Point cloud surface reconstruction is a critical technology in lunar exploration， the quality of surface reconstruction is directly dependent on point cloud accuracy. To enhance point cloud precision， preserve surface details， and prevent over-smoothing， a surface reconstruction method based on the Euclidean distance of point cloud normal vectors and surface projection is proposed. This neighborhood-based approach employs an anisotropic projection process to closely approximate the true surface positions of objects， effectively maintaining the geometric integrity of the real object and improving the quality and reliability of the 3D model.Initially， principal component analysis （PCA） is utilized to estimate and normalize normal vectors on unstructured point clouds. Subsequently， voxel filtering is applied for preprocessing to eliminate outliers. Following this， a moving least squares （MLS） method employing anisotropic weight functions is used for point cloud projection， yielding smooth point cloud data while preserving local geometric features. Finally， the Poisson algorithm is implemented for implicit surface reconstruction and triangular mesh generation.The morphological parameters of the Indian Lava Tube are analyzed to assess the influence of algorithmic parameters on resampling and surface reconstruction. Results indicate that a projection radius of 30 mm and a normal vector difference threshold of 5° yield the highest resampling accuracy， with a point cloud root mean square error （RMSE） below 13.9 mm. Optimal surface reconstruction is achieved with a 40 mm projection radius and the same threshold， resulting in a reconstructed surface RMSE under 67.6 mm. Compared to farthest point sampling， normal space sampling， uniform sampling， and voxel sampling methods， the proposed approach attains superior surface reconstruction precision.This algorithm substantially enhances surface accuracy and is well-suited for reconstructing lava tube point clouds in natural environments. It offers a robust solution for high-precision three-dimensional modeling of lunar lava tubes.