Abstrakt
A method for determining a roof coverage type and a building height from a sparse laser scanning point cloud was introduced in Hofman (2008). This model driven approach utilizes 2D building outlines from the Digital Cadastral Map (DCM) and orthoimages in addition to an airborne laser scanning point cloud.
Its results were unsatisfactory, since the determination of roof types was not reliable. Thus, a practical application of this method was not possible.
While searching possibilities for its improvement, it was discovered that derivation of roof edges from orthoimages was the weakest point in the workflow. A new model driven method is presented, which is suitable for buildings with a rectangular plot.
Based on four predefined roof types, a subset of a point cloud is divided into several groups corresponding to roof planes. The best fitting planes are found by means of least squares adjustment with an iterative exclusion of outliers.
The most probable roof type is selected by an evaluation of a number of points excluded from the calculation. Results of this new approach were applied on datasets from three test sites (Brno, Sobotka and Pardubice- Polabiny) which are presented.
In spite of a very low density of laser points (1.5 and 0.25 points per m2) the method reveals very good results. The success rate of correctly determined roof cover types is 91% and 80% for the point cloud density of 1.5 and 0.25 points per m2, respectively.