基于无人机航拍视频的快速特征匹配与相机方位估计方法研究
发布时间：2019-03-16 08:04
【摘要】：无人机航拍是获取空间数据的重要途径,被广泛应用于军事和民用领域中.其中,基于无人机航拍视频影像的三维重建技术,在城市规划、变化检测、灾害评估等应用中发挥着重要作用.在视频影像的三维重建流程中,特征匹配是基础步骤,为相机方位及参数估计提供可靠输入信息;相机方位估计是三维重建的关键环节,其估计精度与三维重建的效果息息相关.因此,如何提升特征匹配的速度和相机方位的估计精度是当前图像处理、三维重建等领域的研究热点.为此,本文针对无人机航拍视频影像的特点,重点研究了快速特征匹配和相机方位估计问题,主要研究工作如下:1、针对航拍视频影像的特征点提取与匹配速度问题,提出了无迹卡尔曼滤波和KLT匹配算法相结合的特征点跟踪算法,实现了相邻帧特征点的方位预测与快速匹配.该算法首先针对航拍视频特点,利用无迹卡尔曼滤波在相邻帧中进行特征点预测,以确定匹配范围;其次,根据KLT匹配算法对特征点进行跟踪,得到的匹配结果作为观测值;最后,通过卡尔曼增益修正得到特征点的准确位置.对比实验证明了该算法不仅高效,而且在匹配精度方面优于KLT算法.2、针对航拍视频帧视差变化小的特点,提出了关键帧筛选算法,以减少频繁相机方位估计带来的累计误差问题.该算法以特征点对数量和运动大小作为衡量标准,首先利用特征点对计算出图像帧之间的平移量和旋转量,然后进行加权综合求出图像间的相异度,设置阈值筛选关键帧.对比实验证明了通过本文算法得到的关键帧的相机方位比ORB-SLAM得到的关键帧的初始相机方位精度高.本文首先提出无迹卡尔曼滤波和KLT光流结合算法实现了特征点的快速匹配,为相机方位估计提供了可靠的数据.然后提出关键帧筛选算法得到了关键帧的相机方位的精确估计值,相关结果可为快速三维重建方法提供基础理论.
[Abstract]:UAV aerial photography is an important way to obtain spatial data and is widely used in military and civilian fields. Among them, 3D reconstruction technology based on UAV aerial video image plays an important role in urban planning, change detection, disaster assessment and other applications. In the three-dimensional reconstruction process of video image, feature matching is the basic step, which provides reliable input information for camera azimuth and parameter estimation. Camera azimuth estimation is the key link of three-dimensional reconstruction, and its estimation precision is closely related to the effect of three-dimensional reconstruction. Therefore, how to improve the speed of feature matching and the accuracy of camera azimuth estimation is a hot topic in the field of image processing, 3D reconstruction and so on. Therefore, aiming at the characteristics of UAV aerial video image, this paper focuses on the fast feature matching and camera azimuth estimation. The main research work is as follows: 1. Aiming at the feature point extraction and matching speed of aerial photograph video image, the main research work is as follows: 1. A feature point tracking algorithm based on unscented Kalman filter and KLT matching algorithm is proposed to realize azimuth prediction and fast matching of feature points in adjacent frames. Firstly, the algorithm uses unscented Kalman filter to predict the feature points in adjacent frames to determine the matching range. Secondly, according to the KLT matching algorithm, the feature points are tracked and the matching results are used as observation values. Finally, the exact position of the feature points is obtained by Kalman gain correction. The experimental results show that the proposed algorithm is not only efficient, but also superior to KLT algorithm in matching accuracy. 2. In view of the small variation of disparity between aerial video frames, a key frame filtering algorithm is proposed. In order to reduce the cumulative error caused by frequent camera azimuth estimation. In this algorithm, the number of pairs of feature points and the size of motion are used as the criterion. Firstly, the translation and rotation between frames are calculated by using the pair of feature points, then the dissimilarity between images is calculated by weighted synthesis, and the threshold value is set to filter the key frames. The experimental results show that the camera azimuth accuracy of the key frame obtained by this algorithm is higher than that of the key frame obtained by ORB-SLAM. In this paper, a combination algorithm of unscented Kalman filter and KLT optical flow is proposed to realize the fast matching of feature points, which provides reliable data for azimuth estimation of camera. Then the key frame filtering algorithm is proposed to obtain the accurate estimation of the camera orientation of the Keyframe. The related results can provide the basic theory for the fast three-dimensional reconstruction method.
