基于单目视觉与惯导融合的无人机位姿估计  

Pose estimation of UAV based on monocular vision and inertial navigation

作　　者：熊敏君 卢惠民[1] 熊丹 肖军浩[1] 吕鸣[1] 

机构地区：[1]国防科技大学机电工程与自动化学院,长沙410073

出　　处：《计算机应用》2017年第A02期127-133,共7页journal of Computer Applications

基　　金：国家自然科学基金资助项目(61403409;61503401);中国博士后科学基金资助项目(2014M562648)

摘　　要：针对无卫星信号条件下仅利用惯性导航进行无人机(UAV)位姿估计精度低的问题,提出了一种基于扩展卡尔曼滤波(EKF)的单目视觉/惯导组合位姿估计算法。首先,借助超声波传感器,基于最小二乘法获取单目视觉的绝对尺度信息,从而计算得到视觉位姿估计结果;其次根据惯性测量单元(IMU)的运动学模型建立系统方程,进行惯导解算,并推导误差状态方程对系统误差状态建模;最后,将视觉位姿估计结果作为观测量,基于扩展卡尔曼滤波得到系统误差状态的最优估计值,从而对惯导解算的结果进行修正。使用该算法在室外150 m范围场景开展实验研究,位置估计的均方根误差(RMSE)低于0.995 m,航向角估计的RMSE为2.235°,水平姿态角估计的RMSE低于1.915°。To deal with the low accuracy of pose estimation for Unmanned Aerial Vehicles（ UAVs） when only using inertial navigation without available Global Navigation Satellite System（ GNSS）, this paper proposed an integrated pose estimation algorithm combining monocular vision and Inertial Measurement Unit（ IMU） based on Extended Kalman Filter（ EKF）. Firstly, the absolute scale in monocular vision system was obtained by using ultrasonic sensor and least square method, so the pose estimation results could be obtained using monocular vision; secondly, the system equation was built according to the kinematic model of the IMU to realize inertial navigation calculation, and the error state equation was derived to model the system＇ s error states; finally, the optimal estimation of the system＇ s error states was completed based on EKF using the pose estimation results of monocular vision as the measurement information, so the pose estimation results obtained by inertial navigation were corrected. Experiments were conducted in large scale environments with the range of 150 meters.The Root Mean Square Error（ RMSE） of position estimation is less than 0. 995 meters, the RMSE of heading estimation is2. 235 degrees, and the RMSE of horizontal attitude is less than 1. 915 degrees.

关 键 词：单目视觉 惯性导航 位姿估计 尺度估计 扩展卡尔曼滤波 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]