Abstract: The advancement of intelligent mining continues unabated, with three-dimensional laser scanning technology laying the foundation for the perception of mines. A robust two-stage denoising method for complex underground tunnel scenes in three-dimensional point cloud data has been proposed to acquire high-quality mine 3D point cloud data. First, three-dimensional laser scanning technology is used to acquire 3D point cloud information of underground tunnel scenes and analyze the different noise characteristics within the tunnel point cloud. Secondly, a two-stage tunnel denoising model is designed. In the first stage, the impact of the angle relationship between the point cloud normals and the tunnel axis is investigated and removed by calculating the normals and constructing the tunnel axis. In the second stage, the optimization of the tunnel point cloud is carried out. By analyzing the distance variations between the tunnel's point cloud pending optimization and the noise point cloud, points in close proximity are integrated back into the denoised point cloud from the first stage. This results in a complete point cloud post-denoising. The method is implemented through secondary development based on the open-source software Cloud Compare. Finally, using a specific case of the main haulage drift and return airway in an underground mine, the processing time for each stage of point cloud denoising, the impact of angle thresholds on denoising results, comparisons between different denoising quantities, point cloud normal calculations, and key parameters for point selection in the optimization process are analyzed in detail. The effectiveness of the method is systematically validated.The research results show that the proposed denoising method for underground tunnel scenes in three-dimensional point clouds effectively removes noise. When the angle threshold is less than 1°, the optimal denoising effect can be achieved. Through the two-stage optimization algorithm, effective repair of surface holes on the tunnel is achieved. This study provides strong guidance for the practical application of denoising of underground tunnel point clouds, demonstrating its potential in improving data quality and reliability. It holds significant practical application significance in the fields of mining intelligence, geological exploration, and safety monitoring.