PTZ) camera is widely used in surveillance systems due to its wide field of view and high resolution. The camera mounted on the gimbal can only make pure rotation motion around the optical center. The optical center overlap of images makes it impossible to use existing Structure from Motion (SfM) methods for purely rotated images orientation. In order to estimate the elements of interior and exterior orientation in the free net of a set of PTZ camera images with pure rotation, a global image orientation method for PTZ camera with pure rotation was proposed, which motivated by the idea of the global SfM. Methods: Compared with the general SfM method, the similarities are feature extraction and feature matching, and the differences are interior orientation, exterior orientation and bundle adjustment. Firstly, partial image pairs are selected for inner orientation to estimate the elements of interior orientation. Then, global exterior orientation was implemented for all images, and the elements of exterior orientation were estimated. Finally, the elements of interior and exterior orientation are optimized by bundle adjustment with no object 3d point. Results: Experiments on simulation data and real data prove the feasibility and accuracy of the proposed method. In the simulation data experiments, the maximum error of the focal length and the principal point is 3.321 pixels, and most of the errors are less than 1 pixel. The maximum rotation errors of the four datasets are 0.116 degrees, 0.320 degrees, 0.103 degrees, 0.125 degrees. In the real data experiments, the maximum reprojection errors of the checkpoints are 4.919 pixels and 4.758 pixels for the two datasets, respectively. Conclusions: Compared with the existing global SfM method, the proposed method can successfully orient the images of PTZ camera with pure rotation. In addition, compared with other commercial software, the exterior orientation accuracy of the proposed method is higher, and it also has the advantage of computational efficiency. The method presented is not limited to PTZ camera images, and it has universality for other purely rotated images orientation problems.