航空视频影像的正射影像制作关键技术研究

发布时间：2019-07-03 18:57

【摘要】：近年来,无人机低空航测正逐步成为与卫星遥感、常规航空摄影测量相并列的航空遥感技术,其主要目标是为了满足现代社会对于及时测绘和精细测绘的应用需求。尽管具有及时性、精确性的优点,但通常需等待无人机降落之后才能获取传感器系统所拍摄的影像数据,无法满足对及时性要求极高的应急测绘需求。针对这一问题,将数码摄像机集成于传感器系统中,并利用无线传输技术将拍摄的视频进行实时下传。然而,受制于视频图像的小像幅,小视场、低分辨率、无法直观反映测区整体概况等缺陷,限制了其潜在的应用价值。因此,研究利用航空视频进行正射影像制作这一新兴领域,具有迫切的应用需求及研究价值。将航空视频拼接为正射影像主要涉及两方面内容：第一,如何将动态的视频数据转换为静态的帧图像数据；第二,如何利用无初始位置信息的帧图像数据进行正射影像制作。针对第一个问题,需研究视频图像的提取技术,同时提取的帧图像之间,其重叠度需满足影像拼接要求。第二个问题则可分为两个子问题加以解决。首先恢复帧图像的空间结构,即影像在拍摄时空中的位置和姿态信息。其次,对恢复姿态后的影像进行正射纠正、拼接生成几何位置一致、颜色变化平滑的正射影像。因此,本文的研究工作及创新之处主要体现在： (1)首先,系统的总结了2D和3D至2D的影像几何变换模型,并分析了两种模型的几何精度差异。其次,在给出两种模型参数估计方法的基础上,结合两种模型的技术特点、几何精度及应用需求,设计了基于像方和基于物方的两种正射影像拼接流程。 (2)研究了摄像机的静态几何检校方法及流程。首先,在分析数码摄像机畸变规律基础上,通过对比经典的畸变模型与Brown模型的关系,提出了一种顾及高阶项和交叉项的畸变模型,使摄像机的静态总体检校精度小于0.5个像元。其次,详细推导了数码摄像机的检校流程,并对提出的畸变模型检校结果进行分析,指出本文所提畸变模型的有效性。 (3)基于曲线拟合原理,提出一种自适应的关键帧提取方法。在分析UAV载航空视频重叠度变化规律的基础上,提出一种两步法关键帧提取算法：即学习阶段和提取阶段。学习阶段主要计算当前地形及飞行条件下,视频重叠度的变化规律。提取阶段主要依据学习阶段提供的初始值,在特定区间内进行重叠度抽样,并根据抽样结果对该范围内的重叠度变化规律进行曲线拟合,进而按照拟合结果计算满足重叠度需求的帧索引位置。 (4)针对运动像移所造成的影像模糊,研究利用图像处理的方法对其进行恢复,获取清晰影像。首先,详细总结了目前国内外关于核函数估计及影像去模糊处理的研究进展。其次,基于信息熵、信噪比、信息量等信息论测度,建立影像的有参和无参评价指标体系。最后以地面试验及空中试验结果为指导,提出适合视频帧图像的运动像移恢复策略。 (5)提出一种基于像方的影像空间结构恢复及优化方法。通过分析基于像方的影像转换模型及其误差传播规律,针对帧图像无POS信息的特点,提出一种利用连接点属性将新加入影像逐渐纳入到已经优化后影像序列中的结构恢复方法,并通过最小二乘原理将累计的误差合理配置于所有参与平差的影像中,使所有像点的总体投影误差最小。最后,通过试验结果验证了本文方法的有效性。 (6)提出一种CPU与GPU协同处理的航空视频拼接流程。针对航空视频拼接的及时性需求,总结国内外关于遥感数据的并行处理研究进展,提出利用GPU技术对视频拼接的关键步骤进行并行加速,并根据CPU模式与GPU模式的特点,设计了利用两种处理器进行协同操作的拼接流程。 (7)通过具体的工程实例,阐述了航空视频拼接的作业流程,展示了相关的处理成果,并分析对比了像方和物方的拼接影像,指出两种方法的优缺点,验证了本文以航空视频进行正射影像制作的可行性。本文以研究解决利用航空视频进行正射影像制作过程中的关键技术为研究目的,结合摄像机检校、关键帧提取、运动像移恢复、影像空间结构重建、GPU并行运算等关键技术,利用关键帧图像完成了基于像方和基于物方的正射影像制作,拓展了航空视频的应用范围。
[Abstract]:In recent years, the low-altitude aerial survey of the unmanned aerial vehicle is becoming an aerial remote sensing technology which is parallel to the satellite remote sensing and the conventional aerial photogrammetry, and the main aim is to meet the application demand of the modern society for timely mapping and fine mapping. Although it has the advantages of timeliness and accuracy, it is generally necessary to wait for the unmanned aerial vehicle to land, acquire the image data taken by the sensor system, and can not meet the emergency mapping requirement with high requirements for timeliness. Aiming at the problem, the digital video camera is integrated in the sensor system, and the captured video is transmitted in real time by the wireless transmission technology. However, the small image, small field of view and low resolution, which are subject to the video image, can not directly reflect the overall situation of the survey area and other defects, and limit its potential application value. Therefore, the research of using aerial video to make the orthophoto image to make this new field, has the urgent application requirement and the research value. The method comprises the following steps of: firstly, converting the dynamic video data into static frame image data; secondly, using the frame image data without the initial position information to perform the orthophoto-image system; For the first problem, the extraction technology of the video image needs to be studied, and the overlapping degree of the extracted frame images is to meet the image splicing requirements. The second question can be divided into two sub-problems. The first step is to restore the spatial structure of the frame image, that is, the position and attitude of the image in the air at the time of the shooting. and then, the image after the restoration attitude is corrected, the geometric position of the splicing generation is consistent, and the color change is smooth and the positive projection As a result, the research work and innovation of this paper are mainly embodied in this paper. (1) Firstly, the geometric transformation model of 2D and 3D-2D is summarized, and the geometry of the two models is analyzed. Secondly, on the basis of the two model parameter estimation methods, the two kinds of