基于单目稀疏法多传感器融合移动机器人定位

Multi-sensor Fusion Mobile Robot Localization Based on Monocular Sparse Method

针对移动机器人室内定位过程中,单目视觉难适应光照变化、里程计累计误差导致定位误差较大问题,提出边缘侧多传感器融合的定位方法。以稀疏直接法(半直接法)作为单目视觉的前端,实时单目相机估计位姿,通过惯性传感器恢复尺度输出位置信息,并且获取IMU的加速度以及偏航角、里程计当前速度,通过扩展卡尔曼滤波算法融合3种传感器信息,实现更加精确的定位。在移动机器人侧处理传感器读取的信息,从而减小机器人体积。边缘侧混合式多传感器信息融合使移动机器人在单个传感器失效以及无法人为干预时,也能够精确实时地在多种复杂环境中完成自主定位。

Aiming at the problem that monocular vision is difficult to adapt to light changes during the indoor positioning of mobile robots,and the cumulative error of odometer leads to positioning errors,a positioning method based on edge measurement and multisensor fusion was proposed.Using sparse direct method(semi-direct method)as the front end of monocular vision,real-time monocu⁃lar camera was used to estimate position and attitude,inertial sensor was used to recovery scale output position information,IMU accel⁃eration,yaw angle and current speed of odometer were obtained.The multiple sensor information was fused through extension Kalman filter algorithm for more precise positioning.The information read by the sensors was processed on the mobile robot side to reduce the robot volume.Edge-side hybrid multi-sensor information fusion enables mobile robots to accurately and real-time complete autonomous positioning in a complex environments even when the sensors are reset and cannot be interfered by human intervention.