基于重叠区域融合的多无人机实时视觉稠密地图构建

Multi-UAV Real-time Visual Dense Mapping Based on Overlapping Region Fusion

针对提高多无人机平台执行协同环境感知任务时的精度与效率问题,开展了多无人机视觉稠密建图与地图融合技术的研究。首先,针对建图误差和传感器误差导致生成无效地图点和局部重影的问题,提出了一种基于视觉的稠密点云地图构建算法,该算法在对点云进行滤波之后通过基于KD-Tree的广义最近邻迭代和体素滤波结合完成稠密点云地图的构建。随后,提出了基于重叠区域的稠密地图融合算法,算法进行增长子地图的融合判断,通过获取重叠区域并查找正确的匹配点对,实现稠密地图的融合。最后,搭建了一个双无人机验证平台,并通过在室外场景下进行协同环境感知验证了所提出的方法可以保证地图构建误差在5%以内的同时有效降低地图点的数量,并且实现了精确的地图融合。

To address the issue of improving accuracy and efficiency in multi-UAV platforms for collaborative environmental sensing tasks,a study on visual dense mapping and map fusion technologies for multi-UAV was conducted.Firstly,to tackle the problems of invalid map points and local ghosting caused by mapping errors and sensor inaccuracies,a visual dense point cloud mapping algorithm was proposed.This algorithm constructs dense point cloud maps by filtering the point clouds and employing a combination of generalized nearest neighbor iteration based on KD-Tree and voxel filtering.Subsequently,a dense map fusion algorithm based on overlapping regions was introduced.This algorithm determines the fusion of growing sub maps by obtaining overlapping regions and identifying correct matching points,thereby achieving dense map fusion.Finally,a dual-UAV validation platform was built.Outdoor scene experiments demonstrated that the proposed method can maintain mapping errors within 5%,effectively reduce the number of map points,and achieve precise map fusion.