不同特征提取算法对相机运动估计的适用性研究  

The applicability of different feature extraction algorithms to camera motion estimation

作　　者：查冰 张力[1] 艾海滨[1] ZHA Bing, ZHANG Li, AI Haibin(Chinese Academy of Surveying and Mapping, Beijing 100830, Chin)

机构地区：[1]中国测绘科学研究院,北京100830

出　　处：《测绘科学》2018年第3期92-98,共7页Science of Surveying and Mapping

基　　金：国家自然科学基金项目(41271374)

摘　　要：针对不同特征提取算法对不同传感器平台获取的影像空间相对关系估计的适用性问题,该文以不同来源的影像数据为研究对象进行实验分析。采用基于特征的影像匹配算法SIFT,SURF,ORB对影像进行特征提取和匹配,利用RANSAC算法进行粗差剔除,随后使用归一化的八点法估计基础矩阵,最后采用辛普森距离统计像素均方根误差。结果表明,在不考虑速度的情况下,SIFT算法对于各类影像的相对位置关系估计有较好的精度;ORB算法在速度上有较大优势,检测和匹配的特征点数目最多;SURF算法的速度和精度介于两者之间。Aiming at the applicability of different feature extraction algorithms for the estimation of the relative relationship of image space obtained by different sensor platforms,this paper analyzes the image data from different sources as the research object.In this paper,feature-based image matching algorithm SIFT,SURF,and ORB are used to extract and match the images,and the RANSAC algorithm is used to eliminate the coarse difference,then the basic matrix is estimated by the normalized 8-point method,and the root mean square error of the Simpson distance is used.The results show that,without considering the speed,the SIFT algorithm has a good accuracy for the relative position relationship estimation of all kinds of images.The ORB algorithm has a great advantage in speed,and the number of feature points detected and matched is the most.The speed and precision of SURF algorithm are between the two algorithm.

关 键 词：特征提取 特征匹配 基础矩阵 极线约束 随机抽样一致性算法 

分 类 号：P235[天文地球—摄影测量与遥感]