基于局部特征与拓扑约束的遥感图像匹配与目标定位

发布时间：2018-02-21 22:31

本文关键词： 遥感图像特征提取拓扑关系图像匹配仿射变换　出处：《东北大学》2014年硕士论文　论文类型：学位论文

【摘要】：地面目标定位是无人机的重要应用之一,也是目前网络处理和模式识别领域的热点研究课题之一。传统的无人机系统依靠GPS导航定位,当GPS失效的情况下,通常利用惯性导航、光电测量等方法进行定位,但是这些方法对设备依赖性较大,易受环境影响。因此本文提出了基于无人机航拍图像与遥感图像匹配的目标定位算法,并通过实验验证了算法的有效性和优越性。本文的工作与成果主要体现在以下几个方面：(1)本文提出了利用局部特征和拓扑约束相结合的目标定位算法框架。基于图像局部特征实现关键点的提取和特征描述,再根据关键点的拓扑关系剔除误匹配点,最后通过估计仿射参数,得到目标的准确位置。由于传统的特征匹配只利用关键点自身的特征,当图像中存在特征相似的关键点时就会出现误匹配,而误匹配点会严重影响定位精度,所以为了提高定位精度,本文兼顾局部特征和拓扑关系,实验证明,这种方法可以提高最终定位的精度。(2)本文提出利用多尺度FAST检测图像角点,并利用关键点局部特征SIFT描述子进行预匹配。多尺度FAST克服了算法本身不具有尺度不变性的缺点,SIFT描述子的生成过程使关键点的特征向量具有旋转不变性,基于欧式距离的评价标准对两幅图像中的关键点进行预匹配。从实验可以看出多尺度FAST角点比SIFT极值点具有更高的准确性、稳定性。(3)本文提出了利用Delaunay三角剖分对匹配点进行拓扑约束,剔除误匹配点。当图像发生尺度缩放和旋转变化时,关键点内部的空间关系保持不变,待匹配图像中关键点组成的三角剖分面的边长与参考图像中对应匹配点构成的边长保持一定的比例关系,利用两者之间的比例关系,将误匹配点剔除。(4)本文提出了用局部区域的RANSAC估计参数模型,实现目标定位。由于拍摄过程中图像可能存在畸变,不同区域的仿射参数不同,所以用RANSAC估计目标点附近局部区域的仿射模型,作为该目标的仿射模型,最终获得目标在遥感图像中的位置坐标,从定位对比实验可以看出局部RANSAC比全局RANSAC具有更高的准确性。
[Abstract]:Ground target positioning is one of the important applications of UAV, and it is also one of the hot research topics in the field of network processing and pattern recognition. Traditional UAV systems rely on GPS navigation to locate, and when GPS fails, inertial navigation is usually used. However, these methods are highly dependent on the equipment and easily affected by the environment. Therefore, a target location algorithm based on the matching of aerial and remote sensing images of UAV is proposed in this paper. The effectiveness and superiority of the algorithm are verified by experiments. The work and achievements of this paper are mainly reflected in the following aspects: 1) in this paper, a target localization framework based on local features and topological constraints is proposed. The local feature of the image realizes the extraction and description of the key points. Then the mismatch points are eliminated according to the topological relation of the key points, and the exact position of the target is obtained by estimating the affine parameters. Because the traditional feature matching only uses the features of the key points, In order to improve the location accuracy, the local feature and topological relationship are taken into account in this paper, and the experimental results show that the mismatch will occur when there are some key points similar to the features in the image, and the mismatch points will seriously affect the location accuracy. This method can improve the accuracy of the final location.) in this paper, we propose to use multi-scale FAST to detect image corners. Multi-scale FAST overcomes the shortcoming that the algorithm itself does not have scale invariance. The generation process of sift descriptor makes the feature vector of key points rotation-invariant. Based on the Euclidean distance evaluation criterion, the key points in the two images are pre-matched. It can be seen from the experiments that the multi-scale FAST corner is more accurate than the SIFT extremum. In this paper, we use Delaunay triangulation to constrain the matching points and eliminate the mismatched points. When the image scale is scaled and rotated, the spatial relationship within the key points remains unchanged. The edge length of the triangulation surface formed by the key points in the image to be matched maintains a certain proportion relationship with the length of the corresponding matching point in the reference image. In this paper, the parameter model of RANSAC estimation of local region is proposed to realize target localization. Because of the possible distortion of the image during shooting, the affine parameters of different regions are different. So using RANSAC to estimate the affine model of the local region near the target point, as the affine model of the target, the position coordinates of the target in the remote sensing image can be obtained. From the localization contrast experiment, it can be seen that the local RANSAC is more accurate than the global RANSAC.
【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【相似文献】