针对在井工矿辅助运输装备作业复杂环境下,传统车道线检测算法精度低、鲁棒性差等问题,文章提出一种基于深度学习的井工矿车道线检测算法,采用UFLD行索引车道线检测算法,并根据井下实际道路场景,自定义标注车道线规则,建立一种涵盖非结构道路下的车道线数据集,从而实现对井下非结构性道路的车道线检测任务。相较于传统车道线检测算法,该算法充分利用了深度学习中先验知识的有效信息,实现了对不同道路场景下的车道线检测任务的兼容性。通过分析基于自建数据集模型训练及检测指标的评估结果表明,该算法提高了对井下车辆车道线检测任务的能力,为井下辅助运输车辆智能驾驶系统的高阶功能应用奠定了基础,且在提高检测精度的同时,保证了其在检测非结构车道场景车道线任务中的鲁棒性,并通过TensorRT架构进行模型推理加速及部署,验证了其在井工矿场景的有效性和工程化部署的可行性。

Aiming at the low accuracy and poor robustness of traditional lane line detection algorithms under the complex environment of auxiliary transportation equipment operation in underground mines, we propose a deep learning-based lane line detection algorithm for underground mines, which adopts the UFLD row-indexed lane line detection algorithm and customizes the labeled lane line rules according to the actual road scenes in underground mines to establish a lane line dataset that covers non-structural roads under the lane line detection task for underground non-structural roads. Compared with the traditional lane line detection algorithm, it makes full use of the effective information of prior knowledge in deep learning to achieve compatibility with the lane line detection task under different road scenes. The analysis of the evaluation results based on the model training and detection metrics of the self - built dataset shows that the algorithm improves the capability for the lane line detection of underground vehicles, lays the foundation for the application of higher - order functions of the intelligent driving system of underground assisted transportation vehicles, and ensures its robustness in the task of detecting lane lines in unstructured lane scenarios while improving the detection accuracy, and is carried out through the TensorRT architecture to model inference acceleration and deployment, which verifies its effectiveness and feasibility of engineering deployment in well mining scenarios.