矿井悬臂式掘进机位姿感知及定位方法研究.pdf

博士学位论文 矿井悬臂式掘进机位姿感知 及定位方法研究 Study on the Pose Perception and Positioning Mechanism of Boom-type Roadheader in Mines 作 者杜雨馨 导 师童敏明教授 中国矿业大学 二○一九年五月 万方数据 学位论文使用授权声明学位论文使用授权声明 本人完全了解中国矿业大学有关保留、使用学位论文的规定，同意本人所撰 写的学位论文的使用授权按照学校的管理规定处理 作为申请学位的条件之一， 学位论文著作权拥有者须授权所在学校拥有学位 论文的部分使用权，即①学校档案馆和图书馆有权保留学位论文的纸质版和电 子版，可以使用影印、缩印或扫描等复制手段保存和汇编学位论文；②为教学和 科研目的，学校档案馆和图书馆可以将公开的学位论文作为资料在档案馆、图书 馆等场所或在校园网上供校内师生阅读、浏览。另外，根据有关法规，同意中国 国家图书馆保存研究生学位论文。 （保密的学位论文在解密后适用本授权书） 。 作者签名导师签名 年 月 日 年 月 日 万方数据 中图分类号 TP274 学校代码 10290 UDC 621.3 密 级 公开 中国矿业大学 博士学位论文 矿井悬臂式掘进机位姿感知 及定位方法研究 Study on the Pose Perception and Positioning Mechanism of Boom-type Roadheader in Mines 作 者 杜雨馨 导 师 童敏明 教授 申请学位 工学博士学位 培养单位 信息与控制工程学院 学科专业 检测技术与自动化装置 研究方向 采掘装备自动监测 答辩委员会主席 徐立中 教授 评 阅 人 盲审 二○一九年五月 万方数据 致谢致谢 在学位论文即将完成之际，首先我要感谢我的导师童敏明教授。师从数载， 收获颇丰，感触亦深。从论文的拟题到研究路线的选择，再到研究结果的分析与 完善，每个细节都倾注了导师的悉心指导。童老师优秀的做人品质、严谨细致的 治学态度、开拓创新的探索精神、高屋建瓴把握全局的能力、忘我的工作精神为 我树立了榜样，激励我更加努力，在科学研究的道路上奋发向上，使自己不辜负 恩师的期望。在此，谨向我的恩师致以崇高的敬意和深深的感谢 感谢课题组唐守锋教授、董海波副教授对论文架构和写作的指导和鼓励，开 拓了我的思维，给予我勇气去克服实验过程和论文撰写中遇到的种种困难，是我 学习和生活的榜样。 感谢美国肯塔基大学工程学院的 Daniel L. Lau 教授与 Yu Ying 博士， 感谢你 们在我交流访问的一年中对我的关心与指导，使我见识到了国外先进的科学技 术，你们一丝不苟、实事求是的科研作风是我学习和奋斗的目标。 感谢课题组李猛博士、梁良博士和魏明生博士给予的建议、帮助和关心，使 我获得了灵感和启发。他们同时也是我生活中的挚友，能跟他们一起学习、共事 和生活，是我人生中的宝贵经历。感谢实验室以及其他课题组的师兄、师姐、师 弟和师妹的无私帮助，让我感觉到了大家庭般的温暖。感谢学院和学校各位老师 对我的关心和爱护。感谢母校十多年培育，在中国矿业大学 110 周年华诞之际， 祝愿母校积历史之厚蕴,宏图更展再谱华章 感谢双方父母对我的养育和照顾，无论我身处何地，每时每刻都感受到你们 对我的关心、理解和支持，你们无私的亲情和朴实的人格使我受益终生。感谢爱 人张贺先生，你的理解与支持是我前进最大的动力，也是这九万字博士论文的所 有起因、经过与结果。感谢在我人生中为我提供帮助和支持的所有老师、朋友和 亲人，感谢你们为我所做的一切。 最后感谢各位专家、教授在百忙之中参加我的论文评阅与答辩，并提出宝贵 的意见， 促使我进一步完善论文。 在此， 谨向他们致以深深的敬意和由衷的感谢 杜雨馨 2019 年春于中国矿业大学 万方数据 I 摘摘 要要 煤矿环境恶劣、安全事故多发，综采工作面智能化、无人化开采是技术发展 的必然趋势，对于提高煤炭生产效率、降低工作强度、保障人员生命安全具有重 要意义。巷道掘进作为煤矿开采的重要环节，是实现矿井建设、采区通风、煤炭 开采与运输的基础。其中，完成掘进机的自动控制是解决巷道自动化综掘施工的 重要途径。悬臂式掘进机工作过程中，行走机构带动履带推进机身前移，截割机 构通过液压缸改变截割头位置，最终在两者的协同工作下完成截割头的钻进作 业。因此，为使机器按要求实现巷道断面的自动切割，首先需要解决掘进机机身 空间位姿的感知以及截割头的定位问题。 在国家 863 项目“薄煤层半煤岩掘进机关键技术研究” （2013BAK06B04） 的支持下，本论文针对上述问题进行了研究。首先，将机器视觉技术应用到掘进 机机身位姿的检测中，以确定截割头在巷道中的坐标基点；随后，针对截割头定 位，设计了基于轮廓识别的双目定位系统，完成了截割头相对机身的三维坐标获 取；最终，结合已经获取到的机身位姿信息，实现了巷道坐标系下悬臂式掘进机 机身位姿与截割头位置的自动实时监测。具体研究工作与成果如下 （1）通过对比分析现有方法，确定了基于机器视觉的悬臂式掘进机空间位 姿感知与定位系统总体方案。首先，根据悬臂式掘进机的基本结构，构建了掘进 机空间位姿检测坐标系统；然后，基于双十字激光仪对机身位姿进行测量，并首 次系统地提出以偏移角 α，俯仰角 β，横滚角 γ，水平偏移位移Δx，垂直偏移位 移Δz 五个参数描述机身在巷道中的位姿；最终，以获取到的机身位姿为基础， 使用双目视觉系统完成截割头在巷道中的空间定位。 （2）为解决煤矿井下采集到的图像照度低、光照分布不均的问题，提出了 一种改进的 Retinex 小波多尺度积边缘提取算法。通过分析不同照度下图像的频 谱分布，完成了小波高低频信息分解。依照高频与低频部分所含信息的特点，采 用改进的多尺度 Retinex 算法删除了低频子图中的照射分量，并采用模糊增强算 法对边缘信息对比度进行加强。在此基础上，提出改进的小波多尺度积完成了非 均匀照度图像边缘检测。 （3）依照总体方案，设计了一种基于双十字激光标靶的悬臂式掘进机机身 位姿监测系统。该系统以双十字激光仪为基准，利用十字激光线在前后标靶上的 成像特征，建立了机身位姿解算模型。通过对激光标靶的设计、监测设备的选型 与安装， 在模拟平台上完成了掘进机机身位姿自动实时检测实验。 实验结果表明， 机身倾角测量精度为 0.16，位移检测精度为 10mm，该测量系统实时性强，满 足巷道掘进过程中对机身位姿精准检测的要求。 万方数据 II （4）在构建悬臂式掘进机截割头双目定位系统的基础上，针对现有的双目 标定算法存在复杂度高、精度低的问题，提出了一种正交消隐点双目视觉系统标 定算法。首先，基于畸变图像中心对称性与透视投影比例不变性完成了单目相机 内参数的标定。随后，根据两组平行直线的正交消隐点几何特性对摄像机外参数 进行标定。并依照左右极线间的几何约束，对双目视觉系统外参数进行优化，最 终完成了立体系统内外参数的精准标定，具有简单、易操作、精度高的优点。 （5）为实现截割头定位，提出了基于轮廓特征的截割头目标匹配算法。针 对现有轮廓描述子的不足， 在非均匀采样的基础上提出了构建轮廓切线夹角描述 子。 随后， 为精准完成轮廓匹配， 引入跳跃惩戒因子与重复匹配惩戒因子对 MVM 算法进行改进， 使得改进后算法可以很好的适应轮廓形变与噪声干扰， 完成匹配。 匹配实验表明，该算法具有一定的抗噪能力，可以实现目标轮廓的快速、精确匹 配，正确检索率高。在 100 组截割头轮廓的匹配实验中，目标识别率为 100。 （6）使用双目摄像仪实时采集截割头图像，完成了巷道坐标系下悬臂式掘 进机截割头的识别与定位。基于双目系统三维坐标解算原理，对截割头轮廓特征 点进行三维重建与平面拟合。在去除距拟合平面较远的奇异点后，将上边缘均值 点视为截割头空间定位点，并使用坐标转换的方法结合已经得到的机身位姿，完 成了截割头在巷道中的定位。实验结果表明，巷道坐标系下截割头定位点在巷道 截面上的投影误差为 3cm，满足巷道掘进过程中截割头位置精准定位的要求。 本文应用机器视觉技术实现悬臂式掘进机的位姿感知与定位，提高了煤矿井 下掘进机机身与截割头的定位实时性与定位精度， 弥补传统检测方法易受巷道粉 尘浓度干扰、测量路径易遮挡、机身剧烈振动等恶劣测量环境影响的不足。对于 提高井下掘进效率，降低施工风险、实现智能化掘进具有重要的理论意义与应用 价值。 该论文有图 81 幅，表 10 个，参考文献 167 篇。 