A Posteriori Weight Estimation in Self-Calibrating Bundle Block Adjustment

A rigorous mathematical model of self-calibrating bundle adjustment consists of three Kinds of observations,i.e.observations of image coordinates,of ground coordinates of control points,and of additional parameters.It is only when their weights are correctly chosen that a stable and reliable result of the self-calibrating adjustment can be ensured and hence,the image systematic errors can be compensated efficiently.In this paper the weight estimation of these three kinds of observations is disscussed by using a posteriori variances estimation in least square adjustment,and an analysis of the experimental results is carried out.The main conclussions are(1) A posteriori variance estimation in self-calibrating bundle adjustment can always have a unique,convergent solution.(2) In any case the estimator of the additional parameters variance is comparatively exact.But when choosing the optimum weight of additional parameters we should properly consider the effect of systematic accumulation of accidental errors.(3) For the case of a sparse ground control,the weight of ground control point observations can not be estimated exactly,because the redundancy of ground point observations is very small.But it will not affect the adjusted results practically.Only for the case of dense ground control,It is neccessary and possible to estimate the weight of ground point observations more exactly.