深度学习下建筑工人高空安全防护装备检测方法

doi:10.16265/j.cnki.issn1003-3033.2022.05.1141

摘要/Abstract

摘要：

为解决现有安全防护装备检测技术动态性、协同性不足的问题,基于深度学习,提出一种建筑工人高空作业安全防护装备的检测新方法;该方法以轻量化网络MobileNetV2置换YOLOv4的主干特征提取网络,实现动态视频状态下安全帽与安全带的综合检测;并开展测试,检验该方法的有效性。结果表明:该检测方法在中央处理器(CPU)运行环境下,检测速度增快2.7倍;在图形处理器(GPU)运行环境下,对于单目标、多目标和小目标的单帧视频检测速度能保持在25~27 ms, 同时能取得91.57%、89.69%和86.63%的平均精度。

关键词: 深度学习, 建筑工人, 高空作业, 安全防护装备, 动态视频检测

Abstract:

In order to address the lack of dynamics and coordination in current detection technology of safety protection equipment, a new detection method for construction workers working at height was proposed based on deep learning. Comprehensive detection of helmet and seat belt in state of dynamic videos was realized by replacing backbone feature extraction network of YOLOv4 with lightweight network MobileNetV2. Then, tests were carried out to verify effectiveness of the method. The results show that detection speed of the proposed method is increased by 2.7 times in central processing unit (CPU) operating environment, and single-frame video detection speed for single target, multi-target and small target can be maintained between 25-27 milliseconds in graphics processing unit (GPU) operating environment. At the same time, mean average precision rate of 91.57%, 89.69% and 86.63% can be achieved.

Key words: deep learning, construction worker, working at height, safety protection equipment, dynamic video detection

张萌, 韩豫, 刘泽锋. 深度学习下建筑工人高空安全防护装备检测方法[J]. 中国安全科学学报, 2022, 32(5): 140-146.

ZHANG Meng, HAN Yu, LIU Zefeng. Detection method of high-altitude safety protective equipment for construction workers based on deep learning[J]. China Safety Science Journal, 2022, 32(5): 140-146.

