| Construction industry is one of the pillar industries of our national economy.While maintaining high-speed development,it is always a labor-intensive mode of production,with low labor productivity.The construction stage is the most active stage of resource flow and project management behavior.The use of information technology means to automatically collect and extract construction site data,providing data support for various fields of site information management,which is conducive to promoting the transformation and upgrading of the industry to quality and intensive.The existing relevant research conclusions have a narrow range of applicability,without good portability and promotion value.The reason is that most of the selected methods have limitations in the scope of application and cost.Based on the above problems,this study proposes a machine-vision based method of automatic data acquisition and processing.This method takes the field monitoring system as raw data input,which has been widely used,uses the object detection model in the field of machine vision to process the data,and obtains the type,quantity,and location of the field object(such as workers,materials and machines).After that,we use the human object interaction model to classify the interaction between workers and materials and equipment,and get the information about working activities on site.In addition,the 3D pose estimation algorithm can also be used to realize the workers’ pose based on surveillance video,and to assist the interaction classification progress.Based on the above methods,a conceptual model of data automatic processing and information extraction system based on site image is proposed and its structure is described.The system integrates target detection,interactive recognition and pose estimation methods,and contains five functional modules of field management from the perspective of its requirements,which provides a framework for automatic field management.Thanks to the modular design of the system and the maturity of the migration learning theory,the system can be easily expanded to various subdivisions of the field management and can keep up with the development trend of the field of information technology.In this study,the feasibility of the selected method and model is trained and verified by collecting the original data,making the annotation rules,and building the data set,and the experimental as well as application results are analyzed and discussed.The experimental results show that the above methods can basically meet the requirements of field data extraction at this stage.With the continuous development of data sets and computer hardware in the future,the system has considerable potential for performance improvement.This study proposes and validates three methods of data acquisition and processing on construction site,and integrates these methods to design and analyze the architecture of automatic data acquisition and processing system on construction site,which provides the possibility of collecting and analyzing on-site data on a low-cost,large-scale and uninterrupted basis.The system can serve multiple fields of construction site management,make up for the existing methods of construction site management,and has certain theoretical and practical value.These methods play the role of information technology in the field of field management,and can improve the level of field management information. |
