一种基于LSD改进的室内直线特征匹配算法

An Improved Indoor Straight Line Feature Matching Algorithm Based on LSD

由于室内长廊等弱纹理场景下点特征匹配效果差,线特征匹配成为了特征匹配研究领域新的热点。现有线特征匹配算法如LSD在弱纹理场景下表现不佳,提取了大量无效特征如重复描述的线特征、断连的线特征、长度过短无法参与匹配的线特征。为此,本文针对LSD进行研究改进,提出了Light-LSD,改进了特征重复、特征断连、特征过短等问题。实验结果表明,在弱纹理场景下Light-LSD可以剔除大量无效特征,使总特征数量减少60%,提高算法整体运行效率。Light-LSD对于SLAM等依赖特征匹配等技术有很大研究价值。

Due to the poor effect of point feature matching in weak texture scenes such as indoor corridors,line feature matching has become a new hot spot in the field of feature matching research.Existing line feature matching algorithms such as LSD perform poorly in weak texture scenes,and extract a large number of invalid features such as repeated description line features,disconnected line features,and line features that are too short to participate in matching.For this reason,the research and improvement of LSD are carried out,and Light-LSD is proposed to improve the problems of feature duplication,feature disconnection,and feature too short.Experimental results show that in a weak texture scene,Light-LSD can eliminate a large number of invalid features,reduce the total number of features by 60%,and improve the overall operating efficiency of the algorithm.Light-LSD has great research value for SLAM and other technologies that rely on feature matching.