The Research And Implement On Camera Calibration Technology Based On Trifocal Tensor

In recent years, computer technology and digital products are developing rapidly, the research on computer vision is to catch people more and more attention.As one of the important research directions of the computer vision, camera calibration is the basic and essential step for two-dimensional digital images restore and rebuild the characteristics of three-dimensional space objects.The existing camera calibration methods include traditional method, based on active vision method and the camera self-calibration method.The traditional method requires the reference calibration block whose shape and geometric information is known, it is difficult to apply in practice; the method based on active vision requires the camera to do some particular movement to implement the calibration, it needs relatively high accuracy of the experimental equipment; camera self-calibration method can work when there is no reference calibration block or the movement of camera is arbitrary,but this method always has poor robustness and the precision of parameters obtaind are generally not high.Therefore,the paper studies the estimation algorithm of trifocal tensor and proposes a camera calibration method based on trifocal tensor. Specific work done as follows:The paper first briefly introduces the basic theory of computer vision, which focuses on the camera model and the relations in multi-view geometric.Camera imaging geometry model includes linear and nonlinear models, linear model is relatively simple, nonlinear model can better simulates and compensates the complicated aberrations of the camera, but often has high computational complexity, and it is relatively difficult to realize.Secondly, the paper does the research about the estimation algorithm of fundamental matrix, the fundamental matrix can represent the projective geometry relationship which is independent of the sence between the images of two viewpoints. Matching algorithm based on gray correlation can only obtain the initial matching points, the RANSAC algorithm can not only strike the fundamental matrix, but also extract the accurate corresponding feature points.Thirdly, the estimation algorithm of trifocal tensor is studied in detail, at the same time,the paper proposes an estimation algorithm of trifocal tensor based on the combination of genetic algorithm and LM algorithm.The general algorithms include the direct linear algorithm, the iterative algorithm and the RANSAC algorithm. The direct linear algorithm is relatively simple, but difficult to obtain the most optimum solution, the iterative algorithm has high complexity and depends on the selection of initial value, RANSAC algorithm has a higher robustness, but the threshold is not easy to determine. The algorithm proposed combines the global search advantage of genetic algorithm and the properties that the decreased iteration speed of LM algorithm is fast,experimental results are more precise.Finally, the camera calibration method based on the trifocal tensor is proposed,this calibration method can be classified into the scope of camera self-calibration method based on absolute dual quadric,compared with general self-calibration method, the method based on the trifocal tensor can obtain the three camera projection matrix in the same reference coordinate system, more importantly, the nonlinear equations are linearized for iterative optimization of the initial results,so the camera obtained are more accurate. The experimental results show that the proposed self-calibration method is correct and effective.