Advanced Matching Techniques for High Precision Surface and Terrain Models

Image matching is currently gaining increased interest in the photogrammetric community to provide accurate and very dense point clouds for Digital Elevelvation Models (DEMs). Besides the traditional Digital Terrain Models (DTMs) the focus is more and more on Digital Surface Models (DSMs). It can be seen as an alternative technique or even competition to airboren laserscanning (LIDAR) which is since the last decade dominating the extraction of DEMs. The increasing application of digital filmless cameras with high radiometric resolution, increased number of images and low ground sampling distances is nowadays exploited by image matching software by different vendors. This paper presents results of two investigations with the new MATCH-T DSM 5.2 version of inpho GmbH. The first investigation deals with automatic image matching in the difficult terrain of open pit mining including change detection and volume calculations using six sequential flights. The results show, that a very high level of automation can be reached by using filmless digital cameras and the features of the new MATCH-T DSM software. The DEMs are generated completely automatically without collection of morphological data. The results are accurate enough for this application as compared to the standard workflow with human operator stereo measurements. The automatic volume determination of cutting and filling shows in average only 0.2-1.0% difference to the manually derived volumes. A second investigation evaluates the software Building Generator of inpho GmbH to perform 3D building extraction using given 2D ground plans of buildings and MATCH-T DSM point clouds. Extensive empirical investigations of parameter settings in urban areas of different complexity and density show, that there are only few decisive parameter settings needed to yield promising results. In areas with simple to moderate complexity building types that are Rectange shaped, L-shaped, U-shaped or T-shaped can be reconstructed in LoD 2 (Level-of-Detail) with up to more than 95% success rate. If the extraction in LoD 2 fails, a LoD 1 building model is derived. The success rates using LIDAR point clouds can be almost reached also by using dense image matching point clouds. In case of complex buildings and high density of buildings the success rates are lower with strong dependency on a subdivision of the given 2D ground plans, which is also the case in LIDAR point clouds.