Site Model Construction Using Geometric Constrained Optimization

This paper presents an optimization approach for estimating the 3D parameters of buildings from multiple images by fitting the observations to the geometric building models. Through multi-image traingulation, the coordinates of the building corners and their covariances are estimated from the located image corresponding points. Then by a constrained optimization the estimated point positions and the partial models of the buildings are integrated. Numerical methods arc used to solve the constrained optimization problem by iteratively updating an approximate solution. To provide an initial solution, the maximum likelihood estimators for the parameters of the planes and the lines are used. The covariances of the estimated parameters can be determined by propagating the input covariance matrix through the linearized optimization model. Experimental results on the RADIUS model boards are then given.