Safety helmet wearing detection algorithm in complex construction environment based on improved YOLOv3

doi:10.16265/j.cnki.issn1003-3033.2022.05.0714

Abstract

Abstract:

In order to address problems of inaccurate or failed detection of safety helmet wearing under intelligent monitoring due to complex environment in construction sites, an improved YOLOv3 detection algorithm was proposed. Focal Loss was adopted to train difficult positive samples so as to improve the model's robustness in complex environment. Then, its multi-scale detection capabilities were improved by using spatial pyramid multi-level pooling based on initial network. Thirdly, attention mechanism was introduced, and channel and spatial attention modules were respectively integrated into YOLOv3's backbone and residual structure of detection layer network, so that it would focus on helmet feature learning. Finally, GIoU was utilized to improve positioning accuracy, and the algorithm's effectiveness was verified under different visual conditions in a complex construction environment. The results show that the improved model's mean accuracy reaches 88%, 13.3% higher than the original one, among which the precision of person and helmet are increased by 17.2% and 9.5%, while recall rate is increased by 15.3% and 7.6%.

Key words: you only look once(YOLOv3), complex construction environment, helmet wearing, detection algorithm, Focal Loss, spatial pyramid pooling(SPP), attention mechanism, generalized intersection over union (GIoU)

ZHAO Hongcheng, TIAN Xiuxia, YANG Zesen, BAI Wanrong. Safety helmet wearing detection algorithm in complex construction environment based on improved YOLOv3[J]. China Safety Science Journal, 2022, 32(5): 194-200.

Figures/Tables 12

Fig.1

Fig.2

Fig.3

Fig.4

Fig.5

Fig.6

Fig.7

Tab.1

Fig.8

Fig.9

Fig.10

Fig.11

References 20

[1]	张伟, 朱双娜, 张潇, 等. 建筑施工安全事故致因系统模型与实证分析[J]. 中国安全科学学报, 2019, 29(6):56-62. doi: 10.16265/j.cnki.issn1003-3033.2019.06.010
	ZHANG Wei, ZHU Shuangna, ZHANG Xiao, et al. Systematic model of construction accident causation andits empirical analysis[J]. China Safety Science Journal, 2019, 29(6): 56-62. doi: 10.16265/j.cnki.issn1003-3033.2019.06.010
[2]	WEN Chenyen. The safety helmet detection technology and its application to the surveillance system[J]. Journal of Forensic Sciences, 2004, 49(4): 770-780. pmid: 15317192
[3]	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, 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, 2014, 40(3): 365-370.
[4]	GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2014: 580-587.
[5]	GIRSHICK R. Fast R-CNN[C]. Proceedings of the IEEE International Conference on Computer Vision, 2015: 1440-1448.
[6]	REN Shaoqing, HE Kaiming, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[C]. Advances in Neural Information Processing Systems, 2015: 91-99.
[7]	徐守坤, 王雅如, 顾玉宛, 等. 基于改进区域卷积神经网络的安全帽佩戴检测[J]. 计算机工程与设计, 2020, 41(5): 1385-1389.
	XU Shoukun, WANG Yaru, GU Yuwan, et al. Construction site safety helmet wearing detection based on improved region convolutional neural network[J]. Computer Engineering and Design, 2020, 41(5): 1385-1389.
[8]	CHU Jun, GUO Zhixian, LENG Lu. Object detection based on multi-layer convolution feature fusion and online hard example mining[C]. IEEE Access, 2018: 19 959-19 967.
[9]	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]. IEEE Conference on Computer Vision and Pattern Recognition, 2016: 779-788.
[10]	REDMON J, FARHADI A. YOLO9000: better, faster, stronger[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 7263-7271.
[11]	REDMON J, FARHADI A. YOLOv3: an incremental improvement[C]. Computer Vision and Pattern Recognition, IEEE, 2018: 3523-3541.
[12]	LIU Wei, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector[C]. European Conference on Computer Visio, 2016: 21-37.
[13]	LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017: 2999-3007.
[14]	施辉, 陈先桥, 杨英. 改进YOLOv3的安全帽佩戴检测方法[J]. 计算机工程与应用, 2019, 55(11): 213-220.
	SHI Hui, CHEN Xianqiao, YANG Ying. Safety helmet wearing detection method of improved YOLOv3[J]. Computer Engineering and Applications, 2019, 55(11): 213-220.
[15]	REZATOFIGHI H, TSOI N, GWAK J Y, et al. Generalized intersection over union: a metric and a loss for bounding box regression[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2019: 658-666.
[16]	赵欣然, 张琪, 王卫东, 等. 可燃性粉尘云的图像检测方法[J]. 中国安全科学学报, 2020, 30(4): 8-13. doi: 10.16265/j.cnki.issn1003-3033.2020.04.002
	ZHAO Xinran, ZHANG Qi, WANG Weidong, et al. Image detection method of combustible dust cloud[J]. China Safety Science Journal, 2020, 30(4): 8-13. doi: 10.16265/j.cnki.issn1003-3033.2020.04.002
[17]	李云, 张晴, 沈子豪, 等. FMF的火焰显著性检测[J]. 中国安全科学学报, 2019, 29(5): 56-61. doi: 10.16265/j.cnki.issn1003-3033.2019.05.010
	LI Yun, ZHANG Qing, SHEN Zihao, et al. Flame saliency detection based on FMF[J]. China Safety Science Journal, 2019, 29(5): 56-61. doi: 10.16265/j.cnki.issn1003-3033.2019.05.010
[18]	HE Kaiming, ZHANG Xiangyu, REN Shaoqing, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2014, 37(9): 1904-1916.
[19]	W OO S, PARK J, LEE J Y, et al. CBAM:convolutional block attention module[C]. Proceedings of the European Conference on Computer Vision(ECCV), 2018: 3-19.
[20]	KINGMA D, BA J. Adam: a method for stochastic optimization[C]. Computer Science, 2014: 1-15.

模型类型	person(精度)	helmet(精度)	person(召回率)	helmet(召回率)	平均精度
YOLOv3	0.707	0.787	0.724	0.818	0.747
YOLOv3-SPP	0.748	0.86	0.786	0.887	0.804
YOLOv3-SE	0.733	0.851	0.767	0.863	0.792
YOLO-A	0.877	0.871	0.881	0.89	0.874
YOLO-B	0.863	0.86	0.84	0.844	0.861
YOLO-C	0.86	0.877	0.871	0.87	0.868
本文模型	0.879	0.882	0.877	0.894	0.88