Based On Intelligent Video Analysis Research On The Perimeter Prevention System Of Construction Site

With the rapid development of the national construction industry,the safety supervision of the construction site has gradually increased.At present,the traditional video monitoring can only provide video display,transmission,capture,save and other functions,but the analysis of content needs to be implemented manually,which has the problems of high cost,heavy workload and high rate of missed detection.Therefore,the use of intelligent video analysis technology to improve the efficiency and accuracy of construction site safety supervision has become a topic of extensive research.In order to meet the needs of construction and safe production in construction sites,a perimeter protection system supported by intelligent video analysis technology is constructed,including two modules: perimeter intrusion detection and helmet wearing detection.In the warning range,the system can detect the intrusion target and the personnel who do not wear the helmet according to the regulations,and generate alarm information in time for the supervision personnel to deal with in time,and improve the safety production and work efficiency.Therefore,the main work and innovation of this thesis are as follows:(1)In view of the lack of data sets related to construction site scenes,30 video sequences from different angles were collected for the training and debugging of the moving target algorithm.(2)In view of the problem that the traditional intrusion detection method is easily affected by the weather,illumination change,camera jitter,large target scale change and other factors,which leads to a high model false detection rate,the moving target detector based on neural network is used for training on the video sequence,and the FMeasure index reaches 90.76%.At the same time,aiming at the problems of large number of parameters and long training time of the current classifier model,a simple and effective classifier was built.After training on the standard pedestrian data set Daimaler,the test accuracy was increased by 1.82% and the number of parameters was reduced to 1/2 of the original.(3)Aiming at the problem that the safety helmet sample data set is too small,which leads to the low accuracy of network detection,the sample expansion algorithm based on scene enhancement is proposed,and the safety helmet wearing detection is carried out by combining with the YOLO v4 network model.Experiments show that the detection accuracy of the helmet detection model on the Helmetwear dataset is improved by 6.39%,which provides an important reference for solving the problem of insufficient target detection training data in other fields.(4)To solve the problem that the Fg Seg Net＿v2 network can perform well only when the test set and the training set belong to the same scene,a cyclic iterative debugging method is proposed,that is,the segmentation network is used to assist the detection of Fg Seg Net＿v2 in new scenes,so as to improve the adaptability of the system to new scenes.