Question about the logic behind determining covariance of priors in Gauss Newton optimizations.

[ybkurt]

I could not understand the covariance assignment of priors in Gauss-Newton optimization codes of Sparse Image Alignment part.

rpg_svo_pro_open/svo_img_align/src/sparse_img_align_base.cpp

Lines 83 to 84 in ca371f3

	H_max_diag_trans = std::max(H_max_diag_trans, std::fabs(H_(j,j)));
	Matrix3d I_trans = Matrix3d::Identity()*prior_lambda_trans_*H_max_diag_trans;

What is the logic behind this operation, i.e. looking at Hessian matrix entries to determine the information's weight? Same operation is also applied in reprojection error minimization part for rotation priors.

rpg_svo_pro_open/svo/src/pose_optimizer.cpp

Lines 321 to 322 in ca371f3

	H_max_diag = std::max(H_max_diag, std::fabs(H_(j,j)));
	I_prior_ *= H_max_diag*prior_lambda_;