无人机遥感图像拼接关键技术研究

发布时间：2018-03-14 11:34

本文选题：无人机　切入点：遥感图像　出处：《东北农业大学》2016年博士论文　论文类型：学位论文

【摘要】：无人机遥感是一种新的遥感手段,因其具有高效、灵活、快速、低成本及高分辨率等特点,近年来呈现良好的发展势头。无人机遥感在很多领域都取得了较为广泛的应用,包括农业、林业、电力、国土资源、城市规划等。尤其在农业方面,无人机广泛应用于农情监测、病虫害监测、作物养分分析、作物长势监测等方面,可以帮助农业生产人员及时、准确地获取作物生长信息,可为减灾降灾和应急指挥提供科学依据,现已取得了较为理想的应用效果,具有广阔的应用前景。但是,由于无人机遥感平台在航拍过程中,受到飞行高度和相机焦距等的限制,所获取的单幅图像覆盖范围较小,往往无法覆盖整个所需区域,因此,为获取整个目标区域信息,就需将获取的多张遥感图像拼接融合成一幅全景图像。无人机遥感图像拼接是后续进行遥感解译与分析的基础,拼接的质量与效果直接决定着解译与分析的准确性,因此,开展无人机遥感图像拼接技术研究具有重要的现实意义。本文对国内外遥感图像拼接技术进行了深入研究与学习,比较和分析了现有的遥感图像拼接技术的特点,基于遥感图像拼接流程,重点开展了无人机遥感图像预处理技术、遥感图像配准技术和遥感图像融合技术的研究,主要研究工作如下:(1)遥感图像预处理。遥感图像预处理是遥感图像拼接非常重要的一步,直接决定着图像拼接的精度。无人机在两次航拍间隔,由于光照强度和入射角变化,相邻两幅影像可能会产生较大色差,表现为图像亮度、饱和度不一致等,因此,必须对图像进行辐射校正,以消除图像色差对拼接的影响。鉴于无人机在航拍过程中,相邻的两幅图像拍摄时间很短,受大气和光照等条件的影响差异相对较小,因此,本文采用直方图匹配法进行相对辐射校正,通过调整参考图像的直方图使之与待拼接图像直方图匹配,从校正结果看,校正后的图像与待拼接图像在亮度和色调上基本趋于一致,很好地实现了两幅图像的平滑过渡。此外,由于无人机体积小,重量轻,受气流影响较大,稳定性和抗风能力较差,飞行姿态倾斜、抖动现象难以避免,这些都会对获取的遥感图像产生直接的影响,导致图像发生畸变,本文通过像点坐标变换、图像重采样等对畸变图像进行几何校正,以消除几何畸变对图像的影响,满足图像配准的需要。几何校正具有计算量大、耗时长的特点,为了提高几何校正的处理速度,针对无人机遥感图像,本文提出了一种基于分布式处理的并行几何校正算法。该算法使得处理器计算所需的数据均存储在本地,避免了处理器之间通信,很好地解决了数据的局部性问题,提高了并行处理效率,在机群系统中,对并行几何校正和串行几何校正算法进行了测试、比较与分析,实验结果表明,并行几何校正算法具有较好的并行性能,大大提高了遥感图像的几何校正的处理速度。(2)遥感图像配准技术研究。图像配准是图像拼接技术体系的核心,图像配准的精度直接决定着图像拼接的质量。由于SIFT算法对旋转、平移、缩放、亮度变化保持不变性,对视角变化、仿射变换、噪声也保持一定程度的稳定性,因此,该算法非常适于无人机遥感图像配准,本文基于SIFT算法进行特征点提取。特征点匹配过程中,由于相邻图像的自相似性,经常会出现一对多或多对一的错误匹配,必须加以剔除。本文提出了采用对极几何约束和单应性约束的方法剔除错误匹配特征点,首先通过对极几何约束缩小搜索范围消除一部分错误特征点,然后基于获得的正确的匹配点对建立单应性变换矩阵模型,进一步消除图像中的错误特征匹配,提高匹配精度,并采用RANSAC算法剔除外点,对匹配特征点进行提纯,实现了图像特征点的精匹配。(3)遥感图像融合技术研究。图像融合的目的就是将配准后的图像基于一定的变换模型合并为一幅图像,两幅图像应做到无缝拼接,不应该出现拼接痕迹,过渡自然。无人机获取的遥感图像,由于受到光照变化和传感器入射角度变化影响,相邻图像间会存在亮度与色度差异,图像若直接用于拼接,则在两幅图像的重叠区域会形成明显的拼接痕迹,出现“鬼影”现象。针对这一问题,本文提出了采用最佳缝合线的方法来消除“鬼影”,首先采用基于动态规划的最佳缝合线搜索,但是采用该搜索策略,若某一条缝合线存在一些较大的误差点,将导致“鬼影”无法彻底消除,得到的缝合线是局部最优化缝合线,不能获取全局最优缝合线。为此,提出了基于图切割的最佳缝合线搜索,缝合线搜索中,基于最大流最小割定理获取有向图的最大流,即最佳缝合线,两幅图像的拼接位置定位比较精确,接缝两侧的错位和“鬼影”现象得到较好解决,真正实现了无缝拼接。同时针对待拼接图像可能因曝光原因等存在的亮度与色度差异,采用泊松融合方法实现了图像的平滑过渡。
[Abstract]:UAV remote sensing is a new method, because of its high efficiency, flexible, fast, low cost and high resolution, in recent years, showing a good momentum of development. The UAV Remote Sensing in many fields has been widely used, including agriculture, forestry, electricity, land resources, city planning and so on. Especially in agriculture, drones are widely used in agricultural monitoring, pest monitoring, crop nutrient analysis, crop growth monitoring, can help agricultural production personnel in a timely and accurate crop growth information, can provide a scientific basis for reducing the disaster and emergency command, has achieved the ideal application effect, with application broad prospects. However, because of the UAV remote sensing platform in the process of aerial photography, by the flight height and the focal length of camera and other restrictions, the coverage of small single image acquired, often can not cover the whole The required area, therefore, in order to obtain the target region information fusion requires multiple remote sensing image mosaic will get into a panoramic image. The UAV remote sensing image mosaic is the basis of remote sensing interpretation and analysis of the quality and effect of splicing directly determines the accuracy of interpretation and analysis, therefore, to carry out Study on remote sensing image mosaic technology machine has important practical significance. In this paper, the domestic and foreign remote sensing image mosaic technology has conducted in-depth research and study, comparison and analysis of the characteristics of remote sensing image mosaic technology of the existing process of remote sensing image mosaic based on focus on the UAV remote sensing image preprocessing technology, remote sensing image registration and fusion technology the technology of remote sensing image research, the main research work is as follows: (1) remote sensing image preprocessing. Image preprocessing is a very important step in remote sensing image mosaic, direct decision The accuracy of image mosaic. Two aerial drones in the interval, due to light intensity and the incident angle changes, two adjacent images may have a greater difference, performance for image brightness, saturation is not consistent, therefore, it is necessary to do image radiation correction, in order to eliminate the influence of image color difference on stitching. In view of the UAV in the process of aerial photography, two images of adjacent shots in a very short time, influenced by the atmosphere and the