摘要
建立了基于运动结构恢复方法和特征点匹配测速技术的无人机巡航测速系统,实现了对河流大范围表面流场的长距离巡航监测.首先建立地面控制点,并向水中投入示踪物体;利用无人机巡航俯视拍摄河流表面,连续采集照片;采用运动恢复结构方法估算各时刻相机的姿态参数,建立目标区域正射投影网格,转换得到正射影图像;再利用加速鲁棒特征算法识别和匹配连续粒子图像中的特征点,得到特征点坐标和位移;进行尺度转化,最终得到流场.该方法成功应用于瑞士苏黎世利马特河表面流场测量,结果表明该方法得到的正射投影图像质量高,流场计算结果合理,与PIV方法相比误差为3.4%,计算速度提升8倍;计算结果与水文站断面实测值吻合良好.
This study presents an airborne image velocimetry technique that combines the structure from motion technique and the feature point tracking method to determine the river’s surface velocity field in a cruising mode.This new velocimetry approach is as follows:firstly,tracker particles are seeded onto water surface and the seeding flow scenery is recorded by a camera mounted to a low-cost quadcopter.Captured image frames are orthorectifed and georeferenced by the transfer relations between camera extrinsic parameters and 3D mesh computed by structure from motion algorithm.Then,scale-invariant keypoints on successive particle images are detected and matched by feature tracking algorithm.The related shifts of feature points in metric space divided by the video frame rate finally give the velocity vector.The technique is applied for measurements at Limmat river(Zurich,Switzerland).The results show that this new airborne image velocimetry gives reasonable surface velocity field with an error of 3.4%and high speed compared to PIV algorithm.A comparison with the velocity profiles measured by a propeller current meter indicates that the proposed airborne velocimetry system is capable of measuring with relatively high accuracy.
作者
曹列凯
Martin Detert
李丹勋
CAO Liekai;Martin Detert;LI Danxun(State Key Laboratory of Hydroscience and Engineering,Tsinghua University,Beijing 100084,China;Laboratory of Hydraulics,Hydrology and Glaciology VAW,ETH Zurich,Zurich 8093,Switzerland)
出处
《应用基础与工程科学学报》
EI
CSCD
北大核心
2020年第6期1271-1280,共10页
Journal of Basic Science and Engineering
基金
国家重点研发计划项目(2016YFC0402308)
国家公派留学基金(201706210226)
关键词
运动恢复结构
无人机
水面流场
特征点
大尺度图像测速
正摄影图像
structure from motion
unmanned aerial vehicles
surface velocity field
feature points
large scale image velocimetry
orthoimage
