煤炭是我国的主要能源，随着工业互联网、人工智能、物联网、智能机器人等技术的发展，建设智能矿山符合国家发展战略，是煤炭行业的必然选择。因此，通过研究煤矿井下空间的实时三维建图、定位和导航技术，可以解决井下机器人和人员的定位及导航问题，有利于保障人员安全和提高智能化开采效率，对建设智能矿山具有重要意义。
        同步定位与地图构建（Simultaneous Localization and Mapping，SLAM）技术在机器人领域用于在未知环境下的定位及地图构建，为机器人的自主定位及导航奠定基础。本文在建设智能矿山的背景下，提出基于SLAM技术的煤矿井下空间重建的方法。本文通过国家煤炭工业采矿工程重点实验室模拟煤矿井下环境，采用RGB-D类深度相机中的Kinect2.0作为传感器获取图像。
针对在图像信息获取过程中受矿井低光照、粉尘大等环境因素影响的问题，通过张正友标定法对Kinect2.0相机进行标定，提高了获取的图像信息的精度。并利用不同去噪算法进行实验，根据实验测试和评价指标，最终选择高斯去噪算法。
       针对在煤矿井下获得的图像缺少纹理和模糊造成特征点误匹配对过多，导致地图构建失败的问题，采用特征点法实现视觉里程计。对比分析SIFT、SURF、ORB特征提取算法特点，提出采用改进的ORB算法，提高了特征点提取的速度。对比分析GMS、RANSAC特征配准算法，采用最近邻法对RANSAC算法进行改进，根据实验对比分析，采用改进的RANSAC算法进行剔除误匹配，减少了迭代次数，提高了机器人在煤矿环境下进行地图构建的鲁棒性。
       针对在三维地图构建过程中存在信息冗余和累计误差的问题，通过选取构建位姿图的关键帧，剔除重复的图像帧。在地图优化方面，通过BA算法和BoVW方法对位姿进行调整优化，提高构建三维点云地图的全局一致性。为满足机器人自主导航对地图的需求，在三维稠密点云地图的基础上，将其转化为三维八叉树地图，为机器人自主导航奠定基础。
       本文提出的基于SLAM的煤矿井下空间重建的方法可以稳定高效地构建出煤矿井下环境的三维地图，实现对矿井下工作场景的三维再现，为井下机器人进行智能化开采、建设智能矿山提供了理论创新性和应用价值。

       Coal is the main energy source in China. With the development of industrial Internet, artificial intelligence, Internet of Things, intelligent robots and other technologies, the construction of intelligent mines conforms to the national development strategy and is the inevitable choice of the coal industry.Therefore, by studying the real-time three-dimensional mapping, positioning and navigation technology of underground space of coal mine, the positioning and navigation problems of underground robots and personnel can be solved, which is conducive to ensuring personnel safety and improving the efficiency of intelligent mining, and is of great significance to the construction of intelligent mines.
        Simultaneous localization and mapping technology (SLAM) is used in the field of robots for positioning and map building in unknown environments, which lays a foundation for autonomous positioning and navigation of robots. Under the background of intelligent mine construction, this paper proposes a method of underground space reconstruction based on SLAM technology.In this paper, the underground environment of coal mine is simulated by the State Key Laboratory of Mining Engineering of Coal Industry, and Kinect2.0 in RGB-D depth camera is used as the sensor to obtain the image.
         Aiming at the problem of environmental factors such as low illumination and large dust in the process of image information acquisition, the Kinect2.0 was calibrated by Zhang Zhengyou calibration method to improve the accuracy of image information acquisition.Different denoising algorithms are used for experiments. According to the experimental tests and evaluation indexes, the Gaussian denoising algorithm is finally selected.
         Aiming at the problem that lack of texture and blur in the image acquired in coal mine leads to too many mismatched feature points, which leads to the failure of map construction, the visual odometer is realized by using the feature point method.The features of SIFT, SURF and ORB feature extraction algorithms are compared and analyzed. An improved ORB algorithm is proposed to improve the speed of feature point extraction.GMS and RANSAC feature registration algorithms are compared and analyzed, and RANSAC algorithm is improved by using the nearest neighbor method. According to the comparative analysis of experiments, the improved RANSAC algorithm is used to eliminate mismatches, which reduces the number of iterations and improves the robustness of robot map construction in coal mine environment.
         In order to solve the problems of information redundancy and accumulated error in the process of 3D map construction, the key frames of pose map construction were selected to eliminate the repeated image frames.In the aspect of map optimization, BA algorithm and BOVW method are used to adjust and optimize the pose to improve the global consistency of three-dimensional point cloud map construction.In order to meet the needs of the map for autonomous navigation of robot, the 3D dense point cloud map was transformed into a 3D octree map, which laid the foundation for autonomous navigation of robot.
        In this paper, the method of space reconstruction in underground coal mine based on SLAM is proposed, which can stably and efficiently construct 3D map of underground coal mine environment and realize 3D reproduction of underground working scene, which provides important theoretical innovation and application value for underground robot to carry out intelligent mining and construct intelligent mine.

