| In the wake of developments in science and technolo gy in the21sd century, the sensorhas made great progresses. Ways of obtaining the image information of the measured objectare various. The multiple sensors, multi-angle, large overlap measure way recently are one ofthe most important developing trends of the remote sensing technology. Due to the imageswith high spatial resolution and large overlap, the data volume is more and more large. Forthis reason, the numerous data has been a trend in both aerial images and aerospace images.Under the limitation of the storage capacity and computing power, the ordinary computercan’t meet the request of the huge data processingand highly automatic production.The graphic processor with the increasing high-performance parallel calculating andprogrammability provides wild space for some algorithm parallelization of the digitalphotogrammetry, further more, the rolling out of CUDA provides a powerful support for thedevelopment of GPU in general purpose computation field as well. This text is aimed atresearch on image pre-processing based on GPU parallel computing, image enhancement andimage matching. And this text is also designed to research GPU parallel processing whichboth makes GPU as the core component to set up GPU parallel processing platform and baseson CUDA4.0development environment, by which to improve the data processing efficiency.We picked up a number of algorithms as the object involved with image pre-processing，Wallis filtering algorithm and image matching from the Pixel Grid system to explore all ofwhose GPU parallel processing project. For another, our focus especially is research into theimplementation method and performance optimization based on GPU special systematicstructure, and by which to improve the image processing algorithm efficiency inphotogrammetryprocessingsystem. What’s new and main in this paper are as follows:1.Both the history and development trend of the Parallel Computing Platform aresummarized briefly as well as the basic mode of the digital photogrammetry image data isconcluded. The hard ware framework of GPU, the CUDA software programming model andthe optimization of CUDA program are expounded systematically. Also, the experimentplatform used by the research is showed in this paper.2.A method of image pre-processing GPU parallel processing technique was put forward,which has carried out GPU parallel processing with image rotation and distortion correction,and which realized good granularity parallel processing of re-sampling through GPU. Basicon the software programming model and GPU’s hardware framework, the task partition andexecution configuration are optimized to make full use of GPU’s high parallel computingadvantages.3.A Wallis image enhancement parallel algorithm into which the Adaptive smooth filter based on GPU acceleration is proposed, which means with the help of the powerfulcomputing ability the Wallis filtering and image adaptive smoothing algorithm isaccomplished. Along with the use of global storage, the algorithm is optimized.4.This article has put forward a parallel processing method based on CUDA’s Harrisoperator of image matching, which has completed graying images, distilling Harris angularpoint, resample and matching grey relevance, and also has improved the distribution of thread,the use of RAM, and the shared storage. This processing method has made GPU’s hugeparallel arithmetic pretty well: the experimental result indicated that this processing method ismuch more faster than CPU serial processing. |
PDF Full Text Request