基于机器视觉的罐笼人员异常行为检测算法研究  

Research on Abnormal Behavior Detection Algorithm for Cage Personnel Based on Machine Vision

作　　者：王哲 潘东亚 李青 傅哲 张斌[3] 詹雨衡 WANG Zhe;PAN Dongya;LI Qing;FU Zhe;ZHANG Bin;ZHAN Yuheng(CCTEG Shenyang Research Institute,Fushun 113122,China;CCTEG Fushun Testing Center Co.,Ltd.,Fushun 113122,China;Xi’an Jiaotong University,Xi’an 710049,China;Xi’an Bossun Security Technology Co.,Ltd.,Xi’an 710304,China;Beijing Institute of Technology,Beijing 100081,China)

机构地区：[1]中煤科工集团沈阳研究院有限公司,辽宁抚顺113122 [2]抚顺中煤科工检测中心有限公司,辽宁抚顺113122 [3]西安交通大学,陕西西安710049 [4]西安博深安全科技股份有限公司,陕西西安710304 [5]北京理工大学,北京100081

出　　处：《煤矿机电》2024年第5期14-18,21,共6页Colliery Mechanical & Electrical Technology

摘　　要：煤矿罐笼是矿工进出矿井的主要运输工具,承担着从地面到地下数百米深的运载任务。由于罐笼在运行过程中存在速度快、空间有限等特点,矿工在罐笼内站立不稳、意外摔倒等行为可能导致严重的安全事故。提出了一种基于人员行为检测的罐笼辅助系统,以提升煤矿作业的安全性,特别是保障煤矿工人在井下和运输过程中的安全。系统利用IP摄像头实时监控罐笼内部,通过边缘计算设备实现目标检测和行为识别算法,及时发现和处理矿工的异常行为。采用轻量级的MobileNetv2和改进的TSM模型,并增加多尺度区域特征融合模块(MRFA),以提升行为识别的精确性和鲁棒性。试验结果表明,改进后的TSM模型在准确率、参数量和推理速度方面均优于传统行为识别算法,特别是在计算资源有限的边缘设备上表现出更高的效率和可靠性。改进的TSM模型在矿工行为数据集上的准确率达99.30%,显著高于其他主流行为识别算法,具有较高的实用价值和广泛的应用前景。The coal mine cage is the main transportation tool for miners to enter and exit the mine,undertaking the task of transporting from the ground to hundreds of meters deep underground.Due to the fast speed and limited space of the cage during operation,miners may experience serious safety accidents such as unstable standing and accidental falls inside the cage.A cage assistance system based on personnel behavior detection was proposed to enhance the safety of coal mining operations,especially to ensure the safety of coal miners underground and during transportation.The system uses IP cameras to monitor the inside of the cage in real time,and realizes target detection and behavior recognition algorithms through edge computing equipment,so as to detect and deal with abnormal behaviors of miners in a timely manner.Adopting lightweight MobileNetv2 and improved TSM model,and adding multi-scale region feature fusion module(MRFA)to improve the accuracy and robustness of behavior recognition.The experimental results showed that the improved TSM model outperformed traditional behavior recognition algorithms in terms of accuracy,parameter count,and inference speed,especially exhibiting higher efficiency and reliability on edge devices with limited computing resources.The improved TSM model achieves an accuracy of 99.30%on the miner behavior dataset,significantly higher than other mainstream behavior recognition algorithms,and has high practical value and broad application prospects.

关 键 词：行为识别 边缘计算 罐笼 异常告警 矿井安全 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]