基于融合BEBLID和改进ORB算法的单目视觉里程计研究
Research on Monocular Odometer Based on BEBLID and ORB Algorithm
余正强 1蒋林 2郭宇飞1
作者信息
- 1. 武汉科技大学冶金装备及其控制教育部重点实验室,武汉 430081
- 2. 武汉科技大学冶金装备及其控制教育部重点实验室,武汉 430081;武汉科技大学机器人与智能系统研究院,武汉 430081
- 折叠
摘要
为防止FAST算法在纹理丰富的环境下提取的特征点过于密集,通过预设条件使特征点稀疏化,便于后续对极约束恢复相机的位姿;为保证相机摄入的图像帧中纹理较少,特征提取时不会丢失图像信息,提出根据特征点数量调整检测方式,使其自适应提取图像特征,增强对复杂环境的适应性;为提高图像特征的匹配精度,通过融入BEBLID使特征点描述符充分表达.经实验验证,面对复杂的场景,改进算法有较强的环境鲁棒性.改进算法的匹配精度要高于ORB算法,在算法耗时方面相较SIFT算法有量级提升.经轨迹测算后,相比于ORB算法,改进算法在相机运动上更贴合真实轨迹,在位姿精度上有较明显的提升.
Abstract
In order to prevent the feature points extracted by FAST algorithm in the environment of rich texture from being too dense,the preset conditions are used to make the feature points sparse,so as to facili-tate the subsequent polar constraint recovery of camera pose.In order to ensure that there are fewer textures in the image frames taken by the camera and no image information will be lost during feature extraction,the detection method is proposed to adjust according to the number of feature points,so that it can extract im-age features adaptively and enhance its adaptability to complex environment.In order to improve the matc-hing accuracy of image features,the feature point descriptor is fully expressed by incorporating BEBLID.Experimental results show that the improved algorithm has strong environmental robustness in the face of complex scenarios.The matching accuracy of the improved algorithm is higher than that of ORB algorithm,and the time consumption of the algorithm is higher than that of SIFT algorithm.After trajectory calcula-tion,compared with ORB algorithm,the improved algorithm is more in line with the real trajectory in cam-era motion,and has obvious improvement in position pose accuracy.
关键词
VSLAM/单目视觉里程计/BEBLID/匹配精度/相机轨迹Key words
VSLAM/monocular odometer/BEBLID/matching accuracy/camera track引用本文复制引用
基金项目
国家自然科学基金(51874217)
国家重点研发计划(2019YFB1310000)
出版年
2024
NETL
NSTL
万方数据