基于MMS影像的SIFT算法改进与实现

发布时间：2018-06-29 21:58

本文选题：移动测量系统 + 特征提取　；参考：《北京建筑大学》2013年硕士论文

【摘要】：近些年来，移动测量系统在各个行业中得到了广泛应用，获得了很好的实践成果，促进了社会的进步发展。由于移动测量系统可以采集大量的可量测实景影像数据，广泛应用到影像浏览，以及后期提取影像中地物的空间位置信息及属性信息中，因此影像数据显得越来越重要，受到人们的重视。本文研究所使用的移动测量系统外业采集的影像数据，由于相机的视场角较小，拍摄的影像不能满足项目的需要（宽幅影像）。从这个问题出发，本文研究适合移动测量系统的影像拼接算法。在移动测量系统中，相机拍摄的影像之间都有一定的重叠度，而且这个重叠度基本上是固定的，因此，本文将移动测量系统的影像特点作为限制条件，改进SIFT匹配算法，制作宽幅影像。另外，移动测量系统拍摄影像数量巨大，而且影像有固定的命名及存储格式，因此，本文给出了开发适合移动测量系统的影像自动拼接的系统构思。本文对移动测量系统拍摄影像的特点进行了全面的分析，提取移动测量系统的独特特征数据，比如相机的固定位置及角度关系，影像间的位置及旋转关系,影像的重叠度等等，改进SIFT的特征提取算法以及特征匹配算法，研究适合移动测量系统影像数据的拼接算法。本文选用在不同级别的道路上拍摄的影像做统计实验。首先，本文通过对影像进行模型分析以及统计实验，获取两个相机拍摄影像的重叠度数据，根据获取的重叠度数据制定相应的阈值，踢除不符合要求的特征点数据。其次，通过影像统计实验，确定特征点匹配的搜索区域H。以正前相机拍摄的影像为标准影像，左前跟右前相机拍摄的影像为待拼接影像，通过人工判读的方法，提取影像的同名特征点，分析同名特征点数据，获取与标准影像特征点匹配的待配准影像特征点的大致区域，制定特征点匹配的搜索区域H。在待配准影像中，以H的矩形搜索区域为搜索范围，运用欧式距离匹配影像特征点。这样不仅提高算法的运算速度，而且还有效减少伪匹配特征点的数量。最后，通过最小二乘原理，随机选取一定数量的特征点建立转换模型，剩余的特征点作为检核点，检验转换模型的可靠性，获取可靠性最高的转换模型。通过本文的研究，提高了算法的运算速度，成功降低特征点伪匹配概率，获取了可信度最高的同名特征点，建立了影像拼接模型。本论文改进的匹配算法，适合移动测量系统，而且在此算法基础上开发的影像自动拼接系统，，为街景影像的后期发布提供便利，对整个移动测量系统功能的完善和改进起到了积极的推动作用。
[Abstract]:In recent years, mobile measurement system has been widely used in various industries, obtained good practical results, and promoted the development of society. Because the mobile measurement system can collect a large number of measurable real scene image data, it is widely used in image browsing, as well as the spatial position information and attribute information of the object in the image, so the image data becomes more and more important. Be valued by people. In this paper, the image data collected from the field of view of the mobile measurement system are studied. Because of the small field of view angle of the camera, the image can not meet the needs of the project (wide range image). From this point of view, this paper studies the image stitching algorithm suitable for mobile measurement system. In the mobile measurement system, there is a certain degree of overlap between the images taken by the camera, and the overlap degree is basically fixed. Therefore, this paper takes the image characteristics of the mobile measurement system as the limiting condition, and improves the sift matching algorithm. Make wide images. In addition, the number of images taken by the mobile measurement system is huge, and the images have a fixed naming and storage format. This paper makes a comprehensive analysis of the characteristics of the image taken by the mobile measurement system, extracts the unique characteristic data of the mobile measurement system, such as the fixed position and angle relation of the camera, the position and rotation relationship between the images, the overlap degree of the image, etc. Improved sift feature extraction algorithm and feature matching algorithm, research suitable for mobile measurement system image data splicing algorithm. In this paper, we choose the images taken on different levels of roads to do statistical experiments. Firstly, through the model analysis and statistical experiment of the image, the overlap degree data of the two cameras are obtained, the corresponding threshold is set according to the overlap degree data obtained, and the feature point data that does not meet the requirements is kicked out. Secondly, the search region of feature point matching is determined by image statistical experiment. Taking the image taken by the front camera as the standard image and the image taken by the left front camera and the right front camera as the image to be stitched together, the feature points of the same name of the image are extracted by manual interpretation, and the data of the same name feature point are analyzed. The approximate region of the feature points to be matched with the standard image feature points is obtained, and the search region of the feature points matching is established. In the image registration, the rectangular search area of H is used as the search range and the Euclidean distance is used to match the feature points of the image. This not only improves the speed of the algorithm, but also effectively reduces the number of pseudo-matching feature points. Finally, through the least square principle, a certain number of feature points are randomly selected to establish the transformation model, and the remaining feature points are used as check points to test the reliability of the transformation model and obtain the most reliable transformation model. Through the research in this paper, the algorithm speed is improved, the probability of pseudo-matching of feature points is reduced successfully, the feature points of the same name with the highest credibility are obtained, and the image mosaic model is established. The improved matching algorithm in this paper is suitable for the mobile measurement system, and the automatic image mosaic system based on this algorithm provides convenience for the later release of the street view image. It plays a positive role in improving the function of the whole mobile measurement system.
【学位授予单位】：北京建筑大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：P235.2

【参考文献】