铁路工人人体行为识别模型  

Human activity recognition model of railway workers

作　　者：黄珍珍 肖硕[1] 王钰[3] 陈伟 王升志[1] 江海峰 HUANG Zhenzhen;XIAO Shuo;WANG Yu;CHEN Wei;WANG Shengzhi;JIANG Haifeng(School of Computer Science and Technology,China University of Mining and Technology,Xuzhou Jiangsu 221116,China;Library,China University of Mining and Technology,Xuzhou Jiangsu 221116,China;Casco Signal Company Limited,Beijing 100070,China)

机构地区：[1]中国矿业大学计算机科学与技术学院,江苏徐州221116 [2]中国矿业大学图书馆,江苏徐州221116 [3]卡斯柯信号有限公司,北京100070

出　　处：《中国安全科学学报》2022年第6期17-22,共6页China Safety Science Journal

基　　金：国家自然科学基金资助(62071470,61971421,51874300);徐州市科技计划项目(KC20167)。

摘　　要：为提高铁路工人施工安全系数,采用基于人体行为识别(HAR)的智能化监测方法,估计铁路工人在施工过程中的动作;使用端到端自动提取数据特征的深度学习方法搭建网络,提高行为识别精度和模型泛化性;鉴于循环神经网络并行能力差,收敛时间长,提出结合空洞卷积与自注意力机制的深度学习模型;使用WISDM和MobiAct公开数据集,分别识别2个数据集上的基本动作和跌倒、撞击等行为。结果表明:相比于卷积神经网络(CNN)、长短期记忆(LSTM)网络、深度卷积LSTM网络,该模型具有更好的识别精度和性能,能够实现更准确的工人行为划分。In order to improve the construction safety factor of railway workers,the intelligent monitoring method based on HAR was used to estimate the action of railway workers in the construction process.The deep learning method of end-to-end automatic extraction of data features is applied to build a network to improve the accuracy of behavior recognition and model generalization.In view of the poor parallel ability and long convergence time of the recurrent neural network(CNN),a deep learning model combining cavity convolution and self-attention mechanism is proposed.The WISDM and MobiAct public datasets are used to identify the basic actions and fall and impact behaviors on the two datasets.The results show that compared with convolutional neural network(CNN),long-term and short-term memory(LSTM)network and deep convolutional LSTM neural network,the model has better recognition accuracy and performance,and can realize more accurate division of worker behavior.

关 键 词：铁路工人 人体行为识别(HAR) 深度学习 空洞卷积 自注意力机制 

分 类 号：X910[环境科学与工程—安全科学]