基于机器视觉的水域远程监测应用系统方案设计  

Design of remote monitoring application system for water based on machine vision

作　　者：严成 何宁[1] 庞维庆 邓德迎 YAN Cheng;HE Ning;PANG Weiqing;DENG Deying(School of Information and Communication,Guilin University of Electronic Technology of China,Guilin 541004,China;Guangxi Colleges and Universities Key Laboratory of Unmanned Aerial Vehicle(UAV)Telemetry,Guilin 541004,China)

机构地区：[1]桂林电子科技大学信息与通信学院,广西桂林541004 [2]广西高校无人机遥测重点实验室,广西桂林541004

出　　处：《测绘通报》2019年第8期149-153,共5页Bulletin of Surveying and Mapping

基　　金：广西高校无人机重点实验室主任基金(WRJ2016KF02);桂林电子科技大学研究生教育创新项目(2018YJCX23)

摘　　要：针对水面污染区域的识别及快速测量,通过电子测量手段和图像甄别方法,提出了一种高精度的水面污染面积测量方案,以满足水域管理和监测盲点污染需求。根据机器视觉的单目视觉测距原理,采用无人机航拍为该测量过程提供相应参数及污染区域方位信息。系统通过图像分割与堆叠技术对拍摄图片中油污染区域特征进行人工圈定边界识别,利用适用于该面积测量的视觉图像处理方法,设计了水面污染面积测量的人机交互,实现对污染区域面积进行实时或后期测量。试验结果表明,通过无人机实际飞控测量及软件处理,为水域管理提供有效依据,模拟水面污染面积测量精度为10-3 m^2,测量相对误差低于5%,垂直拍摄平均相对误差低于1.3%。In view of the recognition and rapid measurement of the water surface pollution area, a high-precision measurement scheme of the pollution area of the water surface is proposed to meet the pollution demand of the sea area management and the monitoring of the blind spot, by means of electronic measurement and image discrimination. According to the ranging principle of monocular vision based on machine vision, the UAV aerial photography is used to provide the corresponding parameters and the azimuth information of the polluted area for the measurement model. The image segmentation and stacking technique is used to identify the characteristics of the oil pollution area by artificial delineation boundary. Using the visual image processing method which is suitable for the area measurement, a human-computer interaction for measuring the pollution area of the water surface is designed to realize the real-time or later measurement. The experimental results show that the actual flight control measurement and software processing of UAV provide effective basis for water management. the polluted area is simulated, the accuracy of polluted area measurement is 10-3 square meters, the measurement relative error is less than 5%, and the average relative error of vertical shooting is less than 1.3%.

关 键 词：机器视觉 面积测量 图像处理 人机交互 无人机飞控 

分 类 号：P258[天文地球—测绘科学与技术]