【学位授予单位】：集美大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41
本文编号：2441085
[Abstract]:UAV aerial photography is an important way to obtain spatial data and is widely used in military and civilian fields. Among them, 3D reconstruction technology based on UAV aerial video image plays an important role in urban planning, change detection, disaster assessment and other applications. In the three-dimensional reconstruction process of video image, feature matching is the basic step, which provides reliable input information for camera azimuth and parameter estimation. Camera azimuth estimation is the key link of three-dimensional reconstruction, and its estimation precision is closely related to the effect of three-dimensional reconstruction. Therefore, how to improve the speed of feature matching and the accuracy of camera azimuth estimation is a hot topic in the field of image processing, 3D reconstruction and so on. Therefore, aiming at the characteristics of UAV aerial video image, this paper focuses on the fast feature matching and camera azimuth estimation. The main research work is as follows: 1. Aiming at the feature point extraction and matching speed of aerial photograph video image, the main research work is as follows: 1. A feature point tracking algorithm based on unscented Kalman filter and KLT matching algorithm is proposed to realize azimuth prediction and fast matching of feature points in adjacent frames. Firstly, the algorithm uses unscented Kalman filter to predict the feature points in adjacent frames to determine the matching range. Secondly, according to the KLT matching algorithm, the feature points are tracked and the matching results are used as observation values. Finally, the exact position of the feature points is obtained by Kalman gain correction. The experimental results show that the proposed algorithm is not only efficient, but also superior to KLT algorithm in matching accuracy. 2. In view of the small variation of disparity between aerial video frames, a key frame filtering algorithm is proposed. In order to reduce the cumulative error caused by frequent camera azimuth estimation. In this algorithm, the number of pairs of feature points and the size of motion are used as the criterion. Firstly, the translation and rotation between frames are calculated by using the pair of feature points, then the dissimilarity between images is calculated by weighted synthesis, and the threshold value is set to filter the key frames. The experimental results show that the camera azimuth accuracy of the key frame obtained by this algorithm is higher than that of the key frame obtained by ORB-SLAM. In this paper, a combination algorithm of unscented Kalman filter and KLT optical flow is proposed to realize the fast matching of feature points, which provides reliable data for azimuth estimation of camera. Then the key frame filtering algorithm is proposed to obtain the accurate estimation of the camera orientation of the Keyframe. The related results can provide the basic theory for the fast three-dimensional reconstruction method.
【学位授予单位】：集美大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41
【参考文献】
相关期刊论文 前10条
1 刘华咏;郝会芬;李涛;;基于视频聚类的关键帧提取算法[J];物联网技术;2014年08期
2 陈智;;浅议无人机航拍的应用[J];时代金融;2014年15期
3 吴伟交;王敏;黄心汉;毛尚勤;;基于向量夹角的SIFT特征点匹配算法[J];模式识别与人工智能;2013年01期
4 肖晗;原魁;何文浩;柴晓杰;;基于FPGA的SIFT特征点检测[J];高技术通讯;2012年04期
5 冷大炜;孙卫东;;一种高数值精度的P3P问题半闭式解法[J];计算机应用研究;2012年01期
6 杨现辉;王惠南;;ICP算法在3D点云配准中的应用研究[J];计算机仿真;2010年08期
7 刘晓光;陈曦;陈政伟;孙静波;;基于图像灰度的SSDA匹配算法[J];航空计算技术;2010年01期
8 张曼;沈旭昆;;一种基于尺度空间的三维点云数据配准算法[J];系统仿真学报;2009年S1期
9 许海霞;王耀南;袁小芳;朱江;周维;;基于矢量差分的未标定摄像机P5P问题的求解[J];自动化学报;2009年08期
10 汤建良;;一个关于P5P问题的求解算法[J];广西师范大学学报(自然科学版);2009年01期
相关硕士学位论文 前5条
1 张一;无人机遥感影像点特征匹配算法研究[D];解放军信息工程大学;2015年
2 吴昊;基础矩阵估计方法研究[D];兰州大学;2015年
3 谢榛;基于无人机航拍图像的室外场景三维重建技术研究[D];浙江工业大学;2014年
4 曹长青;基于内容的视频检索中关键帧提取算法研究[D];太原理工大学;2013年
5 明巍;基于内容的视频检索中关键帧提取算法研究[D];武汉工业学院;2010年
，本文编号：2441085
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2441085.html
