Research On Violation Detection Based On Machine Vision For Construction Site

As a pillar industry and an important part of China’s national economy,the construction industry has brought an important material foundation and great social wealth to the development of China.However,the complex environment of the construction site operation area,the uneven level of the operator’s operation specification and the high mobility of personnel have led to the occurrence of construction site safety accidents from time to time.As a common operation link in the construction site,the safety accidents caused by the irregularities of hoisting operators are common.At present,the monitoring of violations in the process of hoisting operation often relies on the monitoring personnel through the site monitoring video records,but with the increase of monitoring time and monitoring screen,there are problems such as leakage and time lag,so the monitoring behavior monitoring needs of hoisting operation site violations can no longer be satisfied through the monitoring personnel records.To this end,for the problem of detecting violations such as standing people under the crane arm,operators leaving their posts,and missing safety signs in the crane operation of construction sites,this paper proposes a machine vision-based method for detecting crane violations in construction site operation areas,which automatically detects the above violations in the crane process and provides effective support for the safety management of crane operations.The main work and innovation points of this paper are summarized as follows.(1)Taking the construction site lifting operation area as the application scenario,three types of common violations are studied: people standing under the lifting boom,lifting boom operators leaving their posts and missing safety signs.Firstly,the data set of various violations is established by shooting video data in the field and filtering after frame extraction.Secondly,for the problems of incomplete types,small quantity and uneven distribution of the collected missing safety signs dataset,a data enhancement method based on perspective transformation is proposed to simulate the camera taking pictures at different angles,while enriching the types and quantity of safety signs to provide data basis for the subsequent algorithm model.(2)To address the problem of difficult detection of small targets in the process of crane operation violation detection,a YOLOv5-Hoist network model based on CBAM(Convolutional Block Attention Module)is proposed.CBAM enhances the learning ability of the network for target features and reduces the influence of irrelevant features.Meanwhile,PRe LU(Parametric Rectified Linear Unit)is introduced to realize the update of negative region weight parameters and the network has stronger nonlinear fitting ability.The experiments show that the YOLOv5-Hoist network model improves the detection rate of small targets such as operators,lifting booms and safety signs while maintaining the detection performance basically unchanged,and improves the overall detection accuracy.(3)For the detection of three types of violations in the lifting operation,such as people standing under the lifting boom,lifting boom operators leaving the job,and missing safety signs,based on the target detection results of the YOLOv5-Hoist network model,the detection methods of people standing under the lifting boom based on scene modeling,the detection method of lifting boom operators leaving the job based on target tracking,and the detection method of missing safety signs based on regional determination are designed and implemented respectively.The detection method of missing safety signs based on area determination.The experiments show that the violation detection and determination method proposed in this paper can effectively identify the three types of typical violations in crane operation and has a good detection effect.