[1]中矿(北京)煤炭产业景气指数研究课题组,岳福斌.2020-2021年中国煤炭产业经济形势研究报告[J].中国煤炭,2021,47(03):53-61.

[2]丁国峰,吕振福,曹进成,武秋杰.我国大型煤炭基地开发利用现状分析[J].能源与环保,2020,42(11):107-110+120.

[3]胡青松,钱建生,李世银,孙彦景.智能煤矿技术研究与政策制定现状[J].工矿自动化,2021,47(03):1-8.

[4]王国法,赵国瑞,任怀伟.智慧煤矿与智能化开采关键核心技术分析[J].煤炭学报,2019,44(01):34-41.

[5]王国法,杜毅博.智慧煤矿与智能化开采技术的发展方向[J].煤炭科学技术,2019,47(01):1-10.

[6]张申,胡青松,王刚.矿山物联网标准建设的思考[J].工矿自动化,2018,44(01):1-5.

[7]胡凌凤. 煤炭企业信息化标准体系构建研究[D].北京：中国矿业大学(北京),2016:43-65.

[8]韩茜. 智慧矿山信息化标准化系统关键问题研究[D].中国矿业大学(北京),2016.

[9]方新秋,梁敏富,李爽,张磊,严黄宝,谢小平,尉瑞,吴刚,吕嘉锟.智能工作面多参量精准感知与安全决策关键技术[J].煤炭学报,2020,45(01):493-508.

[10]王国法,张德生.煤炭智能化综采技术创新实践与发展展望[J].中国矿业大学学报,2018,47(03):459-467.

[11]张申,胡青松,王刚.矿山物联网标准建设的思考[J].工矿自动化,2018,44(01):1-5.

[12]Li Ao et al. DP-SLAM: A visual SLAM with moving probability towards dynamic environments[J]. Information Sciences, 2021, 556: 128-142.

[13]王霞,左一凡.视觉SLAM研究进展[J].智能系统学报,2020,15(05):825-834.

[14]Cadena C, Carlone L, Carrillo H, et al. Past, Present, and Future of Simultaneous Localization and Mapping: Toward the Robust-Perception Age[J]. IEEE Transactions on

Robotics, 2016, 32(6):1309-1332.

[15]Skrzypczynski P, Belter D. Precise self-localization of a walking robot on rough terrain using parallel tracking and mapping[J]. Industrial Robot, 2013. 40(3):229-237.

[16]Izadi S, Kim D, Hilliges O, et al. Kinectfusion: real-time 3D reconstruction and interaction using a moving depth camera[J]. Proc Uist, 2011:559-568.

[17]Newcombe R A, Izadi S, Hilliges O, et al. KinectFusion: Real-time dense surface mapping and tracking[C]// IEEE International Symposium on Mixed and Augmented Reality. IEEE Computer Society, 2011:127-136.

[18]Whelan T, Kaess M, Johannsson H, et al. Real-time large-scale dense RGB-D SLAM with volumetric fusion[J]. International Journal of Robotics Research, 2015, 34(4-5):598-626.

[19]Rublee E, Rabaud V, Konolige K, et al. ORB: An efficient alternative to SIFT or SURF [C]// Proc of International Conference on Computer Vision. Washington DC: IEEE Computer Society, 2011: 2564-2571.

[20]Mur-Artal R, Tardos J D. ORB-SLAM2: an open-source SLAM system for monocular, stereo and RGB-D cameras [J]. arXiv preprint arXiv:1610. 06475, 2016.

[21]Dai A, Nießner M, Zollhöfer M, et al. Bundlefusion: Real-time globally consistent 3d reconstruction using on-the-fly surface reintegration[J]. ACM Transactions on Graphics (ToG), 2017, 36(4): 1.

[22]He Y, Zhao J, Guo Y, et al. Pl-vio: Tightly-coupled monocular visual–inertial odometry using point and line features[J]. Sensors, 2018, 18(4): 1159.

[23]Campos C , Elvira R , Juan J. Gómez Rodríguez, et al. ORB-SLAM3: An Accurate Open-Source Library for Visual, Visual-Inertial and Multi-Map SLAM[J]. 2020.

[24]史瑶. 基于草图输入的煤矿巷道建模方法的研究[D].中国矿业大学,2020.

[25]Anonymous. CAE Sells Datamine to Constellation Software[J]. Engineering and Mining Journal, 2015, 216(10) : 77.

