SIA OpenIR  > 机器人学研究室
旋翼飞行机器人电力巡线中的在线目标识别与跟踪方法研究
其他题名Research on Online Target Recognizing and Tracking Methods in Power Line Inspection with Rotor Flying Robot
曹蔚然1,2
导师王天然 ; 韩建达
分类号TP242
关键词无人机电力线巡线 图像增强 线状目标检测 目标跟踪 视觉实验系统
索取号TP242/C23/2015
页数119页
学位专业机械电子工程
学位名称博士
2015-09-02
学位授予单位中国科学院沈阳自动化研究所
学位授予地点沈阳
作者部门机器人学研究室
摘要视觉系统任务主要可分为定位、识别和建模,本文主要会对识别问题展开研究。视觉系统的识别问题又可分为目标增强、目标检测和目标跟踪。目标增强是提高目标检测正确率和目标跟踪效果的有效手段;目标检测为无人机自主发现目标以及规避障碍提供直观数据结果;目标跟踪则为无人机自主巡线提供实现手段。相较传统图像视觉处理方法,本文重点考虑无人机应用的特点:实时性、鲁棒性。对传统图像处理方法加以改进以提升处理速度并能有效适应各种复杂背景情况。研究内容主要包括以下几个方面:1.研究无人机航拍图像中线状目标的增强及复杂背景抑制问题。基于无人机航拍图像的灰度变化分布及极值特征,采用特征聚类计算,双边滤波以及可迭代的基于多向滤波的自我增强(IMA)等方法对图像中的电力线目标进行增强。2.研究无人机航拍图像中电力线目标的快速检测方法。在传统图像直线检测算法的基础上,增加几何位置约束和线搜索策略,提出了基于边界搜索的Radon变换(BSRT)检测方法。BSRT方法通过选择边界的方法可以有效规避电力线目标在航拍图像中不是全局占优而出现的误检现象,并且可大幅提升线检测速度。针对弯曲电力线提出了基于活动曲线的弯曲电力线检测(BPLD)方法,该方法在已知电力线起点和方向的前提下可有效检测并表达无人机航拍图像中的弯曲电力线。3.研究了无人机航拍电力巡线中电力线目标和电力铁塔的跟踪问题。在伴线飞行中,针对航拍图像中电力线目标可能由于晃动过大而导致电力线检测失败的问题,利用跟踪算法做为补救措施进行找回;在绕塔飞行中,利用跟踪算法实现对电力铁塔目标的确定以引导无人机环绕飞行。在研究图像的区域特征后,通过采用Mean Shift算法融入粒子滤波(MSPF)的办法提高跟踪算法的实时性。4.研究无人机视觉实验系统。分析了无人机系统的硬件及软件组成,确定了视觉实验系统的定位、工作模式和功能模块;编制了地面站人机交互软件,实现了对优化视觉算法进行切换、调整参数、运行、结果显示以及保存的功能。为无人机视觉算法搭建一个良好的实验平台。
其他摘要Power line inspection is an important way to ensure power transmission of electric power industry. Using lineworker to do this job is low efficiency, subject to be effected by geographical factors, high risk, high cost, and has many other problems, while using rotor UAV(unmanned aerial vehicle) to do it can effectively solve these problems. Visible light vision system is an important perceptual component of a UAV, and can effectively improve the recognition ability and autonomous cruise ability of it. In recent years, the method of using UAV with visible light vision system for power line inspection has been widely used in electric power industry. The main task of vision system can be divided into the positioning, identifying and modeling, this paper will study mainly on identifying problem. Furthermore, identifying problem in vision system can be divided into target enhancement, target detection and target tracking. Target enhancement is an effective means to increase target detection accuracy rate and improve target tracking effect; Target detection provides intuitive data results for UAV autonomously finding targets and avoiding obstacles; Target tracking provides an implementation means for UAV autonomous power line inspection. Compared with traditional image and vision processing methods, this paper consider the characteristics of UAV application: real-time and robust. Traditional image processing methods are improved to enhance the processing speed, and thus can adapt to all kinds of complicated background effectively. The research content mainly includes following several aspects: 1. Linear target enhancements and complex background suppression in UAV aerial image. Based on gray distribution and characteristics of extreme value in UAV aerial image, characteristics clustering computation, bilateral filtering and iterable multidirectional autocorrelation (IMA) approach and other methods are applied to enhance target power lines. 2. Fast detection methods of target power lines in UAV aerial images. On the basis of the traditional algorithms of image line detection, with geometry constraints and line search strategy, a boundary search Radon transform (BSRT) detection method is proposed. By selecting boundaries, BSRT method can effectively avoid error detection of power lines target in an aerial image, even power lines energy is not globally dominant in the image, and line detection speed can also be improved dramatically. For bending power line, a bending power line detection (BPLD) method is proposed based on active curves, which can effectively detect and express bending power line in a UAV aerial image with the premise of knowing power line starting point and the initial direction of it. 3. Tracking method of power line and power pylon in UAV power line inspection. In flying along power line, because power line target in UAV aerial image may shake too much to lead to detection fail, tracking algorithm can be used to find it back as a remedial measure; in flying around tower, tracking algorithm can be used to locate power pylon to guide UAV flying. After researching image region features, an algorithm blending Mean Shift into Particle Filter (MSPF) is proposed to improve algorithm performance. 4. UAV vision experiment system. Hardware and software in UAV system are analyzed, and localization, working modes and function modules of vision experiment system are determined; Ground station human-computer interaction software is programed, and algorithm switching, parameters tuning, runing, results showing and saving functions of optimized algorithm are implemented. Thus, a good experiment platform for UAV vision algorithm is established.
语种中文
产权排序1
文献类型学位论文
条目标识符http://ir.sia.cn/handle/173321/17537
专题机器人学研究室
作者单位1.中国科学院沈阳自动化研究所
2.中国科学院大学
推荐引用方式
GB/T 7714
曹蔚然. 旋翼飞行机器人电力巡线中的在线目标识别与跟踪方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2015.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
旋翼飞行机器人电力巡线中的在线目标识别与(16409KB)学位论文 开放获取CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[曹蔚然]的文章
百度学术
百度学术中相似的文章
[曹蔚然]的文章
必应学术
必应学术中相似的文章
[曹蔚然]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。