特征点法与光流跟踪融合的视觉里程计算法
A Visual Odometry Algorithm Based on Feature Point Method and Optical Flow Tracking
宋欣 1梁绍扬 1张倡铖 1娄伟 1杨磊1
作者信息
- 1. 天津农学院工程技术学院,天津 300392
- 折叠
摘要
针对视觉里程计中特征点法因计算量大而导致实时效果不好的问题,提出了一种特征点法与光流跟踪融合的算法.首先,通过度量运动大小的视差阈值和地图点的跟踪衰减系数作为筛选策略选取关键帧;其次,仅对关键帧中的特征点计算描述符,而对于非关键帧中的特征点则是通过基于图像金字塔的稀疏光流法建立起关键点间的对应关系,从而降低计算量;最后,通过最小化重投影误差得到当前帧的位姿估计.通过实验测试证明,该算法的定位精度与ORBSLAM2 相当,但运行速度可提高70%以上,实时性明显优于ORBSLAM2 算法.
Abstract
Aiming at the problem that the feature point method in visual odometry has a large amount of calculation and leads to poor real-time effect,a fusion algorithm of feature point method and optical flow tracking is proposed.Firstly,the keyframes are selected by the parallax threshold that measures the move-ment and by the tracking attenuation coefficient of the map points as the screening strategy.Then,the de-scriptors are only calculated for the feature points in the key frames.And for the feature points in the non-key frames,the correspondence between the key points is established by the sparse optical flow method based on the image pyramid.In this condition,the amount of calculation is reduced.Finally,the pose esti-mation of the current frame is obtained by minimizing the reprojection error.It is showed that the positio-ning accuracy of the proposed algorithm is comparable to the ORBSLAM2 algorithm,and the running speed can be increased by more than 70%,meanwhile the real-time performance is obviously better than the ORBSLAM2 algorithm in the experimental test.
关键词
视觉里程计/特征点法/光流跟踪/关键帧/位姿估计Key words
visual odometry/feature point method/optical flow tracking/key frame/pose estimation引用本文复制引用
出版年
2024
NETL
NSTL
万方数据