纹理感知的倾斜影像自适应密集匹配方法

发布时间：2018-08-08 17:55

【摘要】：倾斜摄影能够全方位、高效地获取建筑物立面信息,在城市三维重建中表现出独特的优势,已成为一种重要的数据获取方式。影像密集匹配技术是对立体像对逐像素地建立对应关系,根据三角测量原理获得像素的三维空间位置。倾斜摄影相对于传统的垂直摄影方式,不同相机的拍摄视角差异较大,影像包含地物种类错综复杂,不同地物深度变化范围较广,地物之间有明显的遮挡,这给倾斜影像密集匹配带来挑战。针对航空倾斜影像深度变化范围大、纹理缺乏导致的自适应匹配难题,本文比较现有的立体像对匹配方法,以高效的半全局立体匹配算法为基础,在匹配过程中引入纹理信息约束,自适应调整匹配参数,提高算法对不同地形、不同地物的匹配能力。本文的主要研究内容如下:(1)通过分析比较不同的影像纹理检测的算法,采用较为鲁棒的方法来定量描述影像的纹理特征,该纹理特征能够在一定程度上反映像素的深度变化,在匹配的过程中顾及到纹理信息,能够显著提高匹配结果精度。(2)为了降低匹配算法对参数的敏感,本文以像素的纹理属性为依据,对不同纹理属性的像素,采用不同的匹配代价组合策略,同时为了增强初始匹配代价的鲁棒性,对弱纹理的区域进行代价聚合。在匹配阶段,自适应地计算像素的半全局匹配的惩罚参数,在不同纹理区域采用不同的平滑性约束,克服点云噪声、无法保留视差断裂带的问题。(3)研究采用GPGPU实现上述匹配过程,提高对无人机倾斜影像的处理效率,使得该算法能够应用在实际的三维生产中。
[Abstract]:The oblique photography can obtain the facade information in all directions and efficiently. It has shown unique advantages in the three dimensional reconstruction of the city. It has become an important method of data acquisition. The image intensive matching technique is a pixel by pixel based on the establishment of the corresponding relationship. Compared with the traditional vertical photography, the different cameras have a large difference in shooting angle, the image contains the complex types of objects, the range of the depth of different objects is wide and there is a clear occlusion between the objects. This brings a challenge to the intensive matching of the inclined images. In order to match the matching problem, this paper compares the existing stereo pair matching method, based on the efficient semi global stereo matching algorithm, and introduces the texture information constraint in the matching process, adaptively adjusts the matching parameters, and improves the matching ability of the algorithm to different terrain and different objects. The main contents of this paper are as follows: (1) through analysis and comparison, the main contents of this paper are as follows: The same image texture detection algorithm uses a more robust method to quantitatively describe the texture features of the image. The texture features can reflect the depth of the pixel to a certain extent, and the texture information is taken into account in the matching process. (2) in order to reduce the sensitivity of the matching algorithm to the parameters, this paper is to reduce the sensitivity of the matching algorithm to the parameters. Based on the texture properties of pixels, the pixels with different texture attributes are combined with different matching costs. At the same time, in order to enhance the robustness of the initial matching cost, the region of the weak texture is aggregated. In the matching phase, the penalty parameters for the half full match of the pixels are calculated adaptively, and the different texture regions are not used. The same smoothing constraint can overcome the point cloud noise and can not retain the problem of parallax fault zone. (3) the study of using GPGPU to achieve the above matching process is used to improve the processing efficiency of the unmanned aerial vehicle (UAV) image, so that the algorithm can be applied to the actual three-dimensional production.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】