Research On Efficient And Dense Matching Method Based On MPI

Abstract:Efficient three-dimensional terrain and surface reconstruction of disaster site is one of the key technologies to quickly get disaster information, is also a prerequisite for orthophoto correction. Unmanned aerial vehicles and other low-altitude and light sensing system can quickly obtain high-resolution images with high degree of overlap, which attracts wide attention in disaster emergency response. The use of a wide range of low-altitude image in quick3D reconstruction in disaster site depends on the efficient dense matching method. This paper presents an efficient and dense semi-global matching method based on MPI (Message Passing Interface). Not only to overcome the traditional image matching methods are difficult to take advantage of a high degree of overlapping images, but also can quickly deal with a large number of images in a short time to get three-dimensional information on the disaster site, which can easily adapt for efficiency requirements of disaster emergency response. The main content of this paper includes:(1) Discuss and analysis the advantages and disadvantages of the common image matching method, in view of dense image matching of large amount of calculation and time-consuming problem, this paper selects the semi-global matching (SGM) method. This paper studies on the algorithms considering the low-altitude image characteristics, then optimally achieve SGM serial computing, and lay the foundation for parallel computing.(2) Analyse the parallel computing platform characteristics and application range, chooses MPI as a parallel computing platform tools. After groping research, using common and cheap computing resources autonomy to build a parallel computing platform based on MPI. The test shows that this platform has the stability and practicability. This platform for the realization of parallel computing based on MPI provides hardware conditions.(3) Aiming at the bottleneck of dense matching serial processing, this paper meticulously analyzes the process of the dense matching, and studies the parallel of dense matching. In order to realize the efficient dense matching based on MPI, on the one hand to reduce redundant computation of stereo image pairs by adaptive selection, on the other hand, through the test to analysis the process of dense matching the best choice of serial or parallel. This method can greatly shorten the calculation time of dense matching process and improve the calculation efficiency.This paper adopts the parallel computing independent platform based on MPI to obtain dense matching result. A set of aerial images and UAV images are used to perform the experiments. The results show that, this parallel computing platform can be stable and efficient complete parallel computing task, the proposed efficient dense matching method based on MPI can greatly reduce the computing time and improve the calculation efficiency, which can meet the requirements of emergency response.