[26]Simit Raval, Bikram Pratap Banerjee, et al. A Preliminary Investigation of Mobile Mapping Technology for Underground Mining[C]// IEEE International Geoscience and Remote Sensing Symposium.IEEE, 2019.

[27]陈凯,杨小聪,张达.采空区三维激光扫描变形监测系统[J].矿冶,2012,21(01):60-63.

[28]杨彪, 王卓. 三维重建中的点云配准方法研究[J]. 计算机与数字工程, 2014, 42(002):300-303.

[29]王保岩. 矿井巷道三维重建及安全监测技术研究[D].山东大学,2018.

[30]杨宇,赵成星,张晓玲.鱼眼相机的视觉标定及畸变校正[J].激光杂志,2020,41(09):20-23.

[31]宋燕飞,罗尧治,沈雁彬,蔡朋程.基于双目视觉与图像识别的网架结构三维重建[J].空间结构,2020,26(04):28-35+74.

[32]王谭,王磊磊,张卫国,段晓韬,王婉丽.基于张正友标定法的红外靶标系统[J].光学精密工程,2019,27(08):1828-1835.

[33] Jürgen Sturm, Engelhard N, Endres F, et al. A benchmark for the evaluation of RGB-D SLAM systems[C]// IEEE/RSJ International Conference on Intelligent Robots & Systems.IEEE, 2012.

[34]吴国斌, 李斌, 阎骥洲. KINECT 人机交互开发实践[M]. 人民邮电出版社, 2013.

[35]陈晓明,蒋乐天,应忍冬.基于Kinect深度信息的实时三维重建和滤波算法研究[J].计算机应用研究,2013,30(04):1216-1218.

[36]Fengcai Huo et al. Two-stage image denoising algorithm based on noise localization[J]. Multimedia Tools and Applications, 2021: 1-22.

[37]Tu W C, Lai Y A, Chien S Y. Constant time bilateral filtering for color images[C]// 2016 IEEE International Conference on Image Processing (ICIP). IEEE, 2016: 3309-3313.

[38]于虹,甄彤.图像去噪经典算法研究[J].信息与电脑(理论版),2020,32(14):66-67.

[39]袁华,庞建铿,莫建文.基于体素化网格下采样的点云简化算法研究[J].电视技术,2015,39(17):43-47.

[40]慈文彦,黄影平,胡兴.视觉里程计算法研究综述[J].计算机应用研究,2019,36(09):2561-2568.

[41]Forster C, Pizzoli M, Scaramuzza D. SVO: Fast semi-direct monocular visual odometry[C]// IEEE International Conference on Robotics and Automation. IEEE, 2014:15-22.

[42]Engel J, Sturm J, Cremers D. Semi-dense Visual Odometry for a Monocular Camera[C]// IEEE International Conference on Computer Vision. IEEE, 2014:1449-1456.

[43]Engel J, Stückler J, Cremers D. Large-scale direct SLAM with stereo cameras[C]// Ieee/rsj International Conference on Intelligent Robots and Systems. IEEE, 2015:1935-1942.

[44]Kerl C, Sturm J, Cremers D. Dense visual SLAM for RGB-D cameras[C]// Ieee/rsj International Conference on Intelligent Robots and Systems. IEEE, 2014:2100-2106.

[45]冯来磊.三维场景几何特征提取算法研究现状与分析[J].科学技术创新,2018(21):3-5.

[46]Rublee E, Rabaud V, Konolige K, et al. ORB: An efficient alternative to SIFT or SURF[C]// IEEE International Conference on Computer Vision. IEEE, 2012:2564-2571.

[47]高翔. 视觉SLAM十四讲：从理论到实践[M]. 电子工业出版社, 2017.

[48]王金龙,周志峰.基于SIFT图像特征提取与FLANN匹配算法的研究[J].计算机测量与控制,2018,26(02):175-178.

[49]Bian J, Lin W Y, Matsushita Y, et al. GMS: Grid-Based Motion Statistics for Fast, Ultra-Robust Feature Correspondence[C]// IEEE Conference on Computer Vision and Pattern Recognition. IEEE Computer Society, 2017:2828-2837.

[50]马丽丽,曹春梅,陈金广,王星辉.基于RANSAC的特征点匹配算法[J].计算机工程与设计,2016,37(07):1794-1797+1856.

[51]Gálvez-López D, Tardos J D. Bags of binary words for fast place recognition in image sequences[J]. IEEE Transactions on Robotics, 2012, 28(5): 1188-1197.

[52]Ondrej Chum, James Philbin, Andrew Zisserman. Near Duplicate Image Detection: min-Hash and tf-idf Weighting[C]// British Machine Vision Conference. DBLP, 2014.781-786.

[53]Hornung A, Wurm K M, Bennewitz M, et al. OctoMap: an efficient probabilistic 3D mapping framework based on octrees[J]. Autonomous Robots, 2013, 34(3): 189-206.