Comparison of camera calibration methods using virtual images and calibration environment.

摘要

For the past few decades, researchers have performed many comparison studies for different camera calibration methods. However, previous comparison studies mainly focus on back projection accuracy of a camera model and very few on accuracy of parameter estimation for the camera models. In this thesis, the comparison study is performed to observe the interaction between camera parameters and the calibration environment. The investigation focuses on Direct Linear Transformation, a popular linear method, and Tsai, a well known nonlinear camera calibration method. The effect of digitization error, resolution noise, radial lens distortion, and scene point coordinates measurement errors on Direct Linear Transformation (DLT) and Tsai methods were analyzed and compared. The calibration environment, true translation, and rotation (extrinsic) parameters, and true camera (intrinsic) parameters were completely controlled using computer simulation of image formation. Performance of the two methods is evaluated in terms of accuracy of parameter estimates for the respective camera models under various noise, distortion, and error conditions. The results of the analysis show that some camera parameters are more sensitive to noise than others. The comparison study also shows that in some cases Tsai's method is more sensitive to noise in the image plane than the DLT method. The experimental results also suggest that Tsai's method can estimate radial lens distortion reasonably well. The analysis indicates that both the DLT and Tsai's methods can still be improved. Suggestions for improvement are provided in the conclusions.