基于深度学习的建筑工人安全帽佩戴识别研究
|
摘要
建筑工人头部伤害是造成建筑伤亡事故的重要原因。佩戴安全帽是防止建筑工人发生脑部外伤事故的有效措施,而在实际工作中工人未佩戴安全帽的不安全行为时有发生。因此,对施工现场建筑工人佩戴安全帽自动实时检测进行探究,将为深入认知和主动预防安全事故提供新的视角。然而,传统的施工现场具有安全管理水平低下、管理范围小、主要依靠安全管理人员的主观监测并且时效性差、不能全程监控等一系列问题。针对上述现状,提出了一种基于Tensorflow框架,具有高精度、快速等特性的Faster RCNN方法,实时监测工人安全帽佩戴状况。为评估模型性能,收集了6 000张图像用于模型的训练与测试,结果表明,该模型识别工人安全监测中佩戴安全帽工人的平均精度达到90. 91%,召回率达到89. 19%;识别未佩戴安全帽工人的精度达到88. 32%,召回率达到85. 08%。同时,针对工人未佩戴安全帽而进入施工现场的违规行为,通过施工现场入口处监控摄像头截取视频流图像帧,设置检验试验,验证了本方法在施工现场实际应用的有效性。
This paper intends to propose an approach to the Faster RCNN with satisfactory accuracy,high speed with no need to add up any more auxiliary devices based on the Tensor Flow Framework to identify and check the safety helmet real time wearing manners for the construction workers. For the said purpose,we have collected over 6 000 photo-images of the construction workers in wearing their helmets and developed a diversified dataset of the photo-images by visiting different construction sites by ways of taking photographs and extracting frames from the surveillance videos. We have also pre-processed and filtered out the noise and scales of the photo-images simultaneously so as to get rid of the influence of noise making on the image detection and dissemination. What is more,the photo-images we have gained have also been turned to train and test the model by classifying them into two categories: the helmets the construction workers are wearing( safe) and those with no helmets( unsafe),which performance has also been registered and accounted based on the average precision and recalling. Thus,the statistical results indicate that the said rate for the workers' wearing safety helmets can be as precise as about 90. 91% and 89. 19%,correspondingly and respectively. The statistical results also show that the average precision and recalling rate identifying the workers' notwearing safety helmets has been done as precise as up to88. 32% and 85. 08%,respectively. Besides,in view of the undue irregularities of the workers' not wearing such helmets at construction sites,we have also extracted a few video streams from the surveillance camera at the entrance leading to the construction sites. Thus,the test experiment results demonstrate that the general precision and recalling rate for identifying the building workers' not wearing safety helmets prove to be as precise as up to 91. 44% and 89. 69%. In addition,the testing experiment and statistical results can also be done to verify the validity of the above mentioned method in the similar practical situations. Therefore,it can be concluded that,the application of the Faster RCNN method based on the deep learning to identify the workers' safety helmets can surely help to give out quick and prompt warning effect in changing such unsafe behaviors of the building workers at the construction sites and offer opportunities for necessary change so as to promote real-time site monitoring and heighten the safety management of the working order of the construction sites.
This paper intends to propose an approach to the Faster RCNN with satisfactory accuracy,high speed with no need to add up any more auxiliary devices based on the Tensor Flow Framework to identify and check the safety helmet real time wearing manners for the construction workers. For the said purpose,we have collected over 6 000 photo-images of the construction workers in wearing their helmets and developed a diversified dataset of the photo-images by visiting different construction sites by ways of taking photographs and extracting frames from the surveillance videos. We have also pre-processed and filtered out the noise and scales of the photo-images simultaneously so as to get rid of the influence of noise making on the image detection and dissemination. What is more,the photo-images we have gained have also been turned to train and test the model by classifying them into two categories: the helmets the construction workers are wearing( safe) and those with no helmets( unsafe),which performance has also been registered and accounted based on the average precision and recalling. Thus,the statistical results indicate that the said rate for the workers' wearing safety helmets can be as precise as about 90. 91% and 89. 19%,correspondingly and respectively. The statistical results also show that the average precision and recalling rate identifying the workers' notwearing safety helmets has been done as precise as up to88. 32% and 85. 08%,respectively. Besides,in view of the undue irregularities of the workers' not wearing such helmets at construction sites,we have also extracted a few video streams from the surveillance camera at the entrance leading to the construction sites. Thus,the test experiment results demonstrate that the general precision and recalling rate for identifying the building workers' not wearing safety helmets prove to be as precise as up to 91. 44% and 89. 69%. In addition,the testing experiment and statistical results can also be done to verify the validity of the above mentioned method in the similar practical situations. Therefore,it can be concluded that,the application of the Faster RCNN method based on the deep learning to identify the workers' safety helmets can surely help to give out quick and prompt warning effect in changing such unsafe behaviors of the building workers at the construction sites and offer opportunities for necessary change so as to promote real-time site monitoring and heighten the safety management of the working order of the construction sites.
