Research And Implementation On Key Technologies Of3D Reconstruction Of Clouds Surface Based On Stereo Images

Relative to the increasingly scarcity of surface and groundwater resources, the atmo-sphere is extremely rich in water resources and has great potential for development and utilization. Measuring and modeling the cloud, getting the basic geometric features and building three-dimensional geometric model of the cloud surface can provide the foundation for cloud meteorological parameters scientific visualization and analysis of atmospheric dy-namics within the cloud. Due to the complex morphology and weak texture of cloud image, existing studies of cloud modeling is few. And these algorithms are complex, the simulat-ing effect is not real too. This paper studies the photography-based three-dimensional cloud surface reconstruction technique. The main work is as follows:For the high cost of existing cloud observation equipment and difficult deployment prob-lem, we design a low-cost and portable cloud measurement system. The system consists of several parts:framing, measurement, support and synchronization. Cloud base height mea-surement and cloud synchronized shooting can be easily done via this system.Considering the cloud have different forms and the cloud image texture is weak, this article performs practical application and comparative analysis on cloud images by using a variety of feature extraction and matching algorithms. One is the comparison of processing time, number of matching points of a variety of algorithms on a cloud image. The second is adjusting the parameters and comparing the results, SIFT algorithm as example. The third is comparison of the extent of coverage of match points for each algorithm results. According to experimental results, this paper argues that the integrated use of a variety of algorithm which have minimum coverage with each other, can significantly increase the number of feature points.Since the less efficient of image feature extraction and matching process, this paper im-proves the incremental reconstruction method based on image sequence. One is using a simplified image selection algorithm, only do feature matching on the adjacent images to reduce workload. Second, improve fundamental matrix calculation method to ensure the success rate. Third, analyze and determine the error threshold to increase the number of point cloud. By comparing the experimental results, it proves that the running time of our algorithm is superior to photosynth, visualSFM software such as the number and visual ef-fects of the point cloud.For the imprecise problem of weak texture stereo matching, this paper presents an image segmentation and pixel gradient based gradient algorithm. We use graph cut based super-pixel segmentation method to segment the image and construct a function based on pixel gradient and regional disparity. The method can extract more refined disparity map, so that a dense cloud of points and accuracy improved.Finally, the paper processes the scattered point cloud and grid reconstruction. We use the local geometry information for noise filtering, down-sample the data and use Delaunay triangulation and region growing approach to complete the surface grid.This study is the foundation of cloud scene simulation and scientific visualization of meteorological parameters. The part about reconstruction of the cloud image has a certain reference value for stereo matching of weak texture images.