Research And Development Of The UAV Photogrammetry System's MPI Clustering Development

In recent years,with the development of UAV photogrammetry platform in the field of photogrammetry and remote sensing,digital image high resolution,large overlap,resulting in increasing the amount of image data,mass data processing has become a trend.Ordinary computers are limited by their storage and computing capabilities,in the face of massive data processing has been inadequate,the use of CPU serial processing mode of production has been difficult to meet the efficient production of massive data needs.In this paper,aiming at the urgent requirement of mass photogrammetric imaging data,the MPI message delivery library is used CPU as the core component,and the parallel computing technology of the photogrammetric data processing algorithm is studied based on the Windows development environment.The main work and innovation of the paper are as follows:1.This paper briefly summarizes the current situation and development trend of high performance parallel computing technology in MPI.Taking the typical image matching algorithm in UAV photogrammetry as an example,this paper discusses the parallelization of MPI and the basic model of aerial image parallelization.Based on the deep study of MPI and image matching,a basic flow of MPI parallel computing in image matching is proposed.Two MPI image matching parallelization algorithms are designed and implemented,namely,the average distribution image mentioned in the article Matching parallel algorithm and balanced load image matching parallel algorithm,and the problem of the trial stage to describe and analyze and solve.2.Based on the above two parallel algorithms,this paper focuses on the load balancing and data transmission problems that affect the parallel efficiency of MPI in image matching and parallel computing.Aiming at these problems,a parallel algorithm of image matching is proposed.In the extraction feature stage,the initial task allocation corresponding to each node is completed according to the image space adjacency relation,and then the final extraction feature task is determined by the secondary partitioning on the basis of the initial division.In the task scheduling,according to the status of the calculation node priority allocation of the corresponding node tasks,the corresponding task of the corresponding node is assigned to the other nodes assigned to the corresponding task.In the matching phase,the matching task is divided according to the extracted feature task,and then the matching task unit corresponding to each node is delineated according to the node number of the feature extraction task.The same task scheduling method is used to complete the image matching of the whole survey area.3.The image matching system based on MPI is designed and implemented.The system satisfies the load balancing and satisfies the requirement of minimizing the transmission of data.Through the study of this paper,we can conclude that the image matching parallel algorithm proposed in this paper can greatly improve the efficiency of image matching.