positive projective images based on the image side and the object side are designed in combination with the technical characteristics, the geometric precision and the application requirement of the two models. Connected to the process. (2) The static geometric calibration of the camera is studied. In this paper, based on the analysis of the distortion law of digital camera, a kind of distortion model, which takes into account the high-order terms and cross terms, is put forward based on the analysis of the distortion law of the digital camera, so that the accuracy of the static total physical examination of the camera is less than 0. . Secondly, the calibration process of the digital camera is derived in detail, and the result of the proposed distortion model is analyzed, and the distortion model proposed in this paper is pointed out. The validity of the model is based on the principle of curve fitting. In this paper, a two-step method of key-frame extraction is proposed based on the analysis of the variation law of the degree of overlap of UAV-loaded aerial video. Stage and extraction phase. The learning phase mainly calculates the current terrain and flight conditions, and the video is heavy The extraction stage is mainly based on the initial value provided in the learning phase, the overlap degree sampling is carried out in a specific interval, and the variation law of the overlap degree in the range is curve-fitted according to the sampling result, and the overlapping degree needs to be met according to the fitting result. (4) image blurring caused by moving image, and the method of image processing is used to study it. Restore and get a clear image. First of all, a detailed summary of the current home and abroad about kernel function estimation and image The research progress of fuzzy processing is based on information theory measure of information entropy, signal-to-noise ratio, information quantity and so on. and finally, based on the ground test and the air test result, a suitable video frame image is proposed, And (5) a video-based image is proposed. The spatial structure recovery and optimization method is presented. By analyzing the image-based image transformation model and its error propagation law, a new image is gradually incorporated into the optimized image by using the attribute of the connection point, aiming at the characteristics of the image-free information of the frame image. The structure recovery method in the image sequence is adopted, and the accumulated error is reasonably arranged in the image of all the participating adjustment by the least square principle, so that all the images The overall projection error of the image point is minimal. Finally, by the test results The effectiveness of this method is verified. (6) A kind of CPU and GPU are put forward. In this paper, the research progress of parallel processing of remote sensing data at home and abroad is summarized, and the key steps of video splicing are accelerated in parallel by using GPU technology. The characteristics of the PU mode and the GPU mode are designed. (7) The operation flow of the aviation video splicing is described through the specific project examples, and the relevant processing results are shown, and the comparison of the image and the object side is also analyzed. The image is spliced and the advantages and disadvantages of the two methods are pointed out. The feasibility of making a positive projection image of an empty video is presented in this paper. The key technologies in the process of using aerial video to make the orthophoto image are studied, and the camera calibration, the key frame extraction, the motion image removal and the image space are combined. The key technologies such as structure reconstruction, GPU parallel operation and other key technologies are used to complete the image-based and object-based direct injection by using the key frame images.
【学位授予单位】：武汉大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：P231

【参考文献】