A Point Cloud Filtering Approach to Generating DTMs for Steep Mountainous Areas and Adjacent Residential Areas

Digital terrain models (DTMs) are considered important basic geographic data. They are widely used in the fields of cartography, land utilization, urban planning, communications, and remote sensing. Digital photogrammetry mainly based on stereo image matching is a frequently applied technique to generate DTMs. Generally, the process of ground filtering should be applied to the point cloud derived from image matching to separate terrain and off-terrain points before DTM generation. However, many of the existing filtering methods perform unsatisfactorily for steep mountainous areas, particularly when residential neighborhoods exist in the proximity of the test areas. In this study, an improved automated filtering method based on progressive TIN (triangulated irregular networks) densification (PTD) is proposed to generate DTMs for steep mountainous areas and adjacent residential areas. Our main improvement on the classic method is the acquisition of seed points with better distribution and reliability to enhance its adaptability to different types of terrain. A rule-based method for detecting ridge points is first applied. The detected points are used as additional seed points. Subsequently, a locally optimized seed point selection method based on confidence interval estimation theory is applied to remove the erroneous points. The experiments on two sets of stereo-matched point clouds indicate that the proposed method performs well for both residential and mountainous areas. The total accuracy values in the form of root-mean-square errors of the generated DTMs by the proposed method are 0.963 and 1.007 m; respectively; which are better than the 1.286 and 1.309 m achieved by the classic PTD method.