基于单目视觉的室内微型飞行器位姿估计与环境构建

针对微型飞行器(Micro air vehicle,MAV)在室内飞行过程中无法获得GPS信号,而微型惯性单元(Inertial measurement unit,IMU)的陀螺仪和加速度计随机漂移误差较大,提出一种利用单目视觉估计微型飞行器位姿并构建室内环境的方法。在机载单目摄像机拍摄的序列图像中引入一种基于生物视觉的方法获得匹配特征点,并由五点算法获得帧间摄像机运动参数和特征点位置参数的初始解;利用平面关系将特征点的位置信息由三维降低到二维,给出一种局部优化方法求解摄像机运动参数和特征点位置参数的最大似然估计,提高位姿估计和环境构建的精度。最后通过扩展卡尔曼滤波方法融合IMU传感器和单目视觉测量信息解算出微型飞行器的位姿。实验结果表明,该方法能够实时可靠地估计微型飞行器的位置和姿态,构建的环境信息满足导航需求,适用于微型飞行器室内环境中的导航控制。

Monocular Vision Based Motion Estimation of Indoor Micro Air Vehicles and Structure Recovery

Micro air vehicles(MAVs) need reliable attitude and position information in indoor environment.The measurements of onboard inertial measurement unit (IMU) sensors such as gyros and accelarometers are corrupted by large accumulated errors,and GPS signal is unavailable in such situation.Therefore,a monocular vision based indoor MAV motion estimation and structure recovery method is presented.Firstly,the features are tracked by biological vision based matching algorithm through the image sequence,and the motion of camra is estimated by the five-point algorithm.In the indoor enviroment, the planar relationship is used to reduce the feature point dimentions from three to two.Then,these parameters are optimized by an local strategy to improve the motion estimation and structure recovery accuracy.The measurements of IMU sensors and vision module are fused with extended Kalman fileter.The attitude and position information of MAVs is estimated.The experiment shows that the method can reliably estimate the indoor motion of MAV in real-time,and the recovered enviroment information can be used for navigation of MAVs.