Western China is the world’s major coal-producing area, accounting for 1/3 of global coal production. Large-scale underground coal mining has led to serious ground deformation that has unique spatial and temporal characteristics. However, there are obvious engineering limitations to using GNSS, InSAR, and other common measurement techniques for monitoring mining subsidence. In recent years, researchers have attempted to use UAV LiDAR to monitor subsidence in mining areas. Point clouds are obtained through periodic ground scanning. Surface subsidence information is extracted through the filtering, modeling, and superposition processes. Conventional point cloud processing usually produces subsidence models that are too noisy and only provide information on surface subsidence. This means that it is unable to capture the horizontal ground displacement accompanying subsidence in mining areas, which hinders the practical application of this technology. A local flat point extraction (LFPE) algorithm based on geomorphic features is proposed. The algorithm extracts flat surface point clouds from the scan data and calculates the precise subsidence value of the ground by superposition to facilitate deformation monitoring. The obtained subsidence volume is spatially interpolated based on the unique spatial distribution characteristics of the mining subsidence basin to generate the initial surface subsidence model. By denoising this model, we can obtain a fine surface subsidence model that avoids significant noise caused by non-ground point clouds participating in modeling and superposition in common processing methods. On this basis, we successfully extracted the horizontal displacement information accompanying subsidence by employing sub-pixel correlation using the surface feature images before and after subsidence. The proposed technical pathway was validated by applying it to a coal mining subsidence area in the Yushen mining area in western China. Results demonstrate that, compared to conventional treatments, the proposed subsidence modeling method using laser scanning point clouds improves monitoring accuracy by over 50% while also obtaining the necessary horizontal displacement information for engineering requirements. These findings confirm the efficiency and accuracy of the proposed method for acquiring 3D surface deformation in mining areas.