What can be done with a badly calibrated camera in ego-motion estimation?

The Research Report of CVLand CMP

Tomas Svoboda and Peter Sturm.

The whole PostScript file compressed by compress.

Abstract: This paper deals with the ego-motion estimation from two views and with the influence of the accuracy in the camera calibration on the estimation of the motion parameters. We assume a raw knowledge about the camera. We present the linear guess of the uncertainty of the motion parameters based on the uncertainty in the calibration parameters. The statistic observation will be also presented. We did many tests with synthetic data. We find the relations between noise in the camera parameters and the acceptability of the translation vector. We show that the linear guess of the translation vector uncertainty is very stable and useful even with a raw calibration. The guess of the noise in the rotation seems to be less stable and the estimation of the rotation is very sensitive to the accuracy in calibration parameters.