Research On Camera Self-Calibration Based On Scene Information

With the rapid development of China's economic construction and its gradually raised international status, there are more and more social activities increasing everyday, and therefore more requirements of information constructions based on geographic information. The 4D data itself lack of environment information and is always with two degrees, although it contains lot of spatial and property information. The close-range images and videos which we are contacted with in daily life, contain a wealth of geometric information, so it is a big technological innovation and breakthrough of building new information services and enrich GIS information by extracting spatial information through photos and Videos. Digital Measurable Image can make up for many shortcomings of traditional GIS4D products, providing cm-level perception of the earth from a distant planet spacing to the ground. Therefore, precise analysis and modeling of image data, is the key of image information mining, and camera self-calibration is a very important step of restoring camera's imaging geometry, which makes it possible to reconstruct the target object.There are a lot of structural geometry objects in natural and humane scenes, this paper firstly gave a summary of existing camera calibration methods by calibration objects or templates, and tried to find the relationship between the objects and natural scene geometrical structures, then constructed natural calibrating objects system based on the scene information.In order to improve the accuracy of camera self-calibration based on scene information, this paper presents method based on the information of the photo itself——Camera radial distortion parameters calculation based on numerical digit. This method is simple, and the conditions are easy to extract. This step by step method approaches to solve the distortion of the traditional calculating without the complicated nonlinear equations, and can guarantee the accuracy in different resolution images. The accuracy can be set according to demand, and it can estimate the number of cycle efficiency of the algorithm, the method is a simple and efficient solution to solve radial distortion of the cameras.Most of the previous camera self-calibration methods use two or more images, which calibrate the camera by the multi-view geometry. Based on the natural calibrating objects system that is already constructed, this paper tries further study on how to do the calibration by limited scene information in a single image. We modeled the objects in the scene which is useful for the calibration, and then do the organization and coding. Finally, we proposed a new single-image calibration algorithm using a variety of multi-constraint geometric features, which improves the availability of image and video data.