Optimization and Simulation Research on Landscape Design of Lanzhou Waterwheel Based on Three-dimensional Modeling Technology

Abstract

This study proposes a method for the digital reconstruction of the Lanzhou waterwheel landscape based on 3D laser scanning technology. First, point cloud data of the waterwheel surface is obtained through multi-station laser scanning. Public control points are used to align and stitch the point clouds, and feature measure segmentation technology is employed to handle complex scenes. For feature extraction, the AKAZE algorithm is selected to construct a nonlinear scale space, combined with the M-LDB descriptor to enhance rotation and scale invariance. The Flann-based matching process is improved through grid-based preprocessing and group RANSAC model optimization to enhance feature matching accuracy. Based on the point cloud geometric framework, the seed point matching method is employed to achieve precise texture mapping, resulting in a high-fidelity 3D model. Performance validation shows that the model accuracy score in subjective evaluation is 94.85 points, with lighting and shadow effects exceeding 93 points. The AKAZE algorithm reduces skeleton extraction time by 12.1–18.0%. The total modeling time of the model is only 15.24 minutes, which is up to 74.7% faster than the comparison method. The average feature point distance error is 2.33 m, a reduction of more than 57.6%. The spatial stereoscopic level reached 84.12%, significantly higher than the 26.34–63.16% achieved by traditional methods.