煤炭是我国经济发展的重要基石，我国以煤炭为主导地位的能源结构短时期无法改变。2020年，国家发改委联合八部发布的《关于加快煤矿智能化发展的指导意见》表明，智能化是我国煤炭产业发展的必然要求。基于视觉的同时定位与建图(Simultaneous Localization and Mapping，SLAM)技术，可以实时准确地构建煤矿井下三维地图，实现井下机器人的避障导航功能，对煤炭产业智能化发展意义重大。然而煤矿井下低照度、多粉尘、弱纹理的恶劣环境，严重影响了视觉传感器数据采集质量以及特征提取匹配精度，极大限制了视觉SLAM在煤矿井下的应用。所以为了提高视觉SLAM算法在煤矿井下的适用性，本文研究了顾及图像增强的煤矿井下视觉SLAM算法，具体研究内容如下：

(1)视觉SLAM以图像信息为基础，其定位和建图精度与图像质量密切相关。煤矿井下低照度、多粉尘、弱纹理的恶劣环境，严重影响了煤矿井下图像质量，导致其存在照度不均、对比度低、细节模糊的问题。所以为了增强煤矿井下图像质量，提高视觉SLAM算法的整体精度与稳定性，本文设计了一种基于改进双边滤波的Retinex算法。

将原始RGB(Red，Green，Blue)图像转换至HSI(Hue，Saturation，Intensity)色彩空间，以改进的双边滤波代替传统Retinex算法的高斯滤波作为中心环绕函数，对图像反射分量进行估计后转换至RGB色彩空间，得到最终增强图像。相较于单尺度Retinex(Single-Scale Retinex，SSR)算法和多尺度Retinex(Muti-Scale Retinex，MSR)算法，该算法处理后的图像未出现明显的泛白及光晕现象，图像质量得到了明显提升。

(2)针对煤矿井下弱纹理、低对比度环境，视觉SLAM容易发生跟踪丢失和建图失败的问题。本文提出一种基于自适应阈值的ORB特征提取匹配算法。根据图像对比度，自适应调整特征提取算法的检测阈值。试验结果表明：与传统的ORB算法相比，本文算法提取特征点数量更多，增强了视觉SLAM前端里程计的鲁棒性，满足了视觉SLAM算法在煤矿井下的特征提取匹配需求。

(3)针对视觉SLAM算法在建图过程中存在的信息冗余以及累积误差问题。本文研究了基于关键帧的局部地图优化以及基于视觉词袋的回环检测模型，有效剔除了重复图像，节约了计算成本，提高了相机位姿精度，并在此基础上构建了煤矿井下的全局一致性地图。

试验结果表明：本文的视觉SLAM在明暗变化区域和弱纹理区域中表现效果更好，相比ORB-SLAM2，本文的绝对位姿误差均值减少了76.2%，具有更好的轨迹精度。同时，本文构建的稠密点云地图及八叉树地图质量良好，具有较高的鲁棒性，可为机器人在煤矿井下的自主导航提供技术参考。

Coal is an important cornerstone of China's economic development, and its dominance in the energy structure cannot be changed in the short term. In 2020, the National Development and Reform Commission and eight other departments jointly issued the "Guidance on Accelerating the Intelligent Development of Coal Mines", which indicated that intelligence is a inevitable requirement for the development of China's coal industry. Simultaneous localization and mapping (SLAM) technology based on vision can be used to construct real-time and accurate 3D maps of underground coal mines, enabling obstacle avoidance and navigation functions for underground robots, which are of great significance for the intelligent development of the coal industry. However, the harsh underground environment of low illumination, high dust, and weak texture seriously affects the quality of the visual sensor data and feature extraction and matching accuracy, greatly limits the application of vision SLAM in coal mines. Therefore, in order to improve the applicability of visual SLAM algorithms in coal mines, this paper studies a coal mine visual SLAM algorithm that takes into account image enhancement. The specific research contents are as follows:

