Algorithm Design of Global Point Cloud Registration Based on BSP Model

Abstract

Registrations of multi-angle point cloud involve enormous point cloud data, complicated and heavy computation. Previous non-parallel method is strong computing resources demanding and low performance of point cloud global registration. Bulk Synchronous Parallel Computing Model (BSP) processes the data with a parallel computing method, which can be applied to huge computation such as point cloud data processing. We design a point cloud global registration algorithm base on BSP model and build a Hama parallel computing cluster with average PCs. The results of four engineering examples show that the registration algorithm base on BSP model reduces computing resources requirements, accomplishes global registration with acceptable accuracy and improves the efficiency of registration. The Hama computing cluster we built is implementation simplicity and ease of point cloud processing. The improvement thought and implementation method can extend to other point clouds processing like filters, rendering, modeling and so on.