Research On Fast Mosaic And Points Cloud Building Based On UAV Images In A Cluster Environment

As an important complement of traditional photogrammetry methods, UAV remote sensinghas been widely applied in many fields. Epecially in the timesensitive area, the UAV remotesensing method takes a non-substitutable role. Cluster parallel computing technique is applied tofast processing of UAV images in this paper. Mosaicing of UAV images and building ofthree-dimensional points cloud are researched by the numbers. The topic selecting bearstheoretical meaning and practical value. The main work done in the thesis is listed as follows:1. The concept and feature of cluster is illuminated. The development status is summarized.And the feature of two typical digital photogrammetric systems based on cluster has beendescribed in detail. Based on these, a parallel computing platform based on HP blade server isbuilt.2. Pointing at the large amount of tie points abstracting work, the method of tie pointsabstracting in the overlapped area of images is researched. A multicore parallel processingmethod is given, and the processing rate has been remarkably improved.3. Based on in-depth researching on matching algorithm based on SIFT features, a parallelprocessing method in master-slave mode for sequential UAV images’ matching based on SIFTfeatures is proposed. A striped partitioning strategy based on delamination is proposed. Toreduce the cost of communication，the technique that overlaps data communication andcomputing is used. The experimental results show that the method has good scalability. However,the master-slave mode is still taking serial processing in the feature matching tache. A matchingparallel algorithm in peer-to-peer mode is designed to improve it. Through reasonable taskdistribution and execution scheduling, parallel processing of the whole matching flow isachieved. The experimental results show that the algorithm can achieve nearly linear speedupeffect.4. The method to solve the problem of the radiometric consistency in the mosaicing ofimages is researched. A parallel processing method of images composing is designed, which isable to improve the processing efficiency and compose the result image with smooth grey leveland balanced hue.5. A three-dimensional points cloud building scheme based on UAV images SIFT featuresmatching is proposed. In the selected reference coordinate system, the first model is obtained byrelative orientation and forward intersection, then the geometric model is being expanding viathe “resection-forward intersection-bundle adjustment without GCP” method until the wholearea’s points cloud is acquired. The experimental results show that this scheme could provide reference for further research of acquiring the three-dimensional geometric information of thetarget with control points support and prove the feasibility.