Research On SIFT Feature Rapid Matching Technology Of Remote Sensing Image

Sharply increase in Remote Sensing data and a variety of application requirements put forward severe challenges on the way of photogrammetry and Remote Sensing data processing.Based on CUDA developing tools,this thesis makes a comprehensive study on the remote sensing image SIFT feature matching CPU-GPU collaborative parallel processing technology.The main work and innovation of this paper is as follows:1.The research background of this thesis and status are briefly analyzed and summarized.And the GPU's parallel processing hardware framework,software programming model,optimization principle and basic strategies are discussed in detail.Moreover,software and hardware experimental platforms are given.2.After discussing the basic theory of SIFT feature matching in detail and analyzing its parallelism,a CPU-GPU collaborative parallel speed up in Heterogeneous parallel processing method was put forward.By Using reasonable thread groups to ensure the correctness of data loading,and using the partitioning of matrix to hide the access latency of Global Memory,finally Optimized the feature point matching.3.Proposed the large-format stereo pair SIFT feature matching strategy.The strategy firstly estimated the overlap regions,and made the overlap regions into blocks.Then conducted SIFT feature matching in the search images with the corresponding block.This method effectively solved the problem in the failure of large-format SIFT feature matching because of limited storage space.4.Analysis the reasons for the failure of inclined image SIFT feature matching,a method to calculate inclination the on the basis of exterior orientation elements is proposed after the exposition of the basic principle of ASIFT operator,and verified ASIFT operator affine deformation by the experiment.