在煤矿井下的作业环境中，移动机器人的应用在保证效率的同时极大降低了人工作业的风险。然而在井下这类复杂环境，经典的即时定位与建图（Simultaneous Location and Mapping，SLAM）算法性能受到挑战。对于视觉SLAM系统，在环境纹理不足、光照不良以及系统剧烈颠簸等条件下存在特征跟踪丢失的情况，进而导致定位与建图失败，严重影响机器人完成后续任务。为应对井下环境中的复杂任务，机器人需要对环境有一定理解能力，而语义地图是机器人理解环境的载体。目前已有的语义地图构建方法不够成熟，往往依赖性能较强的硬件设备，不适用于载荷有限的井下移动机器人。本文就如何提高井下移动机器人的定位精度和语义地图构建的实时性展开研究，主要研究内容如下：

分析视觉SLAM系统在井下环境中应用存在的问题，并提出改进方法。针对ORB-SLAM2算法在井下环境中由于特征跟踪丢失，导致定位与建图精度降低或失败的问题，本文提出了基于ORB-SLAM2的多视觉特征融合算法，该算法借鉴多源融合的思想，将线特征与ORB（Oriented FAST and Rotated BRIEF）特征相融合，提高特征跟踪鲁棒性。仿真结果表明，本文算法与对比算法相比，在光照不良环境和系统剧烈颠簸条件下的特征提取正确率分别提高7.3%和23.8%，机器人定位误差平均降低11.5%。

移动机器人的实时语义地图构建。在获得较高定位精度的基础上，利用邻域最大深度信息查询方法对RGB-D相机获取的深度图像进行处理，完成深度空洞填充，实现较完整的稠密点云地图重建；通过构建两层动态八叉树的点云数据结构，提高点云卷积的实时性和语义关联性，并基于改进的FuseNet实现实时的点云语义分割，完成三维语义地图的构建。在公开数据集和自制井下数据集上对语义地图构建方法进行仿真验证，在分割精度和MinkowskiNet算法相近的基础上达到25帧每秒（FPS）的处理速度。

在煤矿井下环境中进行移动机器人语义建图实验，并以主流的ORB-SLAM2和Structure-SLAM算法为参考进行井下定位精度对比。实验结果表明，本文改进算法定位精度和鲁棒性更高，并且实现煤矿井下环境的语义地图构建。

Mobile robots are increasingly being used in coal mine operations to reduce the risks associated with manual labor and improve efficiency. However, the classic Simultaneous Location and Mapping (SLAM) algorithms often struggle to perform well in complex underground environments. In particular, visual SLAM systems can experience feature tracking loss under conditions of insufficient environmental texture, poor lighting, and severe system vibrations, which can lead to failed localization and mapping, seriously affecting the robot's ability to complete subsequent tasks. To cope with the challenges of operating in underground environments, robots need to have a good understanding of their surroundings, and semantic maps are an important tool for achieving this. Unfortunately, current semantic map construction methods are often not suitable for underground mobile robots with limited payloads, as they rely on powerful hardware devices. This article seeks to address this issue by researching ways to improve the positioning accuracy of underground mobile robots and the real-time construction of semantic maps. The main research content is as follows:

(1) Analyzing the problems of applying visual SLAM systems in underground environments and proposing improvement methods. In response to the issue of reduced accuracy or failure of localization and mapping due to feature tracking loss in the ORB-SLAM2 algorithm in underground environments, this paper proposes a multi-visual feature fusion algorithm based on ORB-SLAM2. The algorithm borrows the idea of multi-source fusion to fuse line features and ORB features, improving feature tracking robustness. Simulation results show that the correct feature extraction rate is improved by 7.3% and 23.8% under conditions of poor lighting and severe system vibrations, respectively, and the robot positioning error is reduced by an average of 11.5%.

(2) Real-time semantic map construction of mobile robots. Building on the higher positioning accuracy achieved in (1), this paper uses the neighborhood maximum depth information query method to process the depth image obtained by the RGB-D camera to complete the depth hole filling and achieve a more complete dense point cloud map reconstruction. By constructing a two-layer dynamic octree point cloud data structure, the real-time and semantic correlation of point cloud convolution are improved, and real-time point cloud semantic segmentation is achieved based on the improved FuseNet to complete the construction of a three-dimensional semantic map. The proposed semantic map construction method is simulated and verified on public datasets and self-made underground datasets, achieving a processing speed of 25 frames per second (FPS) with similar segmentation accuracy to the MinkowskiNet algorithm.

(3) Conducting semantic mapping experiments of mobile robots in coal mine underground environments and comparing underground positioning accuracy with the mainstream ORB-SLAM2 and Structure-SLAM algorithm as references. The experimental results show that the proposed algorithm in this paper has higher positioning accuracy and robustness, and realizes semantic map construction in coal mine underground environments.

