基于改进ORB算法的单目视觉里程计研究

Research on Monocular Visual Odometer Based on Improved ORB Algorithms

针对AGV视觉里程计累积漂移产生的误差问题,提出一种鲁棒性较强的单目视觉定位及优化方法。采用改进的ORB算法对特征点进行快速高效的提取,根据暴力匹配算法和快速最近邻算法以及渐进采样一致性算法,进行特征点匹配及剔除误匹配。通过构建最小二乘问题,采用改良的列文伯格—马夸尔特方法以及非线性优化库g2o对AGV位姿进行优化。实验验证得出,提出方法能够有效缩短特征提取时间,提高AGV定位精度,增强视觉里程计定位鲁棒性。

Aiming at the error caused by the cumulative drift of AGV visual odometer, a robust single-eye visual positioning and optimization method is proposed. The improved ORB algorithm is used to extract feature points quickly and efficiently. According to the brute force matching algorithm, fast nearest neighbor algorithm and progressive sampling consistency algorithm, feature point matching and culling mismatch are performed. By constructing the least squares problem, the improved Levinberg-Marquart method and the nonlinear optimization library g2 o are used to optimize the AGV pose. The experimental results show that the proposed method can effectively shorten the feature extraction time, improve the accuracy of AGV positioning, and enhance the robustness of visual odometer positioning.