无人飞行器自主降落区识别方法研究

为实现无人飞行器的自主降落,针对降落平坦区域包含特征较少、且分布没规律、形状各异等特点,本文设计了一种基于点云几何特征的快速点云分块和平坦区识别方法。该方法通过无人飞行器上的相机获取二维图像,并使用多视角立体三维重建技术获得场景三维点云,提出了以空间距离作为平滑项、以点在Z方向上的高度作为相似项的三维点云滤波算法对三维点云滤波,设计了基于点云法线和曲率的聚类分块对点云进行区域划分,然后改进RANSAC算法拟合点云平面,筛选出无人飞行器飞经场景的平坦区,并最终确定出无人飞行器的最佳降落区。最后,用本文所设计方法对戈壁人工沟壑、戈壁自然沟壑和小区花园等实拍场景图像进行降落区识别,实测结果显示识别出的区域地形起伏均小于0.125m@m~2,满足无人飞行器降落要求。

Method for identifying the landing area of unmanned aerial vehicle

In order to achieve unmanned aerial vehicle self-landing, considering the characteristics of UAVs (unmanned aerial vehicle) landing areas which are lack of features and irregular distribution in different shapes, an UAV landing area identification method is proposed based on point cloud processing technologies in this paper. Multi-view 3D reconstruction algorithm is used to generate point cloud from images captured by a camera fixed on an UAV. Taking the spatial distance as the smoothing term and the height of the points in the Z direction as the similarity term, a 3D point-cloud bilateral filtering algorithm is proposed in this paper to reduce noise points. A cluster segmentation based on the normal and curvature information of point cloud is designed to segment the point clouds. And an improved RANSAC algorithm for plane fitting is used to identify flat areas. Then, a selection method is proposed to select landing area for UAV. At last, several real scenes' images are used to generate point clouds to test the accuracy of the algorithm. The experimental results show that the fluctuation of the identified area is less than 0.125m@m², which meet the requirements of UAV's landing.