关键词关键词悬臂式掘进机；位姿感知；截割头定位；激光指向；机器视觉；轮廓匹 配 万方数据 III Abstract Due to the harsh environment in coal mines and frequent safety accidents, the intelligent and unmanned mining has become an inevitable trend of technological development at the comprehensive mining surface. It is of great significance to improve the coal production efficiency, reduce the working intensity and guarantee the personnel safety. As the most important part, the roadway accurate excavation is the foundation of the mine construction, the mining area ventilation, the coal mining and transportation. Therefore, in order to achieve the roadway intelligent construction, we need to accomplish the roadheader automatical control in the first place. During the boom-type roadheader working process, the walking mechanism drives the tracks to push the fuselage forward, and the cutting mechanism changes the cutting head position through hydraulic cylinders. It finally completes the tunnel drilling operation under the cooperation of above two. Hence, for the purpose of realizing the roadway section automatic cutting as required, it is necessary to solve the problem of the machine body pose perception and the cutting head positioning. With the fast development of science and technology, advanced positioning and measurement techniques have been gradually applied into this field. After analyzing various measuring s, this paper firstly studies the roadheader body position and pose detection technique to determine the coordinate base point for the cutting head location in the tunnel space. Then, a binocular system is designed for cutting head positioning based on contours recognition to acquire the 3D coordinates of cutting head relative to the roadheader body. Finally, a boom-type roadheader pose perception and positioning system is proposed to realize the real-time location monitoring of the machine body and the cutting head in the tunnel coordinate system. The main work and achievements are as follows 1 By comparing and analyzing existing researches, the overall scheme is proposed based on machine vision technique for the roadheader pose perception and positioning. Firstly, the pose detection coordinate systems are constructed in the tunnel space according to the machine mechanical structure. Afterwards, for the first time, five parameters are chosen to describe the roadheader body position, which are the deflection angle α, the pitch angle β, the roll angle γ, the horizontal offset displacement Δx and the vertical offset displacement Δz. Then, a double cross laser instrument is used to measure these parameters. Finally, based on the obtained 万方数据 IV position, the cutting head location can be precisely achieved by using a binocular vision system in the tunnel. 2 In order to finish the low-light-level mine image features extraction, an edge detection is proposed in this paper based on the improved Retinex theory and wavelet multi-scale product algorithm. The particular environment in mines makes it difficult to extract features in underground images with dark hue and low target discrimination. Therefore, after the spectrum analysis of images under different illuminance, captured images are decomposed by a wavelet function into four parts. Following, according to the difference of ination contained in different frequency domains, the multi-scale Retinex theory is exploited to eliminate the radiation component in low frequency domains, and the fuzzy enhancement algorithm is used to improve the edge ination contrast. After these, an improved wavelet multi-scale product algorithm is employed to finish the edge detection for non-uni illumination images. 