Unsafe Behavior Recognition Of Construction Workers Working On The Edge Of Buildings Based On Computer Vision

The construction industry plays an important role in China’s national economy,has a vital impact on the development of the national economy,and is an important pillar industry in China.Therefore,it is of great significance to ensure the steady development of the construction industry.At present,the frequency of safety accidents is still high due to the complexity of equipment,more personnel and poor environment in the construction site.High-frequency safety incidents can cause negative social stress in addition to significant loss of life and property.According to the safety accident statistics,most accidents are caused by people’s unsafe behavior,so effective prevention and control of unsafe behavior of construction workers plays a key role in improving the level of safety management.According to the report of safety accidents in China,falling accidents accounted for the highest proportion of safety occurrences,according to safety occurrences reports.Therefore,the computer vision method combined with surveillance video is used to identify and monitor the unsafe behaviors of construction workers in the common adjacent environment in the construction site,so as to improve the efficiency of safety management.Firstly,according to the characteristics of spatial and temporal information in the dynamic process of human behavior,a behavior recognition model structure including CNN and LSTM was established.By introducing the lightweight ECA channel attention module into the Deeper network architecture of Res Next,the performance of the CNN is enhanced with very little additional parameter addition.At the same time,BI-LSTM,which has a stronger ability to process time series,is used to extract the time features in the behavior process.The type of unsafe behaviors near the construction workers was determined through field investigation,and the unsafe behaviors near the construction workers were simulated in the laboratory environment,and the data set was established.The algorithm proposed in this paper was verified on the public UCF-101 dataset and the self-established worker adjacent behavior dataset,and the recognition rate reached 86% and 76%,respectively,with good detection effect.Construction workers limb homework judgment of unsafe behavior has situational awareness,limb work scenarios,the author of this paper the characteristics of the target improvement of anchor box size and improved by combining Bi-FPN Mask R-CNN algorithm,and realized with limb environment is used to identify the target of the workers and over the border fence and pixel level division,in the edge detection on state data set in the self-built accurate rate was 91.3%,compared with the original Mask R-CNN algorithm is improved by 3.1%.Finally,based on the pixel set of segmented workers and adjacent fence,a method to determine the adjacent state of construction workers is proposed.The discriminative approach of unsafe worker behavior suggested in this paper has a good result on the discriminative of unsafe worker behavior near the edge.It can automatically monitor the typical unsafe behaviors of workers in the adjacent operation,which can be used to reduce the occurrence of unsafe behavior accidents in the adjacent environment and enhance the safety management effect.