基于视觉的无人机大范围室外道路检测及拓扑地图构建  

Large-scale Outdoor Road Detection and Topological Map Building for Vision-based UAVs

作　　者：王玉茜 张雪涛[2] 闫飞 庄严[1,2] WANG Yuqian;ZHANG Xuetao;YAN Fei;ZHUANG Yan(School of Control Science and Engineering,Dalian University of Technology,Dalian 116024,China;School of Artificial Intelligence,Dalian University of Technology,Dalian 116024,China)

机构地区：[1]大连理工大学控制科学与工程学院,大连116024 [2]大连理工大学人工智能学院,大连116024

出　　处：《无人系统技术》2021年第4期30-39,共10页Unmanned Systems Technology

基　　金：国家自然科学基金(U1913201、61973049)。

摘　　要：空地协作中空中机器人对室外可行道路的检测与建图,有利于提升地面机器人的导航能力。为了实现上述功能,提出了一种新的基于道路分割、图像拼接、骨架提取的无人机大范围室外道路检测及拓扑地图构建方法。首先,为了解决无人机基于航拍图像的道路识别问题,通过D-LinkNet网络完成对无人机航拍图像中道路的分割,并在保证分割准确率的情况下提升了分割的实时性。其次,考虑到无人机飞行高度和相机视角有限的问题,设计了基于GPU加速的ORB图像拼接算法,从而实现了室外大范围场景的获取。然后,为了完成道路拓扑关系的建立,借鉴并行二值图像细化算法,设计了用于道路拓扑化的骨架提取算法,进而得到了道路拓扑地图。最后,所提方法在马萨诸塞州道路数据集上进行测试,实验结果验证了该方法在道路分割、拓扑地图构建方面具有较高的准确性,且可满足实时性要求。所建拓扑地图有利于地面机器人导航能力的提升。The detection and mapping of outdoor feasible roads by aerial robots in air-ground cooperation is conducive to improve the navigation capabilities of ground robots.In order to achieve the above functions,a new method of UAV large-scale outdoor road detection and topology map construction based on road segmentation,image stitching,and skeleton extraction is proposed.Firstly,in order to solve the problem of road recognition in UAV’s aerial images,D-LinkNet is used to complete the segmentation of roads in UAV aerial images,which improves the real-time performance of segmentation by ensuring the accuracy.Secondly,considering the limitation of UAV’s flight altitude and camera view angle,an ORB image stitching algorithm based on Graphics Processing Unit(GPU)acceleration is designed to achieve the capture of a large outdoor scene.After that,in order to complete the road topology,a skeleton extraction algorithm for road topology is designed based on the parallel binary image thinning algorithm,and the topological representation of the road is obtained.Finally,the proposed method is tested on the Massachusetts road dataset,and the experimental results show that the method has high accuracy in road segmentation and topological map construction,and can meet the real-time requirements.The topology map is helpful to improve the navigation capabilities of ground robots.

关 键 词：空地协作 拓扑地图 D-LinkNet 道路分割 图像拼接 骨架提取 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]