Research Of Localized Parallelization Of Large Scale Point Cloud Registration Based On PCL

With the rapid development of information technology in Ancient Architecture Reservation,engineers have used reverse engineering technology to model and protect ancient buildings.Two-dimensional images,video resources,artificial measurement,laser scanning,and so on,were used to get a full range of information of ancient buildings which need to be protected.With the scanning measurement instrument technology becomes more and more mature,the number of point cloud datum gets much larger.This paper relied on point cloud resources of the NSFC project and GPU parallelizing technique to accomplish the registration of large-scale point cloud datum and to find a registration scheme which is appropriate for parallel algorithm.Main works of this paper:(1)Use PCL`s(Point Cloud Library,PCL)own API to accomplish point cloud registration,analyze the time-consuming of the steps in the process of serial code,and whether it can be parallelized,then make a localized parallelization on features descriptor in the serialization process,then use Intel Parallel Studio(IPS)to analyze speedup and functional hotpots.(2)As for computing descriptors of Fast Point Feature Histograms(FPFH)which is easy to be paralleled but with more time-consuming,One scheme is to use OpenMP and IPS to analyze the speedup compared with the serial code.Another is to use CUDA(Compute Unified Device Architecture,CUDA)and NVIDIA Nsight performance analysis software to compare with the speedup of serialization program and analyze its utilization.(3)Discuss a parallel speedup scheme which is based on CPU and GPUheterogeneous combination to accelerate the registration.(4)Design a reasonable message type in point cloud server which contains various procedures on point cloud and develop a point cloud program with QT and PCL which contains researched parallelization registration scheme.