Identification Of Construction Safety Risks Based On Event Logic Graph And Image Semantics

The safety issues of construction site have always been a concern for communities.Improving safety monitoring methods of construction site has become one of the critical steps to avoid safety accidents and property losses,and to guarantee the occupational health of construction workers.Given the low efficiency and high cost of traditional manual safety inspections,advanced surveillance systems supported by automation technology are considered to be the most potent alternative.However,taking computer vision technology as an example,the monitoring method has flaws as an inability to understand complex information or a lack of integrated framework.To fix the current technical shortcomings of computer vision technology in the field,this paper constructed a safety risk monitoring framework based on event logic graphs(ELG),integrating machine vision as the front-end input of the framework to realize automated identification and risk management on construction sites.First,using a large amount of construction safety accidents text data,this paper completed tasks including event representation,event extraction,and event relation extraction via deep learning methods and pattern recognition technology.Secondly,through event fusion and transition probability calculations,an ELG of construction site safety was constructed to establish a knowledge base showing the evolution mode of construction safety accidents.Next,the object detection algorithm based on YOLOv3 was involved to obtain the underlying semantic information from the image transformed into high-level image semantics through the construction scene ontology,implementing the image event description with the help of the event representation model.Then,the image event description text was used as the input of risk reasoning,mapped with the ELG.The event transition probability was used as the basis of risk reasoning to assess and identify safety risks.Finally,a case was used to verify the application capacity of the framework.This research was based on the knowledge expression form of ELG in the field of natural language processing,and the risk identification framework established with accident evolution filtering as the core of knowledge base has a certain generalization ability.Although this paper only used the object detection algorithm in machine vision as the technical support and explored the possibility of automated monitoring of construction site risks,any automated technology constructing event expression can be used as the input of the framework.It can be seen that the identification framework constructed in this paper provides a new knowledge-assisted early warning method to prevent construction accidents that can be used to improve the efficiency of construction safety management.