利用深度学习的施工人员安全隐患行为诊断控制方法

为了对建筑施工现场存在安全隐患的行为进行诊断控制,提出通过深度学习的方式对建筑施工现场工人的不安全行为进行识别;第一,要对人体骨骼运动模型进行提取,将提取得到的信息作为人体姿态以及运动发生变化的新模态信息,并针对以人体姿态为依据实现骨架信息提取这一过程进行简单介绍,再进一步提出CNN-LSTM模型,该模型能够对空间特征提取性能进行优化;过利用BN-Inception作为CNN-LSTM行为识别模型所需要的空间特征提取器,对所有视频帧中包含的空间结构信息进行提取过程的训练;再通过借助长短时记忆网络(LSTM)针对完整视频中的所有帧进行时序信息的建模,最终通过模型所得出的结果即为LSTM在最终时刻的预测输出;通过相关研究能够证明,利用CNN-LSTM模型获取的信息准确率能够达到88.67%,能够对单模态行为识别模型在识别过程中的准确率进行优化。

A Diagnosis and Control Method of Potential Safety Hazards for Construction personnel Using Deep Learning

In order to diagnose and control the behaviors with potential safety hazards in the construction site, it is proposed to identify the unsafe behaviors of workers in the construction site through in-depth learning. First, extract the human skeleton motion model, take the extracted information as the new modal information of human posture and movement changes, and briefly introduce the process of skeleton information extraction based on human posture, Furthermore, CNN-LSTM model is proposed, which can optimize the performance of spatial feature extraction. By using BN-Inception as the spatial feature extractor required by CNN-LSTM behavior recognition model, the spatial structure information contained in all video frames is trained in the extraction process. Then, the timing information of all frames in the complete video is modeled with the help of long-term and short-term memory network (LSTM). Finally, the result obtained by the model is the prediction output of LSTM at the final time. Through relevant research, it can be proved that the information accuracy obtained by CNN-LSTM model can reach 88.67%, and the accuracy of single-mode behavior recognition model in the recognition process can be optimized.