二进制特征图像算法性能分析

Performance Analysis of Image Algorithm Based on Binary Feature

为研究几种常用二进制特征提取算法的实时性及复杂环境下的鲁棒性,对ORB、SIFT、SURF、BRISK四种基于二进制特征的主流算法,使用Euroc数据集和真实场景图像对进行匹配和统计,对比分析其实时性。选用Mikolajczyk数据集,在图像旋转变换、模糊变换、光照变换、视角变换场景下对算法的鲁棒性进行比较。采用的评价指标为算法耗时和特征匹配数目。实验结果表明,ORB能够在达到足够匹配图像对的同时运行时间在24.6ms以内;SURF在模糊变换、光照变换、视角变换场景下,当环境变化量达到最大时匹配数不低于123对;ORB、BRISK算法在光照变化情况下能获取足够的匹配对数。在实时性方面,ORB算法在满足足够匹配数情况下运行速度最快,实时性最高;SURF在模糊变换、光照变换、视角变换场景下有良好的鲁棒性,SURF算法鲁棒性最强。

Aiming to study the real-time performance and robustness of several common binary feature extraction algorithms in com⁃plex environment,four mainstream binary feature-based algorithms,ORB,SIFT,SURF and BRISK,were matched and statistically compared with the image pairs in the Euroc dataset and real scene to analyze their real-time performance.Mikolajczyk's data set is se⁃lected to compare the robustness of the algorithm in the scene of image rotation transformation,blur transformation,illumination trans⁃formation and perspective transformation.The evaluation index are the algorithm running time and the number of matched feature points.The experimental results show that ORB can achieve enough matched number,and the running time of the algorithm is not more than 24.6ms;SURF can achieve the maximum matching number of 123 pairs in the scene of blur transformation,illumination transfor⁃mation and perspective transformation;ORB and BRISK can obtain enough matched number in the scene of illumination change.In terms of real-time performance,ORB algorithm has the fastest running speed and the highest real-time performance under the condi⁃tion of enough matches;SURF has good robustness in the scene of blur transformation,illumination transformation and perspective transformation,and SURF algorithm has the strongest robustness.