引文
[1] Ministry of Housing and Urban-Rural Development of the People's Republic of China(中华人民共和国住房和城乡建设部). Notification of safety accidents(安全事故通报)[EB/OL].[2018-03-08]. http://www. mohurd.gov. cn/wjfb/201803/t20180322_235474. html.
[2] KONDA S,TIESMAN H,REICHARD A. Fatal traumatic brain injuries in the construction industry,2003-2010[J]. American Journal of Industrial Medicine,2016,59(3):212-220.
[3] DONG X W S,CHOI S D,BORCHARDT J G,et al. Fatal falls from roofs among U. S. construction workers[J].Journal of Safety Research,2013,44(1):17-24.
[4] PAN Dan(潘丹),LUO Fan(罗帆). Safety performance evaluation of the construction projects based on structural equation model[J]. Journal of Safety and Environment(安全与环境学报),2018,18(2):602-609.
[5] Health and Safety Executive. Statistics on fatal injuries in the workplace in Great Britain 2016[EB/OL].[2016-04-19]. http://www. hse. gov. uk/Statistics/oshman. html.
[6] ZHOU Zhenglong(周正龙),MA Benjiang(马本江),HU Fengying(胡凤英). On the management of the construction project based on the evolutionary stable strategy[J]. Journal of Safety and Environment(安全与环境学报),2016,16(1):168-171.
[7] GUO B H W,GOH Y M,WONG K L X. A system dynamics view of a behavior-based safety program in the construction industry[J]. Safety Science,2018,104:202-215.
[8] HINTON G E,OSINDERO S,THE Y W. A fast learning algorithm for deep belief nets[J]. Neural Computation,2006,18(7):1527-1554.
[9] SEO J,HAN S,LEE S,et al. Computer vision techniques for construction safety and health monitoring[J].Advanced Engineering Informatics, 2015, 29(2):239-251.
[10] HAN S,LEE S,PENA-MORA F. Vision-based detection of unsafe actions of a construction worker:case study of ladder climbing[J]. Journal of Computing in Civil Engineering,2013,27(6):635-644.
[11] YU Y T,GUO H L,DING Q H,et al. An experimental study of real-time identification of construction workers'unsafe behaviors[J]. Automation in Construction,2017,82:193-206.
[12] DING L Y,FANG W L,LOU H B,et al. A deep hybrid learning model to detect unsafe behavior:Integrating convolution neural networks and long short-term memory[J]. Automation in Construction,2018,86:118-124.
[13] HU Tian(胡恬),WANG Xingang(王新刚). Analysis and design of safety helmet recognition system based on wavelet transform and neural networks[J]. Software Guide(软件导刊),2006(23):37-39.
[14] FANG Qi,LI Heng,LUO Xiaochun,et al. Detecting non-hardhat-use by a deep learning method from far-field surveillance videos[J]. Automation in Construction,2018,85:1-9.
[15] DALAL N,TRIGGS B. Histograms of oriented gradients for human detection[C]//Computer vision and pattern recognition:Proceedings of 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition.New York,NY,USA:Institute of Electrical and Electronics Engineers,2005:886-893.
[16] PARK M W,ELSAFTY N,ZHU Z H. Hardhat-wearing detection for enhancing on-site safety of construction workers[J]. Journal of Construction Engineering and Management,2015,141(9):1-16.
[17] MNEYMNEH B E,ABBAS M,KHOURY H. Automated hardhat detection for construction safety applications[C]//Proceedings of the 6th Creative Construction Conference(CCC). Primosten, Croatia:Elsevier Ltd.,2017:895-902.
