一种可提高室内视觉SLAM的结构线特征选择方法  

A Structure Line Feature Selection Method for Improving Indoor Visual SLAM

作　　者：姜可 詹总谦[1] 王鑫[1] JIANG Ke;ZHAN Zongqian;WANG Xin(School of Geodesy and Geomatics,Wuhan University,Wuhan 430079,China)

机构地区：[1]武汉大学测绘学院,湖北武汉430079

出　　处：《测绘地理信息》2024年第1期91-95,共5页Journal of Geomatics

基　　金：国家自然科学基金(61871295,42301507)。

摘　　要：室内场景由大量的平面和直线构成,且经常存在纹理单一而少的情况,这会使得产生的点特征稀少且分布不均匀,从而导致传统的以点特征为观测值的视觉同步定位与建图(simultaneous localization and mapping,SLAM)容易产生不准确的定位结果。本文针对室内场景,在基于点特征的视觉SLAM基础上,引入线特征作为额外的观测值。针对与运动方向平行的线特征对运动估计的几何约束较弱的不足,提出一种能从线特征中筛选出与运动方向不平行的线特征的方法,选择与运动方向不平行的结构线特征参与平差优化,提高室内视觉SLAM的定位精度。实验表明,在公开数据集上,使用本文的方法后,定位精度能在基本不影响实时性的条件下提高15%左右。The indoor scene is typically with a large number of planes and lines,which often contains only few and simple texture,this can result in rare and unevenly distributed point features,therefore,the traditional visual slam based on point features is easy to obtain less accurate pose estimation.To cope with the indoor cases,based on the visual SLAM using point features,line features are used as additional observa⁃tions.However,line features which are parallel to motion di⁃rection can only provide relatively week geometric constraint during pose estimation,thus,this paper present a line feature selection method,and select the structural line features that have a greater effect on pose estimation for subsequent bundle adjustment optimization.Experimental results on public data sets show that the localization precision is improved around 15%after using the proposed method without affecting the real-time performance.

关 键 词：结构线 定向 灭点 室内视觉SLAM 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]