(1) Visual SLAM is based on image information, and its positioning and mapping accuracy are closely related to image quality. The harsh underground environment of low illumination, high dust, and weak texture in coal mines seriously affects the image quality, resulting in problems such as uneven illumination, low contrast, and blurred details. In order to enhance the image quality in coal mines and improve the overall accuracy and stability of visual SLAM algorithms, this paper designs a Retinex algorithm based on improved bilateral filtering. The original RGB image is converted to HSI color space, and the improved Bilateral filter replaces the Gaussian filter of the traditional Retinex algorithm as the central surrounding function. After the image reflection component is estimated, it is converted to RGB color space to obtain the final enhanced image. Compared with the single-scale Retinex (SSR) algorithm and the multi-scale Retinex (MSR) algorithm, the image processed by this algorithm don’t show obvious whitening or halo phenomena, and the image quality is significantly improved.

(2) In view of the problem that visual SLAM is prone to tracking loss and mapping failure in coal mines due to weak texture and low contrast environment, this paper proposes an adaptive threshold ORB feature extraction and matching algorithm. The detection threshold of the feature extraction algorithm is adaptively adjusted according to the image contrast. Experimental results show that compared with the traditional ORB algorithm, the algorithm proposed in this paper extracts more feature points, enhances the robustness of the visual SLAM front-end odometer, and meets the feature extraction and matching requirements of visual SLAM algorithms in coal mines.

(3) To address the problem of information redundancy and cumulative error in the mapping process of visual SLAM algorithms, this paper studies a local map optimization method based on keyframes and a loop closure detection model based on visual bag of words.which effectively eliminates duplicate images, saves computing costs, improves camera pose accuracy, and constructs a globally consistent map of underground coal mines.

Experimental results show that the visual SLAM proposed in this paper performs better in areas of varying illumination and weak textures. Compared with ORB-SLAM2, the mean absolute pose error of this paper is reduced by 76.2%, and has better trajectory accuracy. At the same time, the dense point cloud map and octree map constructed in this paper have good quality and high robustness, providing technical reference for robot's autonomous navigation in coal mines.

[1] 陈云涛, 王汝婷. 分析我国煤炭能源消费现状及节能优化路径[J]. 产业创新研究, 2020(10): 38-39.

[2] 李洪言, 张景谦, 陈健斌等. 2021年全球能源转型面临挑战——基于《bp世界能源统计年鉴(2022)》[J]. 天然气与石油, 2022, 40(06): 129-138.

[3] 杜易培. 中国煤炭工业采煤机械化经济与安全效益的量化分析[D]. 北京: 中国矿业大学, 2022.

[4] 葛世荣, 胡而已, 裴文良等. 煤矿机器人体系及关键技术[J]. 煤炭学报, 2020, 45(1): 455-463.

[5] 杨必胜, 梁福逊, 黄荣刚. 三维激光扫描点云数据处理研究进展、挑战与趋势[J]. 测绘学报, 2017, 46(10): 1509-1516.

[6] 吴东金, 夏林元. 面向室内WLAN定位的动态自适应模型[J]. 测绘学报, 2015, 44(12): 1322-1330.

[7] 符世琛, 李一鸣, 杨健健等. 基于超宽带技术的掘进机自主定位定向方法研究[J]. 煤炭学报, 2015, 40(11): 2603-2610.

[8] 李论, 张著洪, 丁恩杰等. 基于RSSI的煤矿巷道高精度定位算法研究[J]. 中国矿业大学学报, 2017, 46(01): 183-191.

[9] 张羽飞. 煤矿井下移动机器人视觉与惯导信息融合的定位技术研究[D]. 西安: 西安科技大学, 2021.

[10] 江传龙, 黄宇昊, 韩超等. 井下巡检无人机系统设计及定位与避障技术[J]. 机械设计与研究, 2021, 37(04): 38-42.

[11] 齐广峰, 吕军锋. MEMS惯性技术的发展及应用[J]. 电子设计工程, 2015, 23(1): 87-89.

[12] 马宏伟, 王岩, 杨林. 煤矿井下移动机器人深度视觉自主导航研究[J]. 煤炭学报, 2020, 45(06): 2193-2206.

