火电厂监控视频安全服检测方法研究

doi:10.11996/JG.j.2095-302X.2023020291

图学学报 ›› 2023, Vol. 44 ›› Issue (2): 291-297.DOI: 10.11996/JG.j.2095-302X.2023020291

• 图像处理与计算机视觉 • 上一篇下一篇

火电厂监控视频安全服检测方法研究

陈刚¹(), 张培基², 龚冬冬², 于俊清²()

1.国电汉川发电有限公司，湖北汉川 431614
2.华中科技大学网络与计算中心，湖北武汉 430074

收稿日期:2022-08-12 接受日期:2022-10-15 出版日期:2023-04-30 发布日期:2023-05-01
通讯作者: 于俊清(1975-)，男，教授，博士。主要研究方向为智能媒体计算、网络安全与信息化。E-mail：yjqing@hust.edu.cn
作者简介:陈刚(1974-)，男，工程师，本科。主要研究方向为火力发电管理。E-mail：735091398@qq.com
基金资助:
国电科技项目(hckj-2020-03)

Research on safety clothing detection method for surveillance video of thermal power plant

CHEN Gang¹(), ZHANG Pei-ji², GONG Dong-dong², YU Jun-qing²()

1. Hanchuan Power Generation Co., Ltd., China Guodian Corporation, Hanchuan Hubei 431614, China
2. Center of Network & Computation, Huazhong University of Science and Technology, Wuhan Hubei 430074, China

Received:2022-08-12 Accepted:2022-10-15 Online:2023-04-30 Published:2023-05-01
Contact: YU Jun-qing (1975-), professor, Ph.D. His main research interests cover intelligent media computing, network security and informatization, etc. E-mail：yjqing@hust.edu.cn
About author:CHEN Gang (1974-), engineer, undergraduate. His main research interest covers thermal power generation management. E-mail：735091398@qq.com
Supported by:
State Grid Corporation of China Technology Project(hckj-2020-03)

摘要/Abstract

摘要：

在智慧电厂建设中，通常采用计算机视觉技术对部署在工业厂房的监控摄像头传回的监控视频进行检测，监控工人是否规范穿着安全服，由于火电厂中场景较为复杂，直接使用现有数据集与算法，模型准确度无法满足使用要求。基于工业监控视频构建了火电厂场景中的安全服目标检测数据集。针对YOLOv5对安全服目标检测准确率较低的问题，使用EfficientNet，ResNet-50，ShuffleNet与MobileNet等多种算法模型替换原YOLOv5的Backbone模块网络结构，提出了基于YOLOv5的模型融合算法。基于构建的火电厂安全服目标检测数据集，选取目前相关领域的最优算法，与优化改进后的YOLOv5算法进行对比实验。实验结果表明，改进后的YOLOv5+EfficientNet算法在火电厂安全服数据集中的准确率得到显著提升，最高检测准确率达到96.6%。

关键词: 火电厂, 监控视频, 安全服, 目标检测, YOLOv5

Abstract:

In the construction of a smart power plant, computer vision technology is usually utilized to detect the surveillance video returned by surveillance cameras deployed in industrial plants to monitor whether workers are wearing safety clothing. However, due to the complexity of the scenes in the thermal power plant, the existing datasets and algorithms proved inadequate in achieving the desired level of accuracy. Based on the industrial surveillance video, a dataset for safety clothing target detection in the thermal power plant scene was constructed. To address the problem of YOLOv5 being insufficient in the detection accuracy for safety clothing targets, various algorithm models such as EfficientNet, ResNet-50, ShuffleNet, and MobileNet were implemented as alternatives to the original YOLOv5 Backbone module network structure. Additionally, model fusion algorithms based on YOLOv5 were proposed. Based on the industrial scene safety clothing target detection dataset, the optimal algorithms in related fields were selected and compared with the optimized YOLOv5 algorithm. The experimental results demonstrated that the accuracy of the improved YOLOv5+EfficientNet algorithm in the industrial scene security service dataset has been significantly improved, with the highest detection accuracy of 96.6%.

Key words: thermal power plant, surveillance video, safety clothing, object detection, YOLOv5

中图分类号:

TP391

陈刚, 张培基, 龚冬冬, 于俊清. 火电厂监控视频安全服检测方法研究[J]. 图学学报, 2023, 44(2): 291-297.

CHEN Gang, ZHANG Pei-ji, GONG Dong-dong, YU Jun-qing. Research on safety clothing detection method for surveillance video of thermal power plant[J]. Journal of Graphics, 2023, 44(2): 291-297.

