张键,丰继林,袁静,周涵,刘祖阳








摘 要: 液压支架护帮板的打开和闭合是煤矿井下的主要作业之一。为了自动识别监控视频中每个护帮板的工作状态,需研究液压支架护帮板工作状态智能识别算法。结合深度学习和计算机视觉的算法,可采用融合幅度信息的光流直方图(Histograms of Oriented Optical Flow,HOF)提取运动特征。实验结果表明:算法的准确率达到88.69%、精确率达到79.08%、召回率达到76.16%、F1_score值达到76.53%,fps值达到18帧/s,验证了该算法的可行性和有效性。
关键词: 液压支架护帮板; 光流直方图; 运动识别; 计算机视觉; 煤矿作业
中图分类号:TP391.41 文献标识码:A 文章编号:1006-8228(2022)06-23-04
Detection algorithm for working state of hydraulic support guard plate
Zhang Jian, Feng Jilin, Yuan Jing, Zhou Han, Liu Zuyang
(Institute of Disaster Prevention, Langfang, Hebei 062541, China)
Abstract: The opening and closing of the hydraulic support guard plate is one of the main operations in coal mines. In order to automatically identify the working state of each plate in the monitoring video, an intelligent recognition algorithm is studied. Combining deep learning and computer vision algorithms, motion features are extracted using Histograms of Oriented Optical Flow (HOF) fused with amplitude information. The experimental results show that the accuracy rate reaches 88.69%, the precision rate reaches 79.08%, the recall rate reaches 76.16%, the F1 score value reaches 76.53%, and the fps value reaches 18 frames/s, which verifies the feasibility and effectiveness of the algorithm.
Key words: hydraulic support guard plate; Histograms of Oriented Optical Flow; motion recognition; computer vision; coal mine operations
0 引言
為实现煤矿生产无人化,以最大程度的保障人员生产安全,越来越多的矿井安装了大量摄像头用于远程监管煤矿生产活动、规范生产行为,其中液压支架护帮板是保障煤矿生产安全的重要手段之一,其主要功能是防止煤壁片帮,对其工作状态进行实时监控是减少事故率的关键。图1为液压支架的结构简图,而本文识别的主要区域为图1中的装置1:护帮装置的护帮板。
目前,国内外基于液压支架护帮板运动状态非接触式智能识别的研究也主要集中在图像处理和计算机视觉领域。比如,满溢桥[1]联合图像增强技术和护帮板位姿解算模型设计了一套识别算法,该算法监测护帮板图像误差较小,但是邻架护帮板的运动会影响其识别性能,而且需要大量的破坏性实验进行验证,因此,该算法距离实际应用尚且有一定差距。……