Non-contact distance measurement is one of the problems in computer vision. Our main focus of attention was distance measuring with a reference paper. Known size of an A4 paper enable us to calculate the homography matrix of the mapping from the image plane to the real world coordinate system. Frequently, the camera parameters are inaccurately estimated and this inaccuracy consequently contributes to the overall error in the measured distance. We decided to model this error rigorously. First we derived the error analytically. Then we looked into each parameter individually and discussed its contribution to the overall error to distance measuring and the contribution to the error in camera pose estimation. After the theoretical analysis we compared theoretical estimations with empirical data. Empirical data was gathered from several smartphones, which we calibrated using a known calibration pattern. We found out, that our model correctly estimates the location of the calibration pattern mapped to the real world coordinate system. Estimated camera pose does not accurately match the reference pose. The orientation of the camera is very close both to the empirical and the reference orientation.
|