图/表 11

图1 YOLOv5模型网络结构

Fig. 1 YOLOv5 model network structure

图2 YOLOv5模型优化流程

Fig. 2 Optimization process of YOLOv5 model

图3 模型放缩参数

Fig. 3 Model scaling parameters ((a) Baseline; (b) Width scaling; (c) Depth scaling; (d) Resolution scaling; (e) Compound scaling)

图4 YOLOv5+ResNet50

Fig. 4 YOLOv5+ResNet50

图5 MobileNet思想

Fig. 5 Mobilenet idea

图6 监控视频ROI片段

Fig. 6 ROI clip of surveillance video

表1 安全服穿着检测对比实验结果

Table 1 Safety clothing wearing detection and comparison test results

Method	Backbone	P	R	mAP@.5	mAP@.5:.95
文献[2]	Darknet-53	0.883	0.905	0.880	-
文献[3]	VGG16	-	-	0.861	-
文献[3]	ResNet-101	-	-	0.858	-
文献[5]	CSPDarknet-53	0.872	0.896	0.907	-
YOLOv5	CSPDarknet-53	0.886	0.908	0.929	0.701
YOLOv5	ResNet-50	0.911	0.859	0.935	0.671
YOLOv5	ShuffleNet V2	0.911	0.852	0.921	0.639
YOLOv5	MobileNet V3	0.904	0.855	0.933	0.657
YOLOv5	EfficientNet	0.910	0.879	0.951	0.707

表2 安全服穿着检测ORE对比实验结果

Table 2 Safety clothing wearing detection ORE comparison test results

Method	Backbone	P	R	mAP@.5	mAP@.5:.95
文献[2]	Darknet-53	0.881	0.933	0.908	-
文献[5]	CSPDarknet-53	0.890	0.923	0.935	-
YOLOv5	CSPDarknet-53	0.900	0.935	0.919	0.698
YOLOv5	ResNet-50	0.907	0.928	0.962	0.748
YOLOv5	ShuffleNet V2	0.911	0.917	0.960	0.723
YOLOv5	MobileNet V3	0.925	0.914	0.963	0.722
YOLOv5	EfficientNet B8	0.944	0.915	0.966	0.761

表3 安全服穿着规范性检测对比实验结果

Table 3 Safety clothing wearing normative testing and comparative test results

Method	Backbone	P	R	mAP@.5	mAP@.5:.95
文献[2]	Darknet-53	0.791	0.912	0.899	-
文献[3]	VGG16	-	-	0.915	-
文献[3]	ResNet-101	-	-	0.904	-
文献[5]	CSPDarknet-53	0.918	0.939	0.949	-
YOLOv5	CSPDarknet-53	0.937	0.951	0.965	0.734
YOLOv5	ResNet-50	0.924	0.954	0.969	0.701
YOLOv5	ShuffleNet V2	0.954	0.933	0.971	0.781
YOLOv5	MobileNet V3	0.936	0.921	0.964	0.788
YOLOv5	EfficientNet B8	0.933	0.945	0.966	0.815

图7 安全服穿着状态检测结果((a), (c)算法改进前检测结果；(b), (d)算法改进后检测结果)

Fig. 7 Safety clothing wearing status test results ((a), (c) Detection results before algorithm improvement; (b), (d) Detection results after algorithm improvement)

图8 生产环境中安全服状态检测结果((a)穿着安全服检测结果；(b)未穿着安全服检测结果)

Fig. 8 Safety clothing wearing status test results in the production environment ((a) Detection result of wearing safety clothes; (b) Detection result without safety clothes)