[1] 谢和平, 任世华, 谢亚辰, 等. 碳中和目标下煤炭行业发展机遇[J]. 煤炭学报, 2021, 46(7): 2197-2211.

[2] 任满翊. 无人化智能煤矿建设探索与实践[J]. 工矿自动化, 2022, 48(S1): 27-29.

[3] Zhang Y, Tian G, Shao X, et al. Semantic Grounding for Long-Term Autonomy of Mobile Robots Toward Dynamic Object Search in Home Environments[J]. IEEE Transactions on Industrial Electronics, Industrial Electronics, 2022, 70(2): 1655-1665.

[4] Chen W, Zhou C, Shang G, et al. SLAM Overview: From Single Sensor to Heterogeneous Fusion[J]. Remote Sensing, 2022, 14(: 6033-6033.

[5] Zhang H, Sheng V S, Xi X, et al. Overview of RGBD semantic segmentation based on deep learning[J]. Journal of Ambient Intelligence and Humanized Computing, 2022: 1-19.

[6] Brindza J, Kajánek P, Erdélyi J. Lidar-Based Mobile Mapping System for an Indoor Environment[J]. Slovak Journal of Civil Engineering, 2022, 30(2): 47-58.

[7] 吕霖华. 基于视觉的即时定位与地图重建(V-SLAM)综述[J]. 中国战略新兴产业, 2017(4): 67-70.

[8] 李小倩, 何伟, 朱世强, 等. 基于环境语义信息的同步定位与地图构建方法综述[J]. 工程科学学报, 2021, 43(6): 754-767.

[9] 赵喜军. 浅谈如何改善煤矿井下作业环境[J]. 山东煤炭科技, 2017(12): 199-200.

[10] 张恒, 樊晓平, 刘艳丽. 移动机器人同步定位与地图构建研究进展[J]. 数据采集与处理, 2005(4): 458-465.

[11] Rao J, Bian H, Xu X, et al. Autonomous Visual Navigation System Based on a Single Camera for Floor-Sweeping Robot[J]. Applied Sciences (2076-3417), 2023, 13(3): 1562.

[12] 张元勋, 黄靖, 韩亮亮. 星表移动探测机器人研究现状综述[J]. 航空学报, 2021, 42(1): 62-79.

[13] 李延真, 石立国, 徐志根,等. 移动机器人视觉SLAM研究综述[J]. 智能计算机与应用, 2022, 12(7):6.

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

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

[16] Klein G, Murray D. Parallel tracking and mapping on a camera phone[C]//2009 8th IEEE International Symposium on Mixed and Augmented Reality. IEEE, 2009: 83-86.

[17] 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.

[18] Mur-Artal R, Montiel J M M, Tardos J D. ORB-SLAM2: An Open-Source SLAM System for Monocular, Stereo, and RGB-D Cameras[J]. IEEE Transactions on Robotics, 2017, 33(5): 1255-1262.

[19] 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.

[20] 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.

[21] 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.

[22] 郑束蕾, SHULEI Z. 地理空间认知理论与地图工具的发展[J]. 测绘学报, 2021, 50(6): 766.

[23] Nikitina T. Building semantic maps for closely related languages: Words for ‘grain’and their kin in South Mande[J]. Zeitschrift für Sprachwissenschaft, 2022, 41(1): 207-230.

[24] Galindo C, Fernández-Madrigal J A, González J, et al. Robot task planning using semantic maps[J]. Robotics and autonomous systems, 2008, 56(11): 955-966.

[25] Wulf O, Nüchter A, Hertzberg J, et al. Benchmarking urban six‐degree‐of‐freedom simultaneous localization and mapping[J]. Journal of Field Robotics, 2008, 25(3): 148-163.

[26] Pronobis A, Jensfelt P. Large-scale semantic mapping and reasoning with heterogeneous modalities [C]//2012 IEEE international conference on robotics and automation. IEEE, 2012: 3515-3522.

[27] McCormac J, Handa A, Davison A, et al. Semanticfusion: Dense 3d semantic mapping with convolutional neural networks[C]//2017 IEEE International Conference on Robotics and automation (ICRA). IEEE, 2017: 4628-4635.

[28] Narita G, Seno T, Ishikawa T, et al. Panopticfusion: Online volumetric semantic mapping at the level of stuff and things[C]//2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2019: 4205-4212.

[29] Cartillier V, Ren Z, Jain N, et al. Semantic mapnet: Building allocentric semantic maps and representations from egocentric views[C]//Proceedings of the AAAI Conference on Artificial Intelligence. 2021, 35(2): 964-972.

