New Approach for Automatic Detection of Buildings in Airborne Laser Scanner Data Using First Echo Only

Airborne laser scanning has become a significant 3D data acquisition technique in the field of surveying. By measuring point clouds defined in three-dimensional coordinates, this technique provides almost automatically Digital Surface Models (DSMs). But for 3D city modelling, the discrimination between terrain and elevated objects based on this surface model is still a challenging task, since fully automatic extractions are not operational. Moreover, some of the available methods combine several echoes although echo separation is not always obvious and sometimes last echo is not reliable. In this context, the aim of this study is to develop a general automatic segmentation method of Lidar point clouds focussing exclusively on the first echo and without any external data. The result of the proposed methodology is the automatic discrimination of the buildings and the terrain, by excluding vegetated areas. In the first step, terrain and off-terrain clouds are discriminated, based mainly on threshold features as proposed in the literature, but improved and generalized to the case of brutal terrain discontinuities. In the second step, buildings and vegetation are categorized as subclasses of the off-terrain class. The innovation of the exposed approach lies in the exploitation of the whole analysis levels combining points, pixel, segment and spatial information. Thus, the processing chain fully benefits from the planimetric and altimetric information of a point cloud. The complete workflow is presented, as well as its limitations. At last, the satisfying results obtained for three different test sites covered by two cloud densities validate our processing chain.