In open-pit mines, extracting mountain point clouds is critical for the mapping and path planning of autonomous mining trucks. Mainstream point cloud partitioning methods often misclassify features due to the irregular contours of mining environments and the blurred boundaries between mountainous and ground regions. To address the above issues, we propose a novel massif point cloud extraction method based on adaptive voxel partitioning. First, the method divides the point cloud of the single-frame into voxels based on a concentric ring sector model and adaptively adjusts the voxel composition through point cloud density distribution. A hash function is introduced to enhance the access speed to voxel units. Then, principal component analysis (PCA) is used for plane fitting within voxels, and iterative estimation and optimization of ground and mountain feature parameters are carried out. Finally, a multidimensional evaluation function is designed based on the uprightness and flatness of the point cloud for likelihood estimation, achieving mountain point cloud extraction. Data collected by autonomous mining trucks across various mining scenarios are used to construct the dataset of open-pit mines and validate the proposed method. The results demonstrate that the massif point cloud extraction method exhibits a more than 5% improvement in precision compared with existing methods and can meet real-time requirements.