基于点线面特征的无漂移旋转视觉里程计  

Drift-free rotation visual odometry based on point-line-plane feature fusion

作　　者：李用杰 秦广健 武利明[2] 熊军林 Li Yongjie;Qin Guangjian;Wu Liming;Xiong Junlin(School of Information Science&Technology,University of Science&Technology of China,Hefei 230026,China;Wireline Product R&D Institute,ZTE Corporation,Nanjing 210012,China)

机构地区：[1]中国科学技术大学信息科学技术学院,合肥230026 [2]中兴通讯股份有限公司有线研究院,南京210012

出　　处：《计算机应用研究》2023年第12期3805-3809,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(62273320)。

摘　　要：现有基于点特征的视觉SLAM(simultaneous localization and mapping)算法在弱纹理环境中表现不佳,为此提出了一种基于点线面特征融合的视觉里程计算法,能够在弱纹理环境中实现精准定位。首先基于曼哈顿世界假设下,使用线特征与面特征提取曼哈顿世界坐标系,并将线特征与面特征与坐标系联合;其次为了提升系统定位的准确性,使用了一种无漂移旋转的位姿估计算法,将位姿的旋转与平移分开求解;最后利用结构化的线特征与面特征对位姿与曼哈顿轴进行优化,综合考虑图像中的点线面特征,使得位姿估计的结果更加精确。实验表明,该算法在TUM与ICL-NUIM数据集中的表现优于目前的其他方法。The existing visual SLAM algorithm based on point feature performs poorly in weak texture environments.Therefore,this paper proposed a visual odometry based on point-line-plane feature fusion,which could accurately locate in weak texture environment.Firstly,based on the assumption of Manhattan world,it extracted the Manhattan world coordinate system by using line features and plane features,and combined line features and plane features with the coordinate system.Secondly,in order to improve the positioning accuracy,this paper used a drift-free rotation position estimation algorithm,which solved the rotation and translation of pose separately.Finally,it optimized the pose and Manhattan axis by using structured line and plane features.The point,line and plane features in the image are comprehensively considered to make the result of pose estimation more accurate.Experiments show that the proposed algorithm outperforms the state of art on common SLAM benchmarks such as ICL-NUIM and TUM.

关 键 词：点线面特征融合 无漂移旋转 视觉里程计 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]