适用于高精度同时定位与地图构建的均衡化亚像素ORB特征提取方法

Uniform distributed subpixel ORB feature extraction method for high-precision SLAM

在视觉同时定位与地图构建问题中,ORB(Oriented FAST and Rotated BRIEF)特征由于其高效、稳定的优点而受到广泛关注。针对ORB特征提取过程中存在的像点量测精度较低、特征聚集现象明显等问题,提出了一种适用于高精度SLAM的均衡化亚像素ORB特征提取方法。分析了精确特征定位的原理,对误差方程进行合理的简化并采用一种基于模板窗口距离的权函数计算方法,大幅降低了计算负担;设计了一种基于四叉树结构的特征均衡化方案,对包含特征的像平面空间进行有限次数的迭代分割,然后选取具有最优响应的特征。试验表明,本文方法进行特征提取的额外计算负担小于2.5 ms,在运行TUM和KITTI数据集时,ORB特征的量测精度分别为0.84和0.62 Pixel,达到亚像素水平,可以降低误差初值,提高光束法平差效率,并能够在满足特征总体分布规律的情况下,显著改善特征聚集的现象,有利于后续问题的稳健、准确求解。

In visual SLAM problems,the ORB feature has drawn much attention because of its high efficiency and stability.To address problems such as the low accuracy of image point measurements and the obvious phenomenon of feature aggregation during ORB feature extraction,a uniform distributed subpixel ORB feature extraction method suitable for high-precision SLAM was proposed.In this study,the principle of precise feature positioning was first analyzed,the error equation was then reasonably simplified,and a weight function calculation method based on template window distance was finally adopted,all of which significantly reduce the algorithm's computational cost.A quadtree-based uniform distribution solution was designed in which the image plane space is segmented with only a limited number of iterations.Features with optimal response are then exported.Experiments show that the additional computational burden of feature extraction for our method is less than 2.5 ms.The measurement accuracy of ORB features is 0.84 and 0.62 pixels on the TUM and KITTI datasets,respectively,reaching the subpixel level.Our method can thus reduce the initial value of errors and increase the efficiency of bundle adjustment.The problem of feature aggregation is effectively solved based on the condition of satisfying the overall distribution of features,which is beneficial to the robust and accurate solution of subsequent problems.