The Study On Camera Self-calibration Techniques In Computer Vision

Computer vision techniques have been developed fast along with those of computer graphics and pattern recognition. The calibration of cameras is an indispensable step to obtain 3D geometric information from 2D images. The calibration of a linear camera has become one of the major research areas in computer vision technology.Camera calibration techniques can be classified into two types: traditional camera calibration and self-calibration. In the case of traditional camera calibration pictures are processed and the internal parameters are therefore obtained for given object shapes and sizes. On the contrary, for camera self-calibration techniques the matrix are solved based on the geometric relationships between the multi-view images and the internal parameters for the camera are thus acquired and the external parameters are retrieved.Focusing on projective geometry, the mathematic basis for camera calibration, this paper discussed the long-established linear calibration method and analyzed the non-linear factors in this technique. The camera self-calibration technique, which is more commonly used nowadays, was also discussed and the reason for its low robustness was analyzed. A new method was developed, which applied (pure rotary movements to retrieve the initial values for the camera's internal parameters and the values were refined using the iterative processes based on the Kruppa equations. Experiments on the images obtained by CCD cameras using this method were conducted and the results were presented and discussed.