[13] FUENTES-PACHECO J, RUIZ-ASCENCIO J, RENDÓN-MANCHA J M. Visual simultaneous localization and mapping: a survey[J]. Artificial Intelligence Review, 2015, 43(1): 55-81.

[14] DURRANT-WHYTE H, BAILEY T. Simultaneous localization and mapping: part I[J]. IEEE robotics & automation magazine, 2006, 13(2): 99-110.

[15] 王浩森. 基于深度相机的煤矿井下机器人SLAM技术研究[D]. 西安: 西安科技大学, 2021.

[16] 郭永坤, 朱彦陈, 刘莉萍等. 空频域图像增强方法研究综述[J]. 计算机工程与应用, 2022, 58(11): 23-32.

[17] 田子建, 王满利, 张元刚. 基于双域分解的图像增强算法[J]. 电子学报, 2020, 48(07): 1311-1320.

[18] Wu Q, Wang X, Hua L, et al. Improved time optimal anti-swing control system based on low-pass filter for double pendulum crane system with distributed mass beam[J]. Mechanical Systems and Signal Processing, 2021, 151: 107444.

[19] 冯相辉. 一种改进的同态滤波图像增强算法[J]. 重庆邮电大学学报(自然科学版), 2020, 32(01): 138-145.

[20] 廖斌, 刘鸳鸳. 基于多尺度灰度变换的图像增强研究[J]. 量子电子学报, 2015, 32(05): 550-554.

[21] Mustafa W A, Kader M M M A. A review of histogram equalization techniques in image enhancement application[C]//Journal of Physics: Conference Series. IOP Publishing, 2018, 1019(1): 012026.

[22] 苏康友, 王晓刚, 柳革命. 改进Retinex算法的光照不均匀图像增强研究[J]. 激光杂志, 2022, 43(09): 103-108.

[23] Zotin A. Fast algorithm of image enhancement based on multi-scale retinex[J]. Procedia Computer Science, 2018, 131: 6-14.

[24] 智宁, 毛善君, 李梅. 基于照度调整的矿井非均匀照度视频图像增强算法[J]. 煤炭学报, 2017, 42(08): 2190-2197.

[25] 张立亚, 郝博南, 孟庆勇等. 基于HSV空间改进融合Retinex算法的井下图像增强方法[J]. 煤炭学报, 2020, 45(S1): 532-540.

[26] 权美香, 朴松昊, 李国. 视觉SLAM综述[J]. 智能系统学报, 2016, 11(06): 768-776.

[27] Davison A J, Reid I D, Molton N D, et.al. MonoSLAM: real-time single camera SLAM[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2007, 29(6): 1052-1067.

[28] Klein G, Murray D. Parallel tracking and mapping for small AR workspaces[C]//2007 6th IEEE and ACM international symposium on mixed and augmented reality. IEEE, 2007: 225-234.

[29] Viswanathan D G. Features from accelerated segment test(fast)[C]//Proceedings of the 10th workshop on image analysis for multimedia interactive services, London, UK. 2009: 6-8.

[30] Mur-Artal R, Montiel J M M, Tardos J D. ORB-SLAM: a versatile and accurate monocular SLAM system[J]. IEEE transactions on robotics, 2015, 31(5): 1147-1163.

[31] Rublee E, Rabaud V, Konolige K, et al. ORB: An efficient alternative to SIFT or SURF[C]//2011 International conference on computer vision. IEEE, 2011: 2564-2571.

[32] Mur-ArtalRaúl, D Tardósjuan. ORB-SLAM2: An Open-Source SLAM System for Monocular, Stereo, and RGB-D Cameras[J]. IEEE Transactions on Robotics, 33(5): 1255-1262.

[33] Campos C, Elvira R, Rodríguez J J G, et al. Orb -slam3: An accurate open-source library for visual, visual–inertial, and multimap slam[J]. IEEE Transactions on Robotics, 2021, 37(6): 1874-1890.

