Abstract: Deep goad in mines have become a significant hazard to the safety of personnel and production equipment in mines. In response to the problems of high cost,poor timeliness,and numerous measurement blind spots in existing deep goaf detection methods,a penetrating three-dimensional laser radar scanning system has been designed for goaf detection. This system uses an inexpensive mechanical rotating laser radar to reduce costs;through an independently designed graph optimizationbased Cartographer-SLAM (Simultaneous Localization and Mapping) algorithm,it can quickly process laser radar data,achieving real-time positioning and mapping underground,thereby improving timeliness. Coupled with a penetrating three-dimensional laser radar scanning system bracket,it can measure in dangerous roadways,goafs,and other areas that are difficult for personnel to access,effectively reducing measurement blind spots. To address the issues of unsmooth and hole-ridden void models constructed by existing modeling algorithms,a Poisson surface reconstruction algorithm optimized by the Moving Least Squares (MLS) method is proposed to model the void goaf cloud data. By performing high-order polynomial interpolation around the data points using MLS,and through octree segmentation,vector field calculation,Poisson equation solving,and isosurface extraction,a three-dimensional model of goaf is constructed. The experiment was conducted in a gold mine in Liaoning Province, where data from multiple goaf,roadways,and underground chambers were collected to achieve precise underground space modeling. The experimental results show that the designed system and algorithm can efficiently and accurately achieve three-dimensional modeling of deep and complex goafs,to a certain extent solving the problems of difficult access to goafs,measurement blind spots,and insufficient modeling accuracy in deep and complex goaf detection,providing important technical support for mine goaf management and safe production. 

Key words: deep goaf,Lidar,simultaneous localization and mapping,moving least squares,Poisson surface reconstruction,modeling