考虑特征误匹配的双目视觉里程计

Stereo Visual Odometry Eliminating Mismatched Feature Points

提出一种针对误匹配情况下的双目视觉里程计,以适应复杂环境下的位姿估计。系统在特征匹配的位姿估计模型基础上,考虑复杂环境下错误匹配的情况,对于前后两帧图像提取的特征点,通过最近邻匹配后,建立基于网格的运动统计模型将错误匹配剔除,将置信度较高的匹配对进行跟踪处理,进而提高位姿估计精度。采用局部光束平差法优化位姿估计,达到累积误差最小化。在地图点的初始化过程中,考虑左右视图匹配质量较高的特征点三角化,提高系统的鲁棒性。数据集实验表明,该改进有效地提高了系统复杂环境下的定位精度。

A method of stereo visual odometry for mismatching is proposed to pose estimation in complex environment in this paper.Based on the mathematical model of feature correspondence,the system considers the wrong matching points in complex environment.The experiments based on the datasets show that the method effectively improves the accuracy of the system localization.