[18] REN S Q,HE K M,GIRSHICK R,et al. Faster RCNN:towards real-time object detection with region proposal networks[J]. IEEE Transactions on Patter Analysis and Machine Intelligence,2017,39(6):1137-1149.
[19] CHENG Jiahui(程嘉晖). Aircraft image recognition based on deep convolution neural network(基于深度卷积神经网络的飞行器图像识别算法研究)[D]. Hangzhou:Zhejiang University,2017.
[2] KONDA S,TIESMAN H,REICHARD A. Fatal traumatic brain injuries in the construction industry,2003-2010[J]. American Journal of Industrial Medicine,2016,59(3):212-220.
[3] DONG X W S,CHOI S D,BORCHARDT J G,et al. Fatal falls from roofs among U. S. construction workers[J].Journal of Safety Research,2013,44(1):17-24.
[4] PAN Dan(潘丹),LUO Fan(罗帆). Safety performance evaluation of the construction projects based on structural equation model[J]. Journal of Safety and Environment(安全与环境学报),2018,18(2):602-609.
[5] Health and Safety Executive. Statistics on fatal injuries in the workplace in Great Britain 2016[EB/OL].[2016-04-19]. http://www. hse. gov. uk/Statistics/oshman. html.
[6] ZHOU Zhenglong(周正龙),MA Benjiang(马本江),HU Fengying(胡凤英). On the management of the construction project based on the evolutionary stable strategy[J]. Journal of Safety and Environment(安全与环境学报),2016,16(1):168-171.
[7] GUO B H W,GOH Y M,WONG K L X. A system dynamics view of a behavior-based safety program in the construction industry[J]. Safety Science,2018,104:202-215.
[8] HINTON G E,OSINDERO S,THE Y W. A fast learning algorithm for deep belief nets[J]. Neural Computation,2006,18(7):1527-1554.
[9] SEO J,HAN S,LEE S,et al. Computer vision techniques for construction safety and health monitoring[J].Advanced Engineering Informatics, 2015, 29(2):239-251.
[10] HAN S,LEE S,PENA-MORA F. Vision-based detection of unsafe actions of a construction worker:case study of ladder climbing[J]. Journal of Computing in Civil Engineering,2013,27(6):635-644.
[11] YU Y T,GUO H L,DING Q H,et al. An experimental study of real-time identification of construction workers'unsafe behaviors[J]. Automation in Construction,2017,82:193-206.
[12] DING L Y,FANG W L,LOU H B,et al. A deep hybrid learning model to detect unsafe behavior:Integrating convolution neural networks and long short-term memory[J]. Automation in Construction,2018,86:118-124.
[13] HU Tian(胡恬),WANG Xingang(王新刚). Analysis and design of safety helmet recognition system based on wavelet transform and neural networks[J]. Software Guide(软件导刊),2006(23):37-39.
[14] FANG Qi,LI Heng,LUO Xiaochun,et al. Detecting non-hardhat-use by a deep learning method from far-field surveillance videos[J]. Automation in Construction,2018,85:1-9.
[15] DALAL N,TRIGGS B. Histograms of oriented gradients for human detection[C]//Computer vision and pattern recognition:Proceedings of 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition.New York,NY,USA:Institute of Electrical and Electronics Engineers,2005:886-893.
[16] PARK M W,ELSAFTY N,ZHU Z H. Hardhat-wearing detection for enhancing on-site safety of construction workers[J]. Journal of Construction Engineering and Management,2015,141(9):1-16.
[17] MNEYMNEH B E,ABBAS M,KHOURY H. Automated hardhat detection for construction safety applications[C]//Proceedings of the 6th Creative Construction Conference(CCC). Primosten, Croatia:Elsevier Ltd.,2017:895-902.
[18] REN S Q,HE K M,GIRSHICK R,et al. Faster RCNN:towards real-time object detection with region proposal networks[J]. IEEE Transactions on Patter Analysis and Machine Intelligence,2017,39(6):1137-1149.
[19] CHENG Jiahui(程嘉晖). Aircraft image recognition based on deep convolution neural network(基于深度卷积神经网络的飞行器图像识别算法研究)[D]. Hangzhou:Zhejiang University,2017.