Research On Motion Segmentation And Behavior Recognition Technology And Its Application In Smart Construction Site

Large construction equipment has always been the main driver of construction progress in various construction activities,and is also one of the most energy consuming links in the construction process.Nowadays,deep learning technology has been applied to some repetitive and complex tasks in the construction field,such as facial recognition in construction site access control and helmet recognition in construction site safety monitoring.However,currently the efficiency monitoring of large-scale construction equipment still relies on manual recording and analysis,which is time-consuming and not accurate enough.In order to solve the problem that the efficiency monitoring of large construction equipment is not mature,this paper proposes to apply the behavior recognition technology to the efficiency detection of large construction equipment,by using ASFormer,MS-TCN and other behavior segmentation network to identify the activities of large construction equipment in the construction site.The work efficiency of large construction equipment can be monitored and the construction schedule and scheme can be adjusted by analyzing the action identification results of its activities.(1)In the research process,the excavator working video is collected from the construction site and the 3D convolutional network is used to extract the video time sequence features from the adjusted video data,and the data set of excavator time sequence action segmentation is produced.A formula method for calculating the working efficiency of excavator based on the results of action segmentation is proposed.(2)To complete the improvement of MS-TCN network model,add the Stage Cascade module and Dual Dilation Layer to the original model,and improve the frame accuracy by increasing the network depth,computing power and receptive field during convolution operation.The Local Boundary Pooling module is added to find the action boundary through the local boundary pooling module,which makes the network model output smoother time sequence action segmentation results.(3)Combining the field of time sequence action segmentation with the field of construction,the work efficiency of excavator is calculated through the result of time sequence action segmentation,and the reasonable work arrangement is analyzed.When the construction site monitoring equipment is more,the speed of machine vision processing video data is faster than the manual speed,which can greatly reduce the labor cost of construction site management,improve the efficiency of equipment use,reduce safety risks,and provide a reliable technical way for the smart site digitization.