基于2点RANSAC的无人直升机单目视觉SLAM

1点随机抽样一致性(RANSAC)算法是一种准确度高、计算量小的数据关联算法,但是其在摄像机多个轴上的角速度都快速变化时会失效,用在以无人直升机为载体的单目视觉同步定位与地图构建(SLAM)上存在滤波发散的风险.针对该问题,提出2点RANSAC算法,结合EKF运动模型的先验信息,用只抽样2个匹配点的RANSAC去除野点.在微小型无人直升机平台上进行了基于2点RANSAC算法的单目视觉SLAM实验,实验结果表明2点RANSAC算法工作可靠,SLAM的位姿估计精度可以达到自主飞行需要.

Monocular Visual SLAM of Unmanned Helicopter Based on 2-point RANSAC

The 1-point random sample consensus(RANSAC) algorithm is a data association algorithm with high accuracy and low compaction cost.However,it fails when angular velocities around multiple axes of the camera change quickly,and causes the risk of filter divergence when applied to the monocular visual simultaneous localization and mapping(SLAM) of unmanned helicopter.For this problem,2-point RANSAC algorithm is proposed,which incorporates a priori information from the EKF(extended Kalman filter) motion model,and uses RANSAC,in which only 2 matched points are used for sampling,to remove the outliers.Monocular visual SLAM based on 2-point RANSAC algorithm is performed on a mini unmanned helicopter(MUH) platform.The field-experiment results show that 2-point RANSAC algorithm works reliably, and the SLAM’s pose estimation is precise enough for autonomous flight.