[30] 姬嗣愚, 王永生. 基于单目视觉和激光雷达的测距方法研究[J]. 舰船电子工程, 2022(002):042.

[31] 胡光桃, 张元, 王强. 移动机器人传感器与导航控制[J]. 制造业自动化, 2020, 42(6): 66-70+78.

[32] 陈宇翔, 王仁, 张健. 基于RGB-D相机的SLAM算法优化[J]. 计算机工程与设计, 2020, 41(5): 1445-1451.

[33] Ng P C, Henikoff S. SIFT: Predicting amino acid changes that affect protein function[J]. Nucleic acids research, 2003, 31(13): 3812-3814.

[34] Bay H, Tuytelaars T, Van Gool L. Surf: Speeded up robust features[J]. Lecture notes in computer science, 2006, 3951: 404-417.

[35] 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.

[36] Fularz M, Kraft M, Schmidt A, et al. A high-performance FPGA-based image feature detector and matcher based on the FAST and BRIEF algorithms[J]. International Journal of Advanced Robotic Systems, 2015, 12(10): 141.

[37] Debeunne C, Vivet D. A review of visual-LiDAR fusion based simultaneous localization and mapping[J]. Sensors, 2020, 20(7): 2068.

[38] Liu M, Sheng L, Yang S, et al. Morphing and sampling network for dense point cloud completion[C]// Proceedings of the AAAI conference on artificial intelligence. 2020, 34(07): 11596-11603.

[39] Wiesmann L, Milioto A, Chen X, et al. Deep compression for dense point cloud maps[J]. IEEE Robotics and Automation Letters, 2021, 6(2): 2060-2067.

[40] Long J, Shelhamer E, Darrell T. Fully convolutional networks for semantic segmentation[C]// Proceedings of the IEEE conference on computer vision and pattern recognition. 2015: 3431-3440.

[41] Gomez-Ojeda R, Moreno F A, Zuniga-Noël D, et al. PL-SLAM: A stereo SLAM system through the combination of points and line segments[J]. IEEE Transactions on Robotics, 2019, 35(3): 734-746.

[42] Zuo X, Xie X, Liu Y, et al. Robust visual SLAM with point and line features[C]//2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2017: 1775-1782.

[43] 王丹, 黄鲁, 李垚. 基于点线特征的单目视觉同时定位与地图构建算法[J]. 机器人, 2019, 41(3): 392-403.

[44] Song X, Zhou Z, Guo H, et al. Adaptive retinex algorithm based on genetic algorithm and human visual system[C]//2016 8th International Conference on Intelligent Human-Machine Systems and Cybernetics (IHMSC). IEEE, 2016, 1: 183-186.

[45] Vakhitov A, Lempitsky V. Learnable line segment descriptor for visual slam[J]. IEEE Access, 2019, 7: 39923-39934.

[46] Schubert D, Goll T, Demmel N, et al. The TUM VI benchmark for evaluating visual-inertial odometry [C] //2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2018: 1680-1687.

[47] Hazirbas C , Ma L , Domokos C , et al. FuseNet: Incorporating Depth into Semantic Segmentation via Fusion-Based CNN Architecture[C]// Asian Conference on Computer Vision. Springer, Cham, 2016.

[48] Qi C R, Su H, Mo K, et al. Pointnet: Deep learning on point sets for 3d classification and segmentation [C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 652-660.

[49] Qi C R, Yi L, Su H, et al. Pointnet++: Deep hierarchical feature learning on point sets in a metric space[J]. Advances in neural information processing systems, 2017, 30.

[50] Wu W, Qi Z, Fuxin L. Pointconv: Deep convolutional networks on 3d point clouds[C]//Proceedings of the IEEE/CVF Conference on computer vision and pattern recognition. 2019: 9621-9630.

[51] Wang P S, Liu Y, Guo Y X, et al. O-cnn: Octree-based convolutional neural networks for 3d shape analysis[J]. ACM Transactions On Graphics (TOG), 2017, 36(4): 1-11.

[52] Zhang J, Zhu C, Zheng L, et al. Fusion-aware point convolution for online semantic 3d scene segmentation[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020: 4534-4543.

[53] Huang S S, Ma Z Y, Mu T J, et al. Supervoxel convolution for online 3d semantic segmentation[J]. ACM Transactions on Graphics (TOG), 2021, 40(3): 1-15.

[54] Shi H, Lin G, Wang H, et al. Spsequencenet: Semantic segmentation network on 4d point clouds[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020: 4574-4583.

[55] Tiderko A , Hoeller F , T Röhling. Robot Operating System (ROS); The Complete Reference (Volume 1)[M]. 2016.