基于改进YOLOv8m-PSC的露天煤矿危险驾驶行为检测系统

Hazardous driving behavior detection system of open-pit coal mine based on improved YOLOv8m-PSC

  • 摘要: 露天煤矿环境复杂多变,安全生产面临诸多挑战,其中危险驾驶行为是威胁矿区安全生产的重要因素之一。矿区内重型机械和运输车辆频繁作业,由于道路条件复杂、作业环境恶劣,驾驶员一旦出现疏忽或违规操作,极易引发重大事故,造成严重的人员伤亡和经济损失。目前,矿区通常采用人工巡检和车载视频回放检查等传统安全监控手段,但这些方法存在效率低、实时性差,耗费大量人力且有监控盲区等缺点,难以有效应对由于危险驾驶行为而导致的事故风险。随着深度学习和计算机视觉技术的迅速发展,利用先进技术手段提升矿区危险驾驶行为检测效率成为重要方向。相比传统手段,智能检测系统能够实现对危险行为的实时检测和预警,不仅提高了监控覆盖率,还有效减少了人力资源的投入。因此,设计并开发适用于露天煤矿环境的危险驾驶行为检测系统,不仅是当前矿区安全管理的迫切需求,也是推动矿山智能化发展的重要方向。在现有YOLOv8m模型的基础上,针对现有模型的参数量大、计算量高及夜晚车内光线较差导致检测精度低等问题,开展了面向露天煤矿危险驾驶行为检测方法研究,运用对比实验、消融实验及理论分析等方法,提出了一种结合部分卷积、自注意力机制及通道混洗的目标检测改进模型YOLOv8m-PSC。在此基础上,设计并实现了基于改进模型的危险驾驶行为检测系统。系统通过PySide6构建直观简便的图形用户界面,能够快速部署到实际生产环境中,弥补了传统监控手段的不足,对露天矿危险驾驶行为进行快速且准确的监控。实验结果表明,改进后的模型在性能上有显著提升:危险驾驶行为的平均检测精度达到83.7%,参数量减少了40.2%,浮点计算量下降了62.6%,推理速度提升了10.7%,实现了在露天矿复杂场景下对疲劳驾驶、分心驾驶等危险驾驶行为高效、实时的检测。改进模型的提出不仅在技术上实现了目标检测模型的进一步优化,也为危险驾驶行为的实时检测提供了理论和实践的参考。此外,系统开发流程中引入了图形界面设计思路,为智慧矿山建设中的人机交互系统设计提供了创新性方案。

     

    Abstract: Open-pit coal mines are characterized by complex environments that pose significant safety challenges. Hazardous driving behaviors among heavy machinery and transport vehicles are a major threat to operational safety, often leading to severe accidents and substantial losses. Factors such as poor road conditions and harsh working environments exacerbate these risks. At present, traditional safety monitoring methods such as manual inspection and vehicle-mounted video replay inspection are usually adopted in mining areas. However, these methods have disadvantages such as low efficiency, poor real-time performance, large manpower consumption and blind monitoring areas, which make it difficult to effectively deal with accident risks caused by hazardous driving behaviors. Advances in deep learning and computer vision offer promising solutions for improving the detection of hazardous driving behaviors. Compared to conventional methods, advanced detection systems enable real-time monitoring and alerts, enhancing coverage while reducing labor demands. Developing a detection system tailored to the conditions of open-pit mines is critical for ensuring safety and advancing intelligent mining practices. High parameter load, computational demands, and poor light inside the car at night leading to low detection accuracy in existing detection models are addressed. An improved object detection model, YOLOv8m–PSC, is proposed, incorporating partial convolution, self-attention mechanisms, and channel shuffle. A detection system based on the improved model is designed with a user-friendly graphical interface developed using PySide6, enabling rapid deployment in real-world scenarios and overcoming limitations of traditional monitoring methods. Experimental results show significant performance improvements: an average detection accuracy of 83.7% for hazardous behaviors, a 40.2% reduction in parameters, a 62.6% decrease in FLOPs, and a 10.7% increase in inference speed compared to the original model. The system efficiently detects fatigue and distracted driving in complex mining scenarios. The improved model advances object detection technology and provides practical insights for real-time hazardous behavior monitoring. Additionally, the system’s GUI design contributes innovative approaches to human-machine interaction in intelligent mining.

     

/

返回文章
返回