参考文献 13

[1]	HU J L, LIN X B, LI C H, et al. Protective clothing detection of substation workers using S-HOG+C operator[C]// 2018 International Conference on Power System Technology. New York: IEEE Press, 2019: 4181-4185.
[2]	WANG X Y, NIU D, LUO P X, et al. A safety helmet and protective clothing detection method based on improved-yolo V₃[C]// 2020 Chinese Automation Congress. New York: IEEE Press, 2021: 5437-5441.
[3]	陈双, 何利力. 基于Faster R-CNN的服装目标检测改进方法[J]. 软件导刊, 2020, 19(4): 42-45.
	CHEN S, HE L L. Improved method for garment target detection based on faster R-CNN[J]. Software Guide, 2020, 19(4): 42-45. (in Chinese)
[4]	DE SOUZA INÁCIO A, LOPES H S. EPYNET: efficient pyramidal network for clothing segmentation[J]. IEEE Access, 2020, 8: 187882-187892. DOI URL
[5]	LEE C H, LIN C W. A two-phase fashion apparel detection method based on YOLOv4[J]. Applied Sciences, 2021, 11(9): 3782. DOI URL
[6]	WANG Z J, WU Y M, YANG L C, et al. Fast personal protective equipment detection for real construction sites using deep learning approaches[J]. Sensors: Basel, Switzerland, 2021, 21(10): 3478.
[7]	KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84-90. DOI URL
[8]	SAUDI M M, HAKIM A, AHMAD A, et al. Image detection model for construction worker safety conditions using faster R-CNN[J]. International Journal of Advanced Computer Science and Applications, 2020, 11(6): 246.
[9]	HUMPHREY E J, BELLO J P. Rethinking automatic chord recognition with convolutional neural networks[C]// The 11th International Conference on Machine Learning and Applications. New York: IEEE Press, 2013: 357-362.
[10]	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]// The IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2016: 770-778.
[11]	MA N N, ZHANG X Y, ZHENG H T, et al. ShuffleNet V2: practical guidelines for efficient CNN architecture design[C]// Computer Vision - ECCV 2018:15th European Conference. New York: ACM, 2018: 122-138.
[12]	HOWARD A, SANDLER M, CHEN B, et al. Searching for MobileNetV3[C]// 2019 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2020: 1314-1324.
[13]	ZHONG Z, ZHENG L, KANG G L, et al. Random erasing data augmentation[C]// The AAAI Conference on Artificial Intelligence. Palo Alto: AAAI Press, 2017: 1-11.

[1]	李利霞, 王鑫, 王军, 张又元 . 基于特征融合与注意力机制的无人机图像小目标检测算法 [J]. 图学学报, 2023, 44(4): 658-666.
[2]	郝帅, 赵新生, 马旭, 张旭, 何田, 侯李祥. 基于 TR-YOLOv5 的输电线路多类缺陷目标检测方法 [J]. 图学学报, 2023, 44(4): 667-676.
[3]	邓渭铭 , 杨铁军 , 李纯纯 , 黄琳 . 基于神经网络架构搜索的铭牌目标检测方法[J]. 图学学报, 2023, 44(4): 718-727.
[4]	胡欣, 周运强, 肖剑, 杨杰. 基于改进YOLOv5的螺纹钢表面缺陷检测[J]. 图学学报, 2023, 44(3): 427-437.
[5]	毛爱坤, 刘昕明, 陈文壮, 宋绍楼. 改进YOLOv5算法的变电站仪表目标检测方法[J]. 图学学报, 2023, 44(3): 448-455.
[6]	郝鹏飞, 刘立群, 顾任远. YOLO-RD-Apple果园异源图像遮挡果实检测模型[J]. 图学学报, 2023, 44(3): 456-464.
[7]	罗文宇, 傅明月. 基于YoloX-ECA模型的非法野泳野钓现场监测技术[J]. 图学学报, 2023, 44(3): 465-472.
[8]	王佳婧, 王晨, 朱媛媛, 王笑梅. 基于民国纸币的图元素匹配检索[J]. 图学学报, 2023, 44(3): 492-501.
[9]	成浪, 敬超. 基于改进YOLOv7的X线图像旋转目标检测[J]. 图学学报, 2023, 44(2): 324-334.
[10]	曹义亲, 伍铭林, 徐露. 基于改进YOLOv5算法的钢材表面缺陷检测[J]. 图学学报, 2023, 44(2): 335-345.
[11]	张伟康, 孙浩, 陈鑫凯, 李叙兵, 姚立纲, 东辉. 基于改进YOLOv5的智能除草机器人蔬菜苗田杂草检测研究[J]. 图学学报, 2023, 44(2): 346-356.
[12]	李小波 , 李阳贵 , 郭宁 , 范震 . 融合注意力机制的 YOLOv5 口罩检测算法[J]. 图学学报, 2023, 44(1): 16-25.
[13]	皮骏, 刘宇恒, 李久昊. 基于 YOLOv5s 的轻量化森林火灾检测算法研究[J]. 图学学报, 2023, 44(1): 26-32.
[14]	谷雨, 赵军. 列车闸瓦钎及闸瓦故障图像检测算法研究 [J]. 图学学报, 2023, 44(1): 88-94.
[15]	崔振东, 李宗民, 杨树林, 刘玉杰, 李华. 基于语义分割引导的三维目标检测[J]. 图学学报, 2022, 43(6): 1134-1142.

火电厂监控视频安全服检测方法研究

Research on safety clothing detection method for surveillance video of thermal power plant

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 11

参考文献 13

相关文章 15

编辑推荐

Metrics

本文评价