基于图像的地铁车站施工人员不安全行为识别研究

为了解决人工监测不能实时看护和及时管理地铁车站施工人员的不安全行为等问题，构建了基于KNN,MLP和LSTM模型的不安全行为识别神经网络模型。首先，通过行为理论研究和现场调查分析，对地铁车站施工不安全行为进行了分类；其次，通过实验构建人体数据集，基于人体骨骼关节点提取不安全行为特征，并进行模型训练；最后，基于MobileNet V1的SSD目标检测算法对施工人员进行定位和追踪，结合射线法判断目标是否跨越不安全区域并发出警报，搭建神经网络模型对施工人员的不安全行为进行识别，并获得计算识别率。结果表明：传统机器学习算法KNN总体准确率为93.45%,优化后的MLP和LSTM两种神经网络模型总体准确率分别达到93.94%和93.68%,相对KNN算法分别提高了0.49%和0.23%。因此所提模型能有效识别施工人员不安全行为，可为地铁施工安全智能识别技术应用提供参考。

Research on identification of unsafe behaviors of construction personnel in subway station based on images

In order to solve the problem that manual monitoring can not monitor and manage the unsafe behavior of subway station construction personnel in real time, a neural network model of unsafe behavior recognition based on KNN, MLP and LSTM models was constructed. Firstly, the unsafe behaviors in subway station construction were classified through behavioral theory research and field investigation analysis. Secondly, the human data set was constructed through experiments, and the unsafe behavior characteristics were extracted based on the human bone junctions, and the model training was conducted. Finally, the SSD target detection algorithm based on MobileNet V1 located and tracked the construction personnel, and judged whether the target crosses the unsafe area and gives an alarm by combining the ray method. The neural network model was built to identify the unsafe behavior of the construction personnel, and the calculation recognition rate was obtained. The results show that the overall accuracy of the traditional machine learning algorithm KNN is 9345%, and the overall accuracy of the optimized MLP and LSTM neural network models is 9394% and 9368%, respectively, which is 049% and 023% higher than that of the KNN algorithm. The proposed model can effectively identify unsafe behaviors of construction personnel, and can provide reference for the application of intelligent identification technology for subway construction safety.