基于稳健特征点的立体视觉测程法
发布时间：2018-11-01 17:48
【摘要】：提出一种基于稳健特征点的立体视觉测程法完成机器人自主高精度定位.从可重复性、精确性和效率3个方面比较多种局部不变特征算法性能,采用稳健特征算法AKAZE(AcceleratedKAZE)提取特征点.提出了一个稳定的特征点匹配框架和改进的随机抽样一致性算法(Random Sample Consensus,RANSAC)去除外点,使文中的视觉测程法可以应用于动态环境中.基于几何约束的分步自运动估计可提供相机运动的精确信息.将提出的方法在KITTI(Karlsruhe Institute of Technology and Toyota Technological Institute)数据集上和复杂校园环境中所采集的立体视觉数据集上进行测试,与经典立体视觉测程方法比较,文中的方法更好地抑制了误差累计,运动估计结果满足实时高精度定位系统需求.
[Abstract]:A stereo vision method based on robust feature points is proposed to achieve autonomous and high precision localization of robot. The performance of several local invariant feature algorithms is compared from the aspects of repeatability, accuracy and efficiency. The robust feature algorithm AKAZE (AcceleratedKAZE) is used to extract feature points. A stable feature point matching framework and an improved random sampling consistency algorithm (Random Sample Consensus,RANSAC) are proposed to remove the external points, so that the visual log method in this paper can be applied to the dynamic environment. Step-by-step self-motion estimation based on geometric constraints can provide accurate information of camera motion. The proposed method is tested on the KITTI (Karlsruhe Institute of Technology and Toyota Technological Institute) dataset and the stereo vision data set collected in the complex campus environment. Compared with the classical stereo vision range method, the method in this paper can better restrain the error accumulation. The result of motion estimation meets the requirement of real-time high-precision positioning system.
【作者单位】： 长安大学信息工程学院;
【基金】：国家自然科学基金(51278058) 国家高等学校学科创新引智计划项目(B14043) 中央高校基金项目(310824151033,310824165024,310824173101,310824173307) 交通部基础应用项目基金(2015319812060)资助
【分类号】：TP391.41
，
本文编号：2304645
[Abstract]:A stereo vision method based on robust feature points is proposed to achieve autonomous and high precision localization of robot. The performance of several local invariant feature algorithms is compared from the aspects of repeatability, accuracy and efficiency. The robust feature algorithm AKAZE (AcceleratedKAZE) is used to extract feature points. A stable feature point matching framework and an improved random sampling consistency algorithm (Random Sample Consensus,RANSAC) are proposed to remove the external points, so that the visual log method in this paper can be applied to the dynamic environment. Step-by-step self-motion estimation based on geometric constraints can provide accurate information of camera motion. The proposed method is tested on the KITTI (Karlsruhe Institute of Technology and Toyota Technological Institute) dataset and the stereo vision data set collected in the complex campus environment. Compared with the classical stereo vision range method, the method in this paper can better restrain the error accumulation. The result of motion estimation meets the requirement of real-time high-precision positioning system.
【作者单位】： 长安大学信息工程学院;
【基金】：国家自然科学基金(51278058) 国家高等学校学科创新引智计划项目(B14043) 中央高校基金项目(310824151033,310824165024,310824173101,310824173307) 交通部基础应用项目基金(2015319812060)资助
【分类号】：TP391.41
，
本文编号：2304645
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2304645.html
