DTM generation from the point cloud using a progressive geodesic morphology and a modified Particle Swarm Optimization

Abstract

ABSTRACTThe digital terrain model (DTM) is an important geospatial product used in many applications of geo-information systems. Advances in light detection and ranging technology leads to generate a dense and accurate point cloud from the Earth’s surface. Because of the existence of various types of non-ground object in the complex terrain, filtering of the point cloud data and DTM interpolation is still a challenging topic. In this article, a progressive geodesic morphology has been proposed to filter the non-ground points from the point cloud. In this regard, the noises were eliminated and the point cloud was gridded and rasterized. Then, an iterative procedure was carried out based on an elevation vector constructed by considering the terrain topography. The geodesic dilation and opening operators were applied in each iteration and some pixels were detected as the non-ground points. These pixels were labelled by the connected component analysis and the generated parcels were investigated using some geometric and structural features such as the object height, the slope of the terrain, the slope of the object boundary, as well as the boundary factor in order to decide about the considered parcel entity. Afterwards, the non-ground points were removed and the DTM was interpolated using the bare Earth points. A modified Particle Swarm Optimization (PSO) has been proposed to optimize the coefficients of the polynomial interpolation for generating a DTM with high accuracy. Results showed that the error of the proposed progressive geodesic morphology was on average 3.42% using an optimized set of parameters and 4.00% using a single set of parameters, respectively. These results represented the stability of the proposed filtering method against the predefined thresholds and this is an important factor to make it a practical method. In comparing the proposed modified PSO with the other interpolation algorithms, the evaluation results indicated that the proposed modified PSO interpolated data only with 0.0884 cm error, averagely. This means the proposed modified PSO has a significant performance in generating DTM.