Abstract
Abstract. In order to improve the Tsai's two-step camera calibration method, we present a camera model which accounts for major sources of lens distortion, namely: radial, decentering, and thin prism distortions. The coordinates of principle points will be calculated at the same time. In the camera calibration model, considering the errors existing both in the observation vector and the coefficient matrix, the Total Least-Squares (TLS) solution is preferred to be utilized. The Errors-In-Variables (EIV) model will be adjusted by the solution within the nonlinear Gauss-Helmert (GH) model here. At the end of the contribution, the real experiment is investigated to demonstrate the improved two-step camera calibration method proposed in this paper. The results show that using the iteratively linearized GH model to solve this proposed method, the camera calibration parameters will be more stable and accurate, and the calculation can be preceded regardless of whether the variance covariance matrices are full or diagonal.
Highlights
Because actual cameras are not perfect and sustain a variety of aberrations, the relationship between object space and image space cannot be described perfectly by a perspective transformation
The conclusions are summarized as follows: 1) Unlike the classical two-step method, which can only handle radial distortion, the improved method proposed here can synthetically establish a camera model that accounts for major sources of camera distortion, namely, radial, decentering, and thin prism distortions
The last step is to optimize all of the calibration parameters with the initial value calculated by the classical two-step calibration method
Summary
Because actual cameras are not perfect and sustain a variety of aberrations, the relationship between object space and image space cannot be described perfectly by a perspective transformation. The two-step method is suitable for most calibration problems, and the iterative convergence speed is fast, since the number of parameters to be estimated through iterations is relatively small. This method can only deal with radial distortion and cannot be extended to other types of distortion (Wenig et al, 1992). The Total Least-Squares (TLS) approach provides a solution, when all the data are affected by random errors and can solve estimation problems in the so-called EIV model (Golub and Van Loan (1980), Van Huffel and Vandewalle (1991)). We considers some of the following disadvantages of the two-step calibration method and LS adjustment, and makes some improvements in camera calibration
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
