复杂场景下的水上桥梁识别技术研究

发布时间：2018-04-21 01:23

本文选题：自动识别 + 遥感图像　；参考：《南京航空航天大学》2017年硕士论文

【摘要】：伴随着计算机视觉技术的发展,遥感图像中的目标自动识别已经成为研究的热点和重点。桥梁作为典型且重要的人工建筑,是交通运输线路的咽喉。对采集图像中的桥梁进行有效识别,在民用及军事上都有着广泛需求。本文以复杂场景下的可见光遥感图像中桥梁目标的自动识别作为研究背景,针对河流的自动提取和河流上桥梁的定位等问题进行了研究。根据复杂场景下水上桥梁图像的特点:不同图像中河流差异较大,水面可能平静,分布均匀,也可能由于波浪和水体浑浊造成河流纹理丰富,分布不均匀。复杂场景下图像的背景区域也多样性,可能包含多种自然景物、农作物或人工建筑,如林地、耕地和居民地等,使得背景中包含多种形式的纹理。在此基础上本文建构了一套完整的水上桥梁自动识别系统。主要研究内容如下:(1)复杂场景下的桥梁图像,河流虽然呈现出不同的形式,其灰度值或高或低,分布或均匀或杂乱。但河流区域之间的颜色相似度较高,河流区域的颜色相比背景区域的颜色差异较大。根据颜色特征相似度,本文提出了K均值聚类与Harris角点相结合的无监督分割法实现河流的自动提取。(2)针对更复杂的桥梁图像,即河流区域内,部分颜色差异较大,河流颜色无明显规律的情况,K均值聚类与Harris角点相结合的方法不能提取出较为完整的河流,但可以提取出河流区域中的部分样本。利用这部分样本提取出能代表本张图像中的河流区域的颜色和纹理特征,进行学习,采用自监督分割方法,对图像中的所有像素点分类,分割出完整的河流区域。(3)对分割出的河流进行形态学操作和干扰区域的剔除,得到较为完整的河流轮廓。对河流二值图像膨胀腐蚀,填补截断位置得到连通的河流二值图,与原来的河流二值图像作差,提取出疑似桥梁的截断区域。利用河流骨架与桥梁相交的特征剔除部分虚假桥梁,再根据桥梁拐点的特征,验证真实桥梁的存在性,完成真实桥梁的获取,将验证后的桥梁对应在原图像的位置进行标记,从而实现桥梁定位。(4)本文提出的桥梁自动识别方法以Visual Studio 2010为开发平台,并结合OpenCV开源视觉库实现了该系统的开发。实验表明,该系统能够自动识别出高空水上桥梁图像中的桥梁目标,并具有一定的适用性。
[Abstract]:With the development of computer vision technology, automatic target recognition in remote sensing images has become a hot spot and focus. As a typical and important artificial building, bridge is the throat of transportation line. It is widely needed in civil and military fields to identify bridges in image collection. In this paper, the automatic recognition of bridge targets in visible light remote sensing images of complex scenes is used as the research background, and the automatic extraction of rivers and the location of bridges on rivers are studied in this paper. According to the characteristics of bridge image in the water of complex scene: the river is different in different images, the water surface may be calm and distributed evenly, or the river texture may be rich and uneven due to the wave and water turbidity. The background areas of the images in complex scenes are also diverse and may include many natural scenes crops or artificial buildings such as forest farmland and inhabitant land which make the background contain many kinds of textures. On the basis of this, this paper constructs a set of complete automatic recognition system of water bridge. The main contents of this study are as follows: (1) Bridge images in complex scenes. Although rivers show different forms, their gray values are high or low, and their distribution is uniform or chaotic. However, the color similarity of river region is higher, and the color of river region is more different than that of background region. According to the similarity of color features, an unsupervised segmentation method based on K-means clustering and Harris corner is proposed to automatically extract river. If there is no obvious rule of river color, the method of K-means clustering combined with Harris corner can not extract more complete river, but can extract some samples from river region. Using this part of the sample to extract the color and texture features which can represent the river region in this image, to learn, to use the self-supervised segmentation method, to classify all the pixels in the image. The whole river area is segmented. (3) the whole river contour is obtained by morphological operation and elimination of the interference area of the segmented river. When the river binary image is dilated and corroded, the connected river binary image is obtained by filling the truncation position, which is different from the original river binary image, and the truncated area of the suspected bridge is extracted. By using the feature of river skeleton intersecting with bridge, some false bridges are eliminated, then the existence of real bridge is verified according to the characteristics of bridge inflection point, and the acquisition of real bridge is completed, and the verified bridge is marked corresponding to the position of the original image. The bridge automatic recognition method proposed in this paper is developed on the platform of Visual Studio 2010 and the open source vision library of OpenCV is used to realize the development of the system. The experimental results show that the system can automatically identify the bridge targets in the image of high altitude water bridge, and it is applicable to some extent.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】