An optimized modeling and visibility analysis approach for natural landscape scenes via Structure-from-Motion and enhanced Dyntacs algorithm

The coupling of various evolutionary factors contributes to the rich and heterogeneous topology and texture features of natural landscapes. However, existing semi-automatic methods struggle to accurately capture the designer’s intent and allow for detailed modeling of large-scale natural landscapes. To address this, this paper proposes a landscape modeling method based on 3D point cloud technology, utilizing Structure-from-Motion (SfM) technology for data collection. The method enhances point cloud data texture through an adaptive partial adjustment strategy. Additionally, we investigate visibility analysis using digital elevation models (DEMs) and propose an improved Dyntacs visibility algorithm. This algorithm reduces inter-visibility computation redundancy through effective reuse strategies and parallel processing, thereby improving computational efficiency. The results indicate that the proposed method achieves a maximum error of only 0.40 mm and significantly reduces processing time compared to existing algorithms. The overall accuracy (OA) of our method is 95.2%, with a mean Intersection over Union (mIoU) of 78.2%. These metrics demonstrate that the proposed model offers significant improvements in performance, effectively enhancing feature extraction and segmentation of point cloud data.