毕业生知识库

矿区开采运输量大、工作环境恶劣、安全性差，自动驾驶技术可以保障车辆在复杂矿区环境下的安全、高效、节能运行，从而显著提升传统采矿业的运输自动化水平，促进矿业开采的转型升级。高精地图是自动驾驶的重要组成部分，可以为自动驾驶矿车提供丰富准确的先验信息，进而提升自动驾驶矿车面对矿区复杂工况与环境的适应能力。本文重点研究高精地图构建的两项核心技术，即定位和建图技术。在定位方面，井工矿环境由于具有无GNSS信号、弱纹理、重复特征等特点，成为自动驾驶矿车定位的挑战性场景。目前，井下车辆定位方法存在前期准备周期长、依赖人工操作、价格昂贵等问题。在建图方面，目前存在尚无面向矿区场景的地图表示方法，以及缺乏面向矿区自动驾驶的地图生成方案等问题。针对上述问题，本文开展了相关研究，主要创新点如下：

1）针对井工矿场景下的自动驾驶矿车定位问题，提出了一种井下同时定位与地图构建（SLAM）方法。首先，提出了一种基于退化检测的激光-惯性-轮速融合里程计，解决了长巷道场景造成的激光里程计退化问题，提高了位姿估计的精度和鲁棒性。然后，提出了一种基于深度学习的显著地点识别方法，降低了虚假闭环的发生率，提高了闭环检测的准确率和鲁棒性。最后，设计了一种多地图融合方案，提高了大规模井工矿场景下的建图效率。实验结果表明，该方法在井工矿环境下取得了良好的定位性能，为矿区高精地图的构建提供了必要的基础。

2）设计了一种基于分层表示的矿区地形建图方法。首先，根据矿区环境的特点，提出了一种矿区语义地图的表示方法，将语义点云地图、多通道栅格地图和三角网格地图整合到一个可扩展的地形建图框架中。然后，为了满足实际应用需求，提出了上述三种地图的构建方法。最后，提出了基于PostgreSQL数据库的持久化方案，实现了对不同地图表示的统一维护，该方案具有良好的灵活性和可扩展性。实验结果表明，该方法能够生成描述矿区复杂地形的语义地图。

3）基于上述研究，提出了一种矿区高精地图构建方法。为提高地图的描述能力，基于矿区语义地图和LaneLet2矢量地图，提出了一种能够描述矿区复杂环境的高精地图表示方法，实现了对矿区环境和道路的建模。基于该表示，提出了一个完整的生成工作流，实现了矿区高精地图的高效低成本构建。最后，在真实矿区中验证了该方法的有效性，为矿区高精地图的应用提供了一种解决方案。

Mines are characterized by a high volume of mining transportation, a harsh working environment and poor safety. Autonomous driving technology can guarantee the safe, efficient and energy-saving operation of vehicles in complex mine environments, thus significantly improving the level of transportation automation in the traditional mining industry and promoting the transformation and upgrading of mining exploitation. High-precision map is an important component of autonomous driving, which can provide rich and accurate prior information for autonomous mine vehicles, thus enhancing the adaptability of autonomous mine vehicles in the face of complex working conditions and environment in the mine. This work focuses on two core technologies for high-precision map construction, namely positioning and map building methods. In terms of positioning, the underground mine environment has become a challenging scene for the positioning of autonomous mine vehicles due to its characteristics, such as no GNSS signal, weak texture, and repetitive features. Currently, underground vehicle positioning technology has problems such as long pre-preparation periods, reliance on manual operation and high prices. Regarding map building, there are problems, such as the lack of map representation methods for mine scenes and the lack of map generation solutions for autonomous driving in mines. In response to the above problems, this work carried out a relevant study with the following main innovations:

1) An underground simultaneous localization and map building (SLAM) technology is proposed for the positioning problem of autonomous mine vehicles in underground mine scenes. Firstly, a laser-inertial-wheel fusion odometry based on degradation detection is proposed to solve the laser odometry degradation problem caused by long alleyway scenes and improve the accuracy and robustness of the pose estimation. Then, a deep learning-based salient place identification method is proposed to reduce the incidence of spurious loop closures and improve the accuracy and robustness of loop closure detection. Finally, a multi-map fusion scheme is designed to improve the efficiency of building maps in large-scale underground mine scenes. The experimental results show that the technology achieves good positioning performance in the underground mine environment and provides the necessary basis for the construction of high-precision maps in mines.

2) A terrain mapping method was designed based on a layered representation for mines. First, based on the characteristics of the mine environment, a semantic map representation method for mines is proposed, integrating semantic point cloud maps, multi-channel grid maps and mesh maps into a scalable terrain mapping framework. Then, to meet the needs of practical applications, methods of building the three maps mentioned above are proposed. Finally, a persistence scheme based on the PostgreSQL database is proposed to achieve unified maintenance of different map representations with good flexibility and scalability. The experimental results show that the method can generate semantic maps describing the complex terrain of mines.

3) Based on the above research, a high-precision map construction technology for mines is proposed. In order to improve the descriptive capability of the map, a high-precision map representation method capable of describing the complex environment of mines is proposed based on the semantic map of the mines and LaneLet2 vector map, and the modeling of the mine environment and roads is realized. Based on this representation, a complete generation workflow is proposed to realize the efficient and low-cost construction of high-precision maps of mines. Finally, the effectiveness of the technology was verified in an actual mine, providing a solution for the application of high-precision maps in mines.