图/表 10

图1

图2

表1

图3

表2

表3

表4

表5

图4

图5

参考文献 23

[1]	张建设, 靳静, 张晶然. 2004年-2013年建筑业高处坠落死亡事故统计分析研究[J]. 建筑安全, 2015, 30(6):42-45.
[2]	孙世梅, 赵金坤, 傅贵. 基于“2-4”模型的高处坠落事故行为原因研究[J]. 中国安全科学学报, 2019, 29(8):23-28. doi: 10.16265/j.cnki.issn1003-3033.2019.08.004
	SUN Shimei, ZHAO Jinkun, FU Gui. Study on behavioral causes of falling accidents based on " 2-4" model[J]. China Safety Science Journal, 2019, 29(8):23-28. doi: 10.16265/j.cnki.issn1003-3033.2019.08.004
[3]	边星, 晋良海, 陈雁高, 等. 施工作业人员佩戴安全帽行为意向研究[J]. 中国安全科学学报, 2016, 26(11):43-48.
	BIAN Xing, JIN Lianghai, CHEN Yangao, et al. Research on workers' behavioral intention of wearing safety helmet[J]. China Safety Science Journal, 2016, 26(11):43-48.
[4]	佟瑞鹏, 张艳伟. 人工智能技术在矿工不安全行为识别中的融合应用[J]. 中国安全科学学报, 2019, 29(1):7-12. doi: 10.16265/j.cnki.issn1003-3033.2019.01.002
	TONG Ruipeng, ZHANG Yanwei. Integration between artificial intelligence technologies for miners' unsafe behavior identification[J]. China Safety Science Journal, 2019, 29(1):7-12. doi: 10.16265/j.cnki.issn1003-3033.2019.01.002
[5]	ZHANG Mingyuan, CAO Zhiying, YANG Zhen, et al. Utilizing computer vision and fuzzy inference to evaluate level of collision safety for workers and equipment in a dynamic environment[J]. Journal of Construction Engineering and Management, 2020, 146(6): DOI: 10.1061/(ASCE)CO.1943-7862.0001802. doi: 10.1061/(ASCE)CO.1943-7862.0001802
[6]	PARK M W, ELSAFTY N, ZHU Zhenhua. Hardhat-wearing detection for enhancing on-site safety of construction workers[J]. Journal of Construction Engineering and Management, 2015, 141(9): DOI: 10.1061/(ASCE)CO.1943-7862.0000974. doi: 10.1061/(ASCE)CO.1943-7862.0000974
[7]	刘晓慧, 叶西宁. 肤色检测和Hu矩在安全帽识别中的应用[J]. 华东理工大学学报:自然科学版, 2014, 40(3):365-370.
	LIU Xiaohui, YE Xining. Skin color detection and Hu moments in helmet recognition research[J]. Journal of East China University of Science and Technology:Natural Science Edition, 2014, 40(3):365-370.
[8]	LECUN Y, BENGIO Y, HINTON G. Deep learning[J]. Nature, 2015, 521(7553): 436-444. doi: 10.1038/nature14539
[9]	KIM M H, SHIN S W, SUH Y Y. Application of deep learning algorithm for detecting construction workers wearing safety helmet using computer vision[J]. Journal of the Korean Society of Safety, 2019, 34(6):29-37.
[10]	FANG Weili, DING Lieyun, LUO Hanbin, et al. Falls from heights: a computer vision-based approach for safety harness detection[J]. Automation in Construction, 2018, 91:53-61. doi: 10.1016/j.autcon.2018.02.018
[11]	张博, 宋元斌, 熊若鑫, 等. 融合人体关节点的安全帽佩戴检测[J]. 中国安全科学学报, 2020, 30(2):177-182. doi: 10.16265/j.cnki.issn1003-3033.2020.02.028
	ZHANG Bo, SONG Yuanbin, XIONG Ruoxin, et al. Helmet-wearing detection considering human joint[J]. China Safety Science Journal, 2020, 30(2):177-182. doi: 10.16265/j.cnki.issn1003-3033.2020.02.028
[12]	肖体刚, 蔡乐才, 高祥, 等. 改进YOLOv3的安全帽佩戴检测方法[J]. 计算机工程与应用, 2021, 57(12):216-223.
	XIAO Tigang, CAI Lecai, GAO Xiang, et al. Improved YOLOv3 helmet wearing detection method[J]. Computer Engineering and Applications, 2021, 57(12):216-223.
[13]	CHENG Rao, HE Xiaowei, ZHENG Zhonglong, et al. Multi-scale safety helmet detection based on SAS-YOLOv3-Tiny[J]. Applied Sciences, 2021, 11(8): DOI: 10.3390/app1108. doi: 10.3390/app1108
[14]	REN Shaoqing, HE Kaiming, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39(6):1137-1149.
[15]	LIU Wei, ANGUELOV D, ERHAN D, et al. SSD: single shot MultiBox detector[C]. European Conference on Computer Vision, 2016:21-37.
[16]	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once:unified, real time object detection[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition(CVPR), 2016:779-788.
[17]	BOCHKOVSKIY A, WANG C Y, LIAO H M. YOLOv4: optimal speed and accuracy of object detection[EB/OL]. (2020-04-23). https:arxiv.org/abs/2004.10934.
[18]	HOWARD A G, ZHU Menglong, CHEN Bo, et al. MobileNets: efficient convolutional neural networks for mobile vision applications[EB/OL].(2017-04-17). https:arxiv.org/abs/1704.04861.
[19]	SANDLER M, HOWARD A, ZHU Menglong, et al. MobileNetV2: inverted residuals and linear bottlenecks[C]. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2018:4510-4520.
[20]	LIN T Y, DOLLAR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017:936-944.
[21]	PAN S J, YANG Qiang. A survey on transfer learning[J]. IEEE Transactions on Knowledge and Data Engineering, 2010, 22(10):1345-1359. doi: 10.1109/TKDE.2009.191
[22]	BYUNG J K. Improved deep learning algorithm[J]. Journal of Advanced Information Technology and Convergence, 2018, 8(2):119-127. doi: 10.14801/JAITC.2018.8.2.119
[23]	EVERINGHA M, GOOL L, WILLIAMS C K, et al. The pascal visual object classes (VOC) challenge[J]. International Journal of Computer Vision, 2010, 88(2):303-338. doi: 10.1007/s11263-009-0275-4

网络模型参数	参数配置值
检测数(Num_classes)	4
输入尺寸(Input_shape)	(608,608)
批尺寸(Batch_size)	16
初始学习率(Learning_rate_base)	0.001
网络模型参数		参数配置值
预热学习率(Warmup_learning_rate)		0.000 1
允许最小学习率(Min_learning_rate)		0.000 01
样本训练轮次(Epoch)		150

检测模型	AP/%	GPU检测速度/ (帧·s^-1)	CPU检测速度/ (帧·s^-1)
YOLOv4	92.55	30.98	1.34
MV2- YOLOv4	90.25	43.49	4.99

检测类别	P/%	R/%	AP/%	t_G/ms	t_C/ms
Safety-hat	98.37	97.47	97.43	15 204.56	121 961.42
No-safety-hat	97.52	92.19	92.01
Safety-belt	98.07	91.04	90.91
No-safety-belt	98.71	86.14	85.91
平均值			91.57	25.34	203.27

检测类别	P/%	R/%	AP/%	t_G/ms	t_C/ms
Safety-hat	97.23	97.64	97.19	15 292.24	123 197.25
No-safety-hat	96.88	91.18	90.59
Safety-belt	93.20	84.28	81.91
No-safety-belt	99.22	89.29	89.08
平均值			89.69	25.49	205.33

检测类别	P/%	R/%	AP/%	t_G/ms	t_C/ms
Safety-hat	88.66	88.44	85.84	15 956.52	128 288.33
No-safety-hat	95.97	98.35	95.92
Safety-belt	84.78	84.19	80.33
No-safety-belt	76.88	89.22	84.41
平均值			86.63	26.59	213.81