Abstract: Aiming at the problems existing in current bench line extraction methods based on open-pit mine point cloud data, such as single judgment basis, poor spatial adaptability, high false detection rate, and low levels of automation and extraction efficiency, a bench line extraction method for open-pit mines based on the dual RANSAC optimized shape detection algorithm is proposed. This method achieves efficient, accurate, and automatic extraction of bench lines from 3D point cloud data of open-pit mines. To reduce the point cloud data volume and computational complexity of single processing, as well as improve parameter control accuracy, algorithm time efficiency, and data adaptability, the large-scale point cloud of open-pit mines is divided into several "point cloud units" with smaller data volumes. On this basis, a DisR-Kt-RANSAC optimized normal vector calculation method is proposed according to the topographic characteristics of open-pit mines, so as to more accurately calculate the normal vectors of points at irregular topographic locations such as bench edges, protrusions, and depressions. Then, further based on the RANSAC algorithm combined with mechanisms such as octree hierarchical weight dynamic update, the target point cloud data is decomposed into point sets of geometric shapes such as flat plates, ramps, slopes, and curved surfaces according to the topographic characteristics of open-pit mines. Based on the downsampling strategy of Delaunay triangulation, internal downsampling and edge point optimization are performed on the extracted shape point sets, reducing data volume while maintaining a stable density of edge points in the shape point sets, which ensures that the improved Alpha Shapes edge detection algorithm can efficiently extract a sufficient number of uniformly distributed edge points. Finally, a spatial connection method based on the k-NN graph and Kruskal's algorithm to generate the minimum spanning tree (MST) was employed, achieving efficient step line construction while effectively avoiding issues such as breakpoints, closed loops, and overlapping multiple lines. Experimental results on step line extraction from point cloud data generated by onboard LiDAR and aerial photography in open-pit mines demonstrate that the proposed step line extraction method based on dual RANSAC-optimized shape detection achieves a completeness of 96.01% and an accuracy of 99.04%. The mean absolute error of the extracted shape point sets is less than 0.23 m, with a standard deviation below 0.22 m, indicating high precision. Additionally, this method exhibits significant temporal efficiency advantages when processing large-scale open-pit mine point cloud datasets exceeding 6×10⁷points.