3 According to the overall scheme, a roadheader body pose monitoring system is designed by using a double cross laser instrument. It takes the cross lasers and laser targets as ination sources to establish the machine body position calculating model by analyzing the characteristics of laser line imaging on the front and rear targets. After the laser target design and the monitoring equipment installation, a simulation plat is utilized to complete the automatic roadheader pose detection experiment in real time. The experimental results show that within 2 to 100 metres measuring range, the angles measurement accuracy is 16 and the displacement detection accuracy is 10mm. We can find that the proposed system can satisfy the requirements of automatic, precise, and real-time positioning for the roadheader during the process of tunnel construction. 4 After the construction of the cutting head binocular positioning system, an accurate calibration is presented for binocular systems based on geometric features of vanishing points ed by two sets of parallel lines. It can solve shortcomings of these existing s which own low precision and high computation complexity. In order to reduce the effect of lens distortion on the reconstruction accuracy, internal parameters of the monocular camera are calibrated by using the central symmetry of distorted images and the proportional in-variance of perspective projection. Afterwards, external parameters are figured out on the basis of geometric properties of orthogonal vanishing points. Finally, geometry constraint 万方数据 V between epipolar lines is adopted for optimizing structure parameters of the binocular system by minimizing matching errors under different views. Since the using of a pair of orthogonal parallel lines in the calibration template, the proposed can exactly finish the parameters calibration for stereo systems with impressive precision and stability. 5 After comparing existing image matching s for target recognition, the cutting head matching algorithm is proposed on the basis of target contour features. The shape descriptor, simplified intersection angle of tangent lines SIATL, is firstly proposed on the basis of spatial positions of contour points, which owns great self-contained property and features describing ability for both partial and whole shapes. To achieve best matching, the MVM algorithm is improved with skipping and multiple matching penalties to complete multiple mapping and skip existing outliers in sequences. More precisely, these advantages make the proposed algorithm less sensible to local variations caused by varied views. Our experimental results demonstrate that the proposed outpers existing ones with strong robustness and high accuracy. In actual pictures test, the recognition rate of cutting head can reach 100, which lays the foundation for accurate location in real-time. 6 By using a binocular vision system installed on the machine to collect cutting head images, the recognition and positioning of the cutting head are completed in the simulated tunnel coordinate system. Based on the 3D coordinate solution principle of binocular system, feature points of the target are picked out for precise 3D reconstructed and plane fitting. After removing outliers from the fitting plane, the mean value of the upper edge is regarded as the anchor point of the cutting head in the tunnel space. Then, the cutting head position can be obtained by using the coordinate transation combined with the machine body pose. In order to verify the feasibility of the proposed scheme, the positioning experiment is carried out by manually manipulating the roadheader in the simulated roadway. The results show that, in the roadway coordinate system, the anchor point