Covariance Computation

Optimal combination of calibration parameters requires the estimation of covariances from the combined effects of previous fits. This process is supported by the following functions.

• init covar. This routine removes the current covariance estimate and replaces it with a "weak" estimate, too small to constrain subsequent fitting except in those circumstances where there is no constraint on the parameters (see EPI min). This process is sufficient to constrain the minimisation solution to one which is close to the initial estimate without unduly affecting the quality of the resulting fit residuals.

• full covar. This routine calculates the stereo camera inverse covariance matrix for the current set of Model paramters around the current estimate. This inverse is then used in all subsequent calibrations until it is either updated or cleared. The integrity of the calculated covariance is checked and the diagonal elements of the product with the inverse is prited in the main tool dialog window. All elements should be close to 1.0 or 0.0. Faliure can sometimes be rectified by modification of the "condition" parameter. The magnitude of the calculated covariance is scaled by the "accuracy" parameter which should be set to the expected feature location accuracy (estimated from the residual plots epi-cerr). This allows combination of data from different feature sources which may have differing intrinsic accuracies.

• epi covar. This routine calculates the stereo camera inverse covariance matrix for the current set of Model paramters around the curent estimate but only for contributions to the model constraint from off epi-polar errors. This routine should thus be used to update the covariance matrix only after minimisation with EPI min.