Camera model selection in high-accuracy 3d vision metrology

Camera model plays a very important part in 3D vision metrology, and it is the main factor determining the accuracy of a camera calibration process. Almost all previous efforts for camera calibration, however, calibrated a fixed, pre-specified camera model. It can be said that the camera model is always selected depending on the engineer's personal experience currently. Is the model been used over or under parameterization in a certain metrology? We don't know! Aiming at answering this question, a camera model selection approach is proposed in this paper, which is simple to implement. In our method, camera model selection is composed of two parts: term selection and order selection. Term selection is fixed, and is the same for all cameras, while order selection can be accomplished according to the measurement noise directly. Both of term and order selections are based on the property of the camera model. Successful experiments are carried out to evaluate the performance of the proposed approach.