无人机电力巡检视觉避障技术研究

发布时间：2018-05-15 05:38

本文选题：双目立体视觉 + 电力巡检　；参考：《北京理工大学》2016年硕士论文

【摘要】：近年来,全球范围内掀起了多旋翼轻小型无人机商业化的热潮。在高压电力巡检应用中,无人机代替人工监察线路作业可极大提高工作效率。为防止巡检无人机与电力设备间发生碰撞事故,本文研究了基于双目立体视觉的无人机避障算法和远程监控系统。避障算法主要完成长基线双目视觉系统下的障碍物深度信息精确感知,为飞行控制系统提供动作依据。远程监控系统以虚拟现实设备为终端,实时监测机载短基线双目系统获取的巡检场景。论文主要展开了如下工作:(1)双目视觉系统设计与标定。在无人机避障工作中,为获得较精确的距离感知效果,设计了大型标定板与长基线双目视觉系统并进行立体标定。利用该视觉系统对远距离范围内的预设标靶进行视觉测距实验,验证了标定结果的正确性与远距离视觉测距的可靠性。在远程监控系统的设计工作中,制作了类似人眼结构的短基线双目视觉系统,以此作为监控系统的机载前端采集装置。(2)实际巡检场景下的双目测距算法设计与仿真。本文依靠长基线垂直架设的双目视觉系统进行场景深度感知,并对如何获取在以下两种巡检场景捕获同名匹配点进行重点分析。a)电力线场景。鉴于电力线几乎不存在特征点信息,本文首先检测公共视野内的电力线,利用视觉极线与电力线近似正交的性质寻找电力线上的同名匹配点对;b)电力塔场景。区别于雷达、结构光等主动式深度感知方法,本文首先利用基于光流法的全被动式景深感知模型初步感知前景电力塔,然后利用SIFT算法提取塔身上的特征点并匹配,匹配结果即为同名匹配点对。同名匹配点对坐标结合相机标定参数即可计算高压线、塔目标的深度信息。(3)基于双目视觉系统的远程监控系统设计与硬件实现。本文对该系统设计思路和主要功能模块(短基线水平架设双目视觉模块,以FPGA与视频压缩芯片为核心的图像处理模块,无线收发模块,虚拟现实终端)进行详细阐述。通过虚拟现实终端,工作人员可以更真实地远程感知巡检情况。本文的研究工作及其主要成果可以作为视觉避障技术在无人机电力巡检应用中的积极尝试。
[Abstract]:In recent years, there has been a worldwide commercial boom of multi-rotor light and small UAVs. In the application of high voltage power inspection, unmanned aerial vehicle (UAV) can greatly improve the working efficiency instead of manual monitoring line operation. In order to prevent collision between patrol UAV and power equipment, the obstacle avoidance algorithm and remote monitoring system of UAV based on binocular stereo vision are studied in this paper. Obstacle avoidance algorithm mainly realizes the accurate perception of obstacle depth information under the long baseline binocular vision system, and provides the basis for flight control system. The remote monitoring system uses virtual reality equipment as the terminal to monitor the inspection scene acquired by the airborne short baseline binocular system in real time. The main work of this paper is as follows: 1) Design and calibration of binocular vision system. In order to obtain accurate range sensing effect, a large scale calibration board and a long baseline binocular vision system are designed and calibrated. The vision system is used to carry out the visual ranging experiment on the preset target in the long range, which verifies the correctness of the calibration results and the reliability of the distance vision ranging. In the design of remote monitoring system, a short-baseline binocular vision system similar to the human eye structure is developed, which is used as the airborne front-end acquisition device of the monitoring system, and the Binocular ranging algorithm is designed and simulated under the actual inspection scene. In this paper, the binocular vision system with long baseline vertical erection is used for depth perception, and how to obtain the matching point of the same name in the following two kinds of patrol scenes is analyzed. A) Power line scene is analyzed. In view of the fact that there is almost no characteristic point information in the power line, this paper first detects the power line in the common field of vision, and uses the property of the approximate orthogonality between the visual pole line and the power line to find the scene of the power tower with the same name matching point on the power line. Different from active depth sensing methods such as radar and structured light, this paper first uses the passive depth sensing model based on optical flow method to perceive the foreground electric tower, and then uses SIFT algorithm to extract and match the feature points on the tower body. The matching result is a pair of matching points with the same name. The design and hardware implementation of remote monitoring system based on binocular vision system can be used to calculate the depth information of high voltage line and tower target based on matching point pair coordinate and camera calibration parameters. In this paper, the design idea and main function modules of the system are described in detail, including the short baseline horizontal binocular vision module, the image processing module with FPGA and video compression chip as the core, the wireless transceiver module, and the virtual reality terminal. Through the virtual reality terminal, the staff can more realistically perceive the inspection situation remotely. The research work and its main achievements in this paper can be used as visual obstacle avoidance technology in the UAV electric power inspection application.
【学位授予单位】：北京理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TM755

【参考文献】