面向室内视觉定位的点特征提取算法比较

Comparative research on point feature extraction algorithms for indoor visual positioning

针对室内定位中特征提取较为困难的问题,对目前已有的室内点特征提取算法进行了评估。比较尺度不变特征变换算法(SIFT)、快速鲁棒性特征算法(SURF)、旋转不变特征算法(ORB)、二进制尺度不变特征算法(BRISK)及快速视网膜特征算法(FREAK)的原理和特点;并自制室内影像数据集,对不同点特征提取算法在不同条件(光照变换、模糊变换、旋转变换、视角变换、尺度变换)下特征提取的计算效率以及提取质量进行了比较。实验结果表明:FREAK算法在5种变换条件下,都具有较高的提取质量且其计算效率也较高,更适合室内视觉定位中的特征提取环节;SURF算法可以获得较好的提取质量,但是计算效率低;ORB算法效率最高,但是其在特征提取性能上有所不足。

Aiming at the difficult problem of feature extraction in indoor positioning,this paper proposes to evaluate the existing indoor point feature extraction algorithms.Compare the principles and characteristics of five point extraction algorithms(SIFT,SURF,BRISK,ORB,FREAK);self-made indoor image data sets,and compared the time cost and extraction quality of different point extraction algorithms under different conditions(lighting,blur,rotation,perspective,scale change).After comparative analysis,it is found that:1)FREAK algorithm has a higher number of correct matches under the five conditions and its calculation efficiency is also higher,so it is more suitable for the feature extraction link in indoor visual positioning;2)SIFT algorithm can obtain better matching quality,but the matching speed is slow;3)ORB algorithm is the most efficient,but it is insufficient in extraction quality.