projection error on the tunnel section is 3 cm, which satisfies the requirement of cutting head precise location in the process of roadway excavation. In this paper, the advanced machine vision technology is applied into coal mines for the roadheader positioning underground, which significantly improves the positioning efficiency, accuracy and real-time of roadheader body and the cutting head. It also makes up shortcomings of traditional s brought by harsh measuring 万方数据 VI environment in mines, such as the easily blocked measuring path, thick dust concentration, excessive vibration of fuselage, etc. In addition, the proposed positioning system has important theoretical significance and application value for improving the efficiency of underground tunneling, reducing construction risk and realizing intelligent tunneling. The present dissertation has 81 figures, 10 tables and 167 references. Keywords boom-type roadheader; pose perception; cutting head positioning; laser pointing; machine vision; contour matching 万方数据 VII 目目 录录 摘摘 要要............................................................................................................................ I 目目 录录........................................................................................................................ VII 图清单图清单........................................................................................................................ XII 表清单表清单..................................................................................................................... XVII 变量注释表变量注释表 .......................................................................................................... XVIII 1 绪论绪论............................................................................................................................ 1 1.1 研究背景及意义 ..................................................................................................... 1 1.2 研究现状 ................................................................................................................. 2 1.3 主要研究内容 ......................................................................................................... 8 1.4 论文组织结构 ......................................................................................................... 9 2 掘进机位姿感知及定位方法分析与总体方案设计掘进机位姿感知及定位方法分析与总体方案设计 ............................................. 12 2.1 悬臂式掘进机基本结构 ....................................................................................... 12 2.2 掘进机空间定位坐标系统 ................................................................................... 13 2.3 掘进机机身位姿数学模型与测量方法对比分析 ............................................... 14 2.4 掘进机截割机构简化模型与定位方法对比分析 ............................................... 21 2.5 悬臂式掘进机空间位姿感知及定位系统总体方案设计 ................................... 25 2.6 位姿感知及定位参数考核指标 ........................................................................... 26 2.7 本章小结 ............................................................................................................... 26 3 基于双十字激光的掘进机机身位姿感知基于双十字激光的掘进机机身位姿感知 ............................................................. 28 3.1 引言 ....................................................................................................................... 28 3.2 基于双十字激光的机身位姿测量方案 ............................................................... 28 3.3 位姿感知系统设备选型与安装 ........................................................................... 30 3.4 基于 Retinex 与小波多尺度积的低照度图像边缘检测算法 ............................ 34 3.5 激光标靶特征点识别与坐标解算 ....................................................................... 41 3.6 悬臂式掘进机机身位姿解算 ...............................