中文摘要：

自然路标提取与匹配是vSLAM的基础。文中提出了一种基于特征点三维信息的自然路标提取、局部特征描述与快速匹配方法。采用双目视觉获取环境图像,提取左右目图像的特征点,并进行匹配。建立左摄像机坐标系下的每个匹配点的三维信息,提出视场约束规则对特征点进行过滤。在此基础上基于改进的MeanShift聚类算法进行自然路标提取。提出一种路标描述符,可以快速进行两个聚类的匹配。该方法可以有效提取非结构化环境中的自然路标,对机器人位姿估计精度要求较低。

英文摘要：

Landmark extraction and matching is basis of vSLAM.A method of landmark extraction,local feature description and fast matching based on 3D information of feature points is proposed.Robot obtains images of environment via binocular vision,extracting feature points from left and right eye images,matching feature points of the two images.Three-dimensional information of each matched points under left camera coordinate system is built.Field of view constraint rule is proposed to filtering points.Then,the method of natural landmark extraction based on improved Mean Shift algorithm is discussed.The paper proposes a landmark descriptor,which can achieve fast matching of the two clustering.This method can extract natural landmarks in unstructured environment,tolerating relatively low accuracy of pose estimation.