弱纹理环境下的视觉惯性里程计优化方法

Optimization Method of Visual Inertial Odometry in Low-Texture Environment

基于视觉的机器人自主定位与导航系统都是利用点特征进行工作,但在弱纹理环境中,无法提取充足的点特征,系统的精度与鲁棒性就会下降。为此,提出了一种弱纹理环境下的视觉惯性里程计优化方法。将点线特征融合,相对于点特征,线特征在弱纹理环境下有较强的鲁棒性,可提供更多的环境几何结构信息,有利于三维地图的构建。为了提高精度,使用紧耦合的方式将相机与IMU采集到的数据融合。利用滑动窗口算法,将IMU预积分后的误差与点线重投影误差,在滑动窗口中用最小化误差函数的方式进行优化。通过EuRoc数据集,将优化后的系统与VINS-Mono系统进行比较。经测试,不同场景中的几组数据的绝对轨迹误差都很小,且均方根误差下降3%左右。验证了算法的精度与鲁棒性。

The autonomous localization and navigation system of robot based on vision works by using point feature,but in low-texture environments,if sufficient point features cannot be extracted,the accuracy and robustness of the system will decline.To solve this problem,a optimization method of visual inertial odometry in low-texture environment is proposed.The point and line features are fused,compared with the point feature,the line feature has strong robustness in the low-texture environment and can also provide more information about the geometric structure of the environment,which is conducive to the construction of 3d map.In order to improve accuracy,the data collected by the camera and the IMU is fused using a tightly coupled method.The sliding window algorithm is used to optimize the IMU pre-integral error and point-line reprojection error by minimizing error function in the sliding window.Through the EuRoc dataset,the optimized system is compared with the VINS-Mono system.After testing,the absolute trajectory error of several groups of data in different scenarios is very small,and the root mean square error decreases by about 3%.The accuracy and robustness of the algorithm are verified.