无人机影像拼接并行计算技术研究

发布时间：2018-06-28 09:33

本文选题：影像拼接 + 特征提取　；参考：《北京建筑大学》2017年硕士论文

【摘要】：无人机遥感作为自然灾害(如地震救援等)现场数据实时获取与应急指挥和快速反应的重要技术,不仅具有成本低、操作简单、机动灵活等特点,而且其获取的图像还具有数据量多、分辨率高和实时性强等优点。同时无人机遥感影像又具有幅面小、像幅多、多重叠等问题,在应急指挥和快速反应的实时影像获取中,如何实现快速、高效的无人机序列图像拼接,就显得尤为重要。在处理图像拼接时,在影像配准这一过程中,不可避免的会产生变形。在处理无人机序列图像拼接时,后面的影像依赖于前面已经配准过的拼接影像,这样就会产生拼接误差逐步积累的问题。因此,在完成拼接这一过程中时,不仅要保证拼接实施过程的效率,还要保证拼接结果的质量。这样才能继续发挥无人机遥感的优势,拓宽无人机序列拼接影像的应用范围、提高无人机序列拼接影像使用价值。随着多处理器计算机的普及以及并行化编程模型的不断优化,多核架构并行化编程已经成为不可阻挡的发展趋势。它不仅能够成功解决复杂冗余的大型计算问题,而且它的运行速度极快。为了进一步解放人力、物力,我们需要充分发挥多核计算机的使用性能,从而去提高无人机序列拼接影像的质量和效率。本论文首先对无人机影像进行预处理操作,主要是进行拉普拉斯锐化,增强影像边缘、采用Wallis滤波对图像均匀分块,根据块内灰度信息自适应匀光匀色、按照重叠度截取感兴趣区。接着提出了一种基于规则格网的Harris特征提取,然后利用马氏距离提取同名点,同时加入核线、唯一性多级约束条件实现由粗到精的匹配过程,最后利用RANSAC算法迭代求出最优单应矩阵,将影像转换到同一参考系统下,利用加权平均值算法,实现融合拼接。在这一基础上,本文综合考虑了影响无人机遥感影像拼接质量的来源,将多航带无人机影像采用多级分组分块进行拼接,对不同组块的拼接误差进行控制;同时引入多核并行编程技术,即保证影像拼接质量,又提高影像拼接效率。为了提高多核计算机运行加速比,本文提出了用动态创建线程和创建线程池来降低创建线程带来的开销的策略。通过实验对比发现,与常用特征提取与匹配算法比较,本文提出的基于规则格网的Harris特征提取算子提取的特征点分布均匀,有利于全局匹配、提高影像配准与拼接质量;本文提出的无人机序列影像的多级分组分块拼接与多核并行编程技术方法,即保证影像拼接质量,又提高影像拼接效率。有效控制了拼接误差传播的积累,同时大大减少了程序运行时间,一定程度上验证了本文提出的基于规则格网的Harris特征提取及多级分组分块并行化编程算法的可行性和有效性。
[Abstract]:UAV remote sensing is an important technology for real-time acquisition of field data of natural disasters (such as earthquake rescue), emergency command and rapid response. It not only has the characteristics of low cost, simple operation, flexibility and so on. Moreover, the obtained images also have the advantages of large amount of data, high resolution and real-time. At the same time, UAV remote sensing images have many problems, such as small format, multi-amplitude, multi-overlapping and so on. In emergency command and real-time image acquisition of rapid response, how to achieve rapid and efficient UAV sequence image mosaic, it is particularly important. In the process of image stitching, deformation will inevitably occur in the process of image registration. In the processing of UAV sequence image stitching, the back image depends on the pre-registered stitching image, which will lead to the problem of gradual accumulation of stitching errors. Therefore, in the process of stitching, we should not only ensure the efficiency of the mosaic implementation process, but also ensure the quality of the stitching results. In this way, the advantages of UAV remote sensing can continue to be brought into play, the application scope of UAV sequence mosaic image can be widened, and the use value of UAV sequence mosaic image can be improved. With the popularization of multiprocessor computer and the optimization of parallel programming model, parallel programming of multi-core architecture has become an irresistible trend. It not only can successfully solve the large-scale computing problem with complex redundancy, but also can run very fast. In order to further liberate manpower and material resources, we need to give full play to the performance of multi-core computer, so as to improve the quality and efficiency of UAV sequence stitching image. In this paper, firstly, the UAV image is preprocessed by Laplacian sharpening and image edge enhancement. Wallis filter is used to divide the image evenly into blocks, and the image is evenly colored adaptively according to the gray level information in the block. Intercept the area of interest according to overlap. Then, a Harris feature extraction method based on regular grid is proposed, and then the point of the same name is extracted by Markov distance, and the kernel line is added at the same time, and the unique multilevel constraint condition is used to realize the matching process from coarse to fine. Finally, the optimal homography matrix is iterated out by using the RANSAC algorithm, and the image is converted to the same reference system, and the weighted average value algorithm is used to realize fusion and stitching. On this basis, this paper synthetically considers the source of affecting the quality of UAV remote sensing image splicing, the multi-band UAV image is spliced by multi-level grouping, and the splicing error of different blocks is controlled. At the same time, multi-core parallel programming technology is introduced to ensure the quality of image stitching and improve the efficiency of image stitching. In order to improve the speedup ratio of multi-core computers, this paper proposes a strategy of dynamically creating threads and creating thread pools to reduce the overhead of creating threads. Compared with the common feature extraction and matching algorithms, the feature points extracted by Harris feature extraction operator based on rule grid in this paper are evenly distributed, which is conducive to global matching and improve the quality of image registration and stitching. In this paper, the multi-level grouping block splicing and multi-core parallel programming method for UAV sequence images is proposed, which ensures the quality of image mosaic and improves the efficiency of image stitching. The accumulation of splicing error propagation is effectively controlled, and the running time of the program is greatly reduced. To some extent, the feasibility and effectiveness of the proposed Harris feature extraction algorithm based on rule-based grid and multilevel block parallel programming algorithm are verified.
【学位授予单位】：北京建筑大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】