基于视觉惯性融合的井下无人车定位方法研究

Research on Localization Method of Underground Unmanned Vehicle Based on the Fusion of Visual Cameras and Inertial Sensors

同步定位与建图(SLAM)技术是矿山井下作业车辆实现复杂环境下自主运行的关键技术之一。基于ORB-SLAM3的算法提出了一种视觉相机和惯性传感器融合的全局定位方法,通过地图点管理机制简化ORB-SLAM3算法中的特征点管理,提高了特征点匹配的准确性。试验结果显示,融合定位新方法在公开数据集的5个序列中,单目、双目以及双目+IMU的模式下平均分别提升了50.5%、8.89%和77.46%,比上一代ORB-SLAM算法更具优势。在矿山井下实地测试中,融合定位新方法在最优模式下实现了更高的全局轨迹精度,全局轨迹误差仅0.4 m,在实际应用中不仅具有更优的性能,而且部署成本更低,有很大的潜力。

The simultaneous localization and mapping(SLAM)technology is one of the key technologies for autonomous operation of mining underground vehicles in complex environments.A global localization method based on the fusion of visual cameras and inertial sensors was proposed by the algorithm of ORB-SLAM3.A map point management mechanism was used to simplify the feature point management in the ORB-SLAM3 algorithm,so as to improve the accuracy of feature point matching.The experimental results show that the new fusion localization method improves the performance by an average of 50.5%,8.89%,and 77.46%in five sequences of public datasets under the modes of single-eye,double-eye,and double-eye+IMU,which is more advantageous than the previous generation of ORB-SLAM algorithm.In the field test of mining underground,the new fusion localization method achieves a higher global trajectory accuracy in the optimal mode,with a global trajectory error of only 0.4 m.In practical application,it not only has better performance,but also has lower deployment cost and great potential.