基于RGB-D图像的机器视觉定位算法  

Machine Vision Positioning Algorithm Based on RGB-D Image

作　　者：林秋叶 胡志恒[1] LIN Qiuye;HU Zhiheng(School of Communication Engineering,Chengdu University of Information Technology,Chengdu 610225)

机构地区：[1]成都信息工程大学通信工程学院,成都610225

出　　处：《现代计算机》2021年第17期140-145,157,共7页Modern Computer

摘　　要：基于视觉的同时定位与地图构建(Simultaneous Localization and Mapping,VSLAM)是近几年机器人领域研究的热点,也是机器人在未知环境中实现自主导航的关键技术。针对现有视觉定位算法运行效率相对较低和鲁棒性较差的问题,本文提出一种改进的RGB-D SLAM算法来进一步提升实时性与鲁棒性。首先,在提取特征点与匹配过程使用四叉树策略,结合自适应阈值函数均匀提取ORB关键点,计算关键点的描述子汉明距离进行粗匹配;其次,采用PROSAC算法进行匹配点筛选,取质量较高的匹配点对结合PNP (Perspective-N-Point)和ICP求解位姿,获取经过PROSAC算法多次迭代后的准确位姿;最后基于光束平差法求解相机位姿的最优解,从而构建全局一致的地图。在TUM RGB-D数据集上进行验证,实验结果表明本文算法在实时性与鲁棒性上综合表现优于传统视觉定位算法。Simultaneous Localization and Mapping( VSLAM) based on vision is a research hotspot in the field of robotics in recent years,and it is also a key technology for robots to realize autonomous navigation in unknown environments. Aiming at the relatively low operating efficiency and poor robustness of existing visual positioning algorithms,this paper proposes an improved RGB-D SLAM algorithm to further improve the real-time and robustness. First,the quadtree strategy is used in the process of extracting feature points and matching,combined with the adaptive threshold function to uniformly extract ORB key points,and calculate the key point descriptor Hamming distance for rough matching;secondly,use the PROSAC algorithm to screen the matching points. High-quality matching point pairs combine PNP( Perspective-N-Point) and ICP to solve the pose,and obtain the accurate pose after multiple iterations of the PROSAC algorithm;finally,the optimal solution of the camera pose is solved based on the beam adjustment method,so as to build a globally consistent map. It is verified on the TUM RGB-D data set,and the experimental results show that the algorithm in this paper is better than the traditional visual positioning algorithm in terms of real-time and robustness.

关 键 词：视觉定位 VSLAM PROSAC 误匹配剔除 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]