An integrated bundle adjustment approach to range camera geometric self-calibration

Abstract

This paper describes a new method for integrated range camera system self-calibration in which both traditional camera calibration parameters and rangefinder systematic error parameters are estimated simultaneously in a free-network bundle adjustment of observations to signalised targets. Its mathematical basis is collinearity and range observation equations augmented with correction models for systematic error sources identified in the data. The self-calibration results from datasets captured with two different range cameras, a SwissRanger SR3000 and a SwissRanger SR4000, are presented and analysed in detail. The method’s effectiveness is demonstrated in terms of systematic error removal and independent accuracy assessment. Up to a 54% reduction in the residual RMS was achieved by inclusion of the proposed error models in the self-calibration adjustment. An improvement of at least 74% in the RMS of object point co-ordinate differences, over that achieved without calibration or provided by the manufacturer’s software (in the case of the SR3000), was realised in an independent accuracy assessment. In addition, the effects of several influencing variables, including the range stochastic error model, the network geometry and the range measurements themselves, on key correlation mechanisms are analysed in detail.