视觉与惯性传感器融合的室内导航算法研究  

Research on Indoor Navigation Algorithm Based on Multi-sensor Fusion

作　　者：杨志芳[1] 潘忠运 YANG Zhi-fang;PAN Zhong-yun(School of Electrical and Information,Wuhan Institute of Technology,Wuhan 430200,China)

机构地区：[1]武汉工程大学电气信息学院,武汉430200

出　　处：《自动化与仪表》2021年第10期44-48,68,共6页Automation & Instrumentation

摘　　要：目前大多数室内移动机器人使用的是视觉传感器的导航算法,其精度低,鲁棒性差,对于快速移动和环境纹理特征不明显的情况,机器人位姿估计的精度会骤然下降。于是该文提出了在ORB-SLAM2算法的基础上,使用双目相机与IMU传感器紧耦合的方式进行状态估计。IMU响应速度快、采集数据不受外界影响、在快速运动和低纹理环境中可以提供较好的估计。此外,在特征点匹配后进行了动态特征点滤除,使其在动态环境中也有较好的表现。在数据集下进行实验测试,结果表明,改进后的算法在上述情况下明显提高了系统的定位精度。At present,most indoor mobile robots use the navigation algorithm of visual sensors.Its accuracy is low and its robustness is poor.For fast moving and environment texture features are not obvious,the accuracy of robot pose estimation will drop sharply.Therefore,based on ORB-SLAM2 algorithm,this paper proposes a tight coupling method of binocular camera and IMU sensor for state estimation.The IMU has fast response time,free data acquisition from external influences,and can provide better estimates in fast moving and low-texture environments.In addition,dynamic feature point filtering is performed after feature point matching,which makes it perform well in dynamic environment.Experimental results under the data set show that the improved algorithm significantly improves the positioning accuracy of the system.

关 键 词：视觉传感器 ORB-SLAM2算法 IMU传感器 紧耦合 动态特征点滤除 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]