[34] Caldato B A C, Achilles Filho R, Castanho J E C. ORB-ODOM: Stereo and odometer sensor fusion for simultaneous localization and mapping[C]//2017 latin American robotics symposium (LARS) and 2017 Brazilian symposium on robotics (SBR). IEEE, 2017: 1-5.

[35] Pumarola A, Vakhitov A, Agudo A, et al. PL-SLAM: Real-time monocular visual SLAM with points and lines[C]//2017 IEEE international conference on robotics and automation (ICRA). IEEE, 2017: 4503-4508.

[36] Yang G, Chen Z, Li Y, et al. Rapid relocation method for mobile robot based on improved ORB-SLAM2 algorithm[J]. Remote Sensing, 2019, 11(2): 149.

[37] Newcombe R A, Lovegrove S J, Davison A J. DTAM: Dense tracking and mapping in real -time[C]. 2011 international conference on computer vision. IEEE, 2011:2320-2327.

[38] Engel J, Schöps T, Cremers D. LSD-SLAM: Large-scale direct monocular SLAM[C]//Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part II 13. Springer International Publishing, 2014: 834-849.

[39] Engel J, Koltun V, Cremers D. Direct sparse odometry[J]. IEEE transactions on pattern analysis and machine intelligence, 2017, 40(3): 611-625.

[40] 杭程. 面向动态场景的视觉SLAM系统研究[D]. 南京: 南京邮电大学, 2022.

[41] Chuang J H, Ho C H, Umam A, et al. Geometry-Based Camera Calibration Using Closed-Form Solution of Principal Line [J]. IEEE Transactions on Image Processing, 2021, PP(99): 2599-2610.

[42] 林慧英, 苏建, 刘玉梅, 陈熔. 基于图像畸变矫正的摄像机标定方法[J]. 吉林大学学报(工学版), 2007(02): 433-437.

[43] 彭祎帆, 陶毅阳, 于超等.基于红外结构光的三维显示用交互装置[J]. 光学学报, 2013, 33(04): 116-122.

[44] 李占利, 周康, 牟琦, 李洪安. TOF相机实时高精度深度误差补偿方法[J]. 红外与激光工程, 2019, 48(12): 263-272.

[45] Joyce Robert A, Leszek, Williams Owen M. Resistor array infrared projector nonuniformity correction: search for performance improvement I II[J]. Defence Science and Technology Organisation(Australia), 2008, 6942.

[46] 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.

[47] 张丽颖. 矿井下目标尺度自适应跟踪算法研究[D]. 北京: 中国矿业大学, 2019．

[48] Cai L, Qian J, Li S, et al. Miner fuzzy detection based on mixture Gaussian model in underground coal mine videos[C]//2010 3rd International Congress on Image and Signal Processing. IEEE, 2010, 1: 437-441.

[49] Cheng X, Cao M, Collier M. An on-line detection system for coal mine dust[C]//2008 7th World Congress on Intelligent Control and Automation. IEEE, 2008: 4166-4171.

[50] 郭晨路. 煤矿采掘输煤除尘技术问题及方案的优化[J]. 矿业装备, 2021(05): 150-151．

[51] 孙晓东, 屈淑亚. 煤矿井下环缝引射水雾除尘器的结构研制及试验[J]. 煤矿机械，2014, 35(04): 50-51．

[52] 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.

[53] 方玉明, 眭相杰, 鄢杰斌等. 无参考图像质量评价研究进展[J]. 中国图象图形学报, 2021, 26(02): 265-286.

[54] 胡志锋. 特征点匹配的SLAM视觉里程计研究[D]. 芜湖: 安徽工程大学, 2022.

[55] 邹雄, 肖长诗, 文元桥, 元海文. 基于特征点法和直接法VSLAM的研究[J]. 计算机应用研究, 2020, 37(05): 1281-1291.

[56] 胡凯, 吴佳胜, 郑翡, 张彦雯, 陈雪超, 鹿奔. 视觉里程计研究综述[J]. 南京信息工程大学学报(自然科学版), 2021, 13(03): 269-280.