illumination condition of the difference is relatively small, therefore, this paper uses the histogram matching method for relative radiometric correction, by adjusting the reference histogram of the image and make the images histogram matching, from the calibration results, the corrected image and the images are basically the same in brightness and color, very good to achieve a smooth transition of the two images. In addition, the UAV has the advantages of small size, light weight, large air current, stability and Poor wind resistance ability, flight attitude tilt, jitter phenomenon is difficult to avoid, which will have a direct impact on the acquisition of remote sensing image, cause image distortion, the image point coordinates transform, image resampling of the distorted image geometric correction, in order to eliminate the influence of geometric distortion of the image, to meet the needs of image registration with a large amount of calculation. The geometric correction, the long time, in order to improve the processing speed of geometric correction, UAV remote sensing images, this paper proposes a parallel geometric correction algorithm based on distributed processing. The algorithm allows the processor to calculate the required data are stored locally, avoid interprocessor communication, very good to solve the problem of local data, improve the efficiency of parallel processing, in cluster system, the algorithm of parallel geometric correction and geometric correction of serial test, comparison and analysis And the experimental results show that the algorithm has good parallel performance, greatly improving the processing speed of remote sensing image geometric correction. (2) research on remote sensing image registration technology. Image registration system is the core of image stitching technology, image registration accuracy directly determines the quality of image stitching. Because the SIFT algorithm of rotation pan, zoom, and invariant to illumination changes, the perspective changes, affine transformation, noise also maintain a certain degree of stability, therefore, the algorithm is very suitable for UAV remote sensing image registration based on SIFT algorithm for feature point extraction. Feature point matching process, due to the self similarity of adjacent images, often appear a to many or many to one matching error, must be removed. This paper proposes using epipolar and homography constraint error elimination method of feature point matching, firstly The epipolar geometry constraint to reduce the search scope to eliminate the wrong part of the feature points, and then get the correct matching points based on the establishment of homography matrix model, to further eliminate the error matching features of the image, improve the matching accuracy, and RANSAC algorithm is used to remove the outer point of matching feature points for purification, the image feature points the precise matching technology. (3) the study of remote sensing image fusion. The purpose of image fusion is image registration based on the transformation of the model into an image, the two images should be seamless, should not appear splicing traces of natural transition. The UAV remote sensing image acquisition, due to changes in illumination change of incident angle and sensor effects, there will be differences between adjacent image brightness and chroma, for direct image mosaic, the overlapping area of two images will form splicing obvious signs appear, The "ghost" phenomenon. To solve this problem, this paper proposes a method using optimal stitching line to eliminate the "ghost", first used to search the optimal stitching line based on dynamic programming, but the search strategy, if a suture line has some errors, will lead to "ghosting" can not completely eliminate. Get the local optimum suture suture, can obtain the global optimal suture line. Therefore, the search for optimal stitching line based on graph cut, suture line search, obtain the maximum flow minimum cut theorem to the flow graph based on the best suture, two image stitching positioning accurately to get a better solution, and joint dislocation on both sides of the "ghost" phenomenon, realize the seamless splicing. At the same time according to the brightness and chrominance difference image mosaic may exist because of the exposure factors, using Poisson image fusion The method realizes the smooth transition of the image.

【学位授予单位】：东北农业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP751

【相似文献】