Abstract: As a crucial energy resource, coal mines are characterized by complex structures, confined spaces, and insufficient illumination in their underground environments, posing significant challenges to fine-grained 3D perception and modeling. High-precision 3D reconstruction is not only a key factor in ensuring mining operation safety but also serves as an essential foundation for analyzing stratigraphic structures, constructing transparent geological systems, and monitoring dynamic environments. To address the limitations of traditional single-sensor approaches in underground scenarios—such as severe occlusion, high noise interference, and limited modeling accuracy—this paper proposes a multi-sensor fusion-based 3D modeling method. The approach involves deploying a multi-line LiDAR array on a roadway cleaning vehicle and integrating visual and millimeter-wave radar data. By employing an automatic stitching strategy that combines local feature description and parameter estimation, stable geometric and texture features from multi-source data are extracted. A two-stage fusion method is adopted, which integrates initial registration based on feature matching and fine registration optimized via probability distribution, effectively overcoming matching ambiguities caused by repetitive structures, dynamic interference, and varying lighting conditions in underground environments. This process progressively constructs a global 3D model of the underground roadway. Experimental results demonstrate that the proposed method achieves an average registration error of less than 0.12 m in coal mine roadway environments, improves stitching efficiency by approximately 47% compared to conventional methods, and reaches a target recognition accuracy of 92.7%. The findings confirm that multi-source information fusion significantly enhances the accuracy and robustness of 3D modeling in underground coal mines, providing reliable data support and technical reference for intelligent stratigraphic modeling, transparent geological construction, and the advancement of smart mining.