[57] Engel J, Stückler J, Cremers D. Large-scale direct SLAM with stereo cameras[C]//2015 IEEE/RSJ international conference on intelligent robots and systems (IROS). IEEE, 2015: 1935-1942.

[58] Kerl C, Sturm J, Cremers D. Dense visual SLAM for RGB-D cameras[C]//2013 IEEE/RSJ International Conference on Intelligent Robots and Systems. IEEE, 2013: 2100-2106.

[59] 王浩森. 基于深度相机的煤矿井下机器人SLAM技术研究[D]. 西安: 西安科技大学, 2021.

[60] Lowe D G. Object recognition from local scale-invariant features[C]//Proceedings of the seventh IEEE international conference on computer vision. IEEE, 1999, 2: 1150-1157.

[61] Bay H, Ess A, Tuytelaars T, et al. Speeded-up robust features (SURF)[J]. Computer vision and image understanding, 2008, 110(3): 346-359.

[62] Li S, Xu C, Xie M. A robust O(n) solution to the perspective-n-point problem[J]. IEEE transactions on pattern analysis and machine intelligence, 2012, 34(7): 1444-1450.

[63] Foresee F D, Hagan M T. Gauss-Newton approximation to Bayesian learning[C]//Proceedings of international conference on neural networks (ICNN'97). IEEE, 1997, 3: 1930-1935.

[64] Moré J J. The Levenberg-Marquardt algorithm: implementation and theory[C]//Numerical Analysis: Proceedings of the Biennial Conference Held at Dundee, June 28–July 1, 1977. Berlin, Heidelberg: Springer Berlin Heidelberg, 2006: 105-116.

[65] Land E H, Mccann J J. Lightness and Retinex theory[J]. JOSA, 1971, 61(1): 1-11.

[66] Land E H. The Retinex theory of color vision[J]. Scientific American, 1977, 237(6): 108-128.

[67] Jobson D J, Rahman Z, Woodell G A. Properties and performance of a center/surround retinex[J]. IEEE Transactions on Image Processing, 1997, 6(3): 451–462.

[68] Jobson D J, Rahman Z, Woodell G A. A multiscale retinex for bridging the gap between color images and the human observation of scenes [J]. IEEE Transactions on Image Processing, 1997, 6 (7): 965–976.

[69] 谢凤英, 汤萌, 张蕊. 基于Retinex的图像增强方法综述[J]. 数据采集与处理, 2019, 34(01): 1-11.

[70] Nuchter A, Surmann H, Lingemann K, et al. 6D SLAM with an application in autonomous mine mapping[C]//IEEE International Conference on Robotics and Automation, 2004. Proceedings. ICRA'04. 2004. IEEE, 2004, 2: 1998-2003.

[71] Endres F, Hess J, Engelhard N, et al. An evaluation of the RGB-D SLAM system[C]//2012 IEEE international conference on robotics and automation. IEEE, 2012: 1691-1696.

[72] Lee G H, Fraundorfer F, Pollefeys M. RS-SLAM: RANSAC sampling for visual FastSLAM[C]//2011 IEEE/RSJ International Conference on Intelligent Robots and Systems. IEEE, 2011: 1655-1660.

[73] Triggs B, McLauchlan P F, Hartley R I, et al. Bundle adjustment—a modern synthesis[C]//Vision Algorithms: Theory and Practice: International Workshop on Vision Algorithms Corfu, Greece, September 21–22, 1999 Proceedings. Springer Berlin Heidelberg, 2000: 298-372.

[74] Sivic J, Zisserman A. Video Google: A text retrieval approach to object matching in videos[C]//Computer Vision, IEEE International Conference on. IEEE Computer Society, 2003, 3: 1470-1470.

[75] Robertson S. Understanding inverse document frequency: on theoretical arguments for IDF[J]. Journal of documentation, 2004.

[76] STURM J, ENGELHARD N, ENDRES F, et al. A benchmark for the evaluation of RGBD SLAM systems[C]//2012 IEEE/RSJ International Conference on Intelligent Robots and Systems, New York: IEEE, 2012: 573-580.