无人机视觉导航算法

为保证无人机着陆精度和安全性，提出了一种无人机自主着陆视觉导航位姿解算方法。首先对机载相机进行标定，获取相机参数；然后综合考虑地标形状和尺寸、地标角点几何分布和角点数量对位姿估计精度的影响，设计了T型着陆地标形状和尺寸参数，将地标轮廓提取和角点检测算法相结合，得到几何分布好、数量适中的8个角点用于位姿解算，保证了位姿解算精度；为减少LK （Lucas-Kanade）光流法稳定跟踪地标的处理时间，直接将提取的这8个角点作为LK光流法检测和跟踪的输入，保证了算法实时性；最后利用三维空间到二维像平面投影关系对飞行位姿参数进行实时解算。实验结果表明:算法具有较高估计精度，算法平均周期为76.756 ms （约13帧/s）,在速度较低的着陆阶段基本满足自主着陆视觉导航的实时性要求。

Unmanned aerial vehicle vision navigation algorithm

In order to ensure accuracy and security of unmanned aerial vehicle(UAV)landing, a UAV autonomous landing with visual navigation pose parameters calculateion method was proposed. Firstly, the airborne camera was calibrated to get the camera parameters, then the important influence of landmark shape and size, angular point geometry distribution and number of points on pose estimation precision were considered, a T type landing landmark was designed with given size parameters, landmark contour extraction with corner detection algorithm was combined to get eight corners with good geometric distribution and the number was reasonable for pose estimation to guarantee the posture calculation accuracy. To reduce the processing time of Lucas-Kanade(LK) optical flow method tracking landmarks stably, the extracted eight corners were used as LK optical flow method input to detect and track, ensuring real-time performance of the algorithm. Finally, real-time flight pose parameters of UAV through the projection relationship between 3D space and 2D image plane were estimated. The results of simulation experiment show that the algorithm has high precision,and the average period is 76.756 ms (about 13 frames per second). The real-time requirements of visual aided navigation of autonomous landing at low speeds of landing stage is satisfied basically.