Optimized Surface Algorithm for Mining Applications Implemented in a GPU

Abstract

This paper describes the design, implementation and experimental results of an original method for real time static surface reconstruction in underground mining applications. The proposed method allows to acquire real time "stop and go" range scans in a low power consumption embedded system, which is an important requirement for long term operations. Our method uses a priori information obtained from the raw data and geometry of the scan, removes overlapped points keeping the neighbor relationship, and uses a downsampling voxel structure to obtain an isotropic surface point cloud. Preliminary results demonstrate that it is possible to achieve real time 3D spheric surface reconstruction from high density Lidar point clouds using low power parallel computing embedded systems. We were able to generate a 3D mesh from a full spherical scan with more than 98000 points in 9ms at full resolution, and perform the mesh down sampling with additional 46ms on a low power computer. These features will allow robotic systems in underground mining activities to perform an efficient surveying using low complexity parallel algorithms.