基于卷积神经网络算法的城市轨道交通施工人员不安全行为智能识别技术

郭飞孔恒乔国刚

Research on Intelligent Identification of Worker′s Unsafe Behavior in Urban Rail Transit Based on Convolutional Neural Network Algorithm

GUO FeiKONG HengQIAO Guogang
摘要:
[目的]施工人员的不安全行为是城市轨道交通施工事故发生的根本因素,传统的管理模式在约束人的不安全行为方面存在不足,因此需借助高精度定位技术和智能识别技术,从主观上消除事故隐患。[方法]介绍了城市轨道交通施工人员不安全行为的产生机理。结合UWB(超宽带无线通信)高精度定位技术、摄像机自标定技术及基于卷积神经网络算法的智能识别技术,搭建了具有定位、感知、识别、预警及通信功能的一体化智能管理平台。以安全帽识别为例,构建了安全帽识别拓扑流程图,对基于卷积神经网络算法的施工人员不安全行为识别的算法进行了测试。[结果及结论]测试结果表明,该算法可实现对施工现场未佩戴安全帽人员的识别,验证了该算法的准确性。该技术实现了对城市轨道交通施工人员不安全行为的智能识别预警。
Abstracts:
[Objective] Worker′s unsafe behavior is the fundamental factor in urban rail transit construction accidents. As the traditional management mode is insufficient in restraining the workers from the unsafe behavior, it is necessary to eliminate the hidden danger of accidents subjectively with the help of high precision positioning and intelligent identification technologies. [Method] The generation mechanism of worker′s unsafe behavior in urban rail transit is introduced. In combination with the technologies of UWB (ultra-wideband) high precision positioning, camera self-calibration and intelligent identification based on convolutional neural network algorithm, an integrated intelligent management platform with functions of positioning, perception, identification, early warning and communication is built. Taking helmet identification as an example, the topology flow chart of helmet identification is constructed, and the algorithm of worker′s unsafe behavior identification based on convolutional neural network is tested. [Result & Conclusion] The test results show that the algorithm can identify the person who does not wear safety helmet on construction site, verifying its accuracy. The technology realizes intelligent identification and early warning of worker′s unsafe behavior in urban rail transit construction.
论文检索