无人机基于视觉引导的移动平台追踪与降落

Tracking and landing of vision-based unmanned aerial vehicles on moving vehicle

针对当前可远距离实现纯视觉引导无人机降落的方案中合作目标尺寸过大的问题,提出一种三维与二维标志组合的合作目标设计思路:通过深度学习加手工特征检测三维标志以进行远距离定位与追踪,接近后切换检测二维标志以进行降落;引入三维合作目标后,在不降低定位精度的前提下,1.5 m高度的可检测距离增加了2.54倍,且可大角度倾斜检测,合作目标尺寸也大幅减小。实验结果表明,远距离阶段平均定位误差为0.05m,近距离阶段在移动平台上的降落点平均误差为0.03m,说明该小尺寸合作目标的视觉定位精度与降落精度很高,可以追踪并降落在移动中的超小尺寸降落平台上。

Aiming at the problem that the size of the cooperative target is often too large in the current schemes of the long-range purely visual-guided drone landing of unmanned aerial vehicles(UAV),the paper proposed a design thought of cooperative targets composed of three-dimensional(3D)and two-dimensional(2D)logos:3D signs were detected through deep learning and manual features for long-distance positioning and tracking,and it was switched to detect 2D signs for landing on the target autonomously after approaching;by introducing the 3D cooperative targets,the detection distance at 1.5 m height was increased by 2.54 times,detecting tilt targets at a large angle was carried out,and the target size was significantly reduced,without decreasing the positioning accuracy.Experimental result showed that the average long-range positioning error would be 0.05 m,and the average landing error on a moving platform in the short range would be 0.03 m,indicating that the small-size cooperative target could have high visual positioning accuracy and landing accuracy,thus making it possible to track and land on the moving ultra small-sized platforms accurately.