Research On Intrusion Detection Of Construction Dangerous Area Based On Binocular Stereo Vision Technology

With the rapid development of construction industry in our country,the scale of civil construction,bridge,tunnels and other projects have continued to expand,engineering construction safety accidents occurred frequently,which will not only have a great impact on the progress of the project and the interests of enterprises,but also threaten the personal safety of construction personnel.An important factor in the accident is that the construction and supervision personnel invaded the dangerous areas of construction machinery without meeting safety requirements.This paper will explore the application of binocular stereoscopic vision technology in construction site safety management,realize man-machine distance detection with binocular stereoscopic vision technology,and introduce convolutional neural network to realize man-machine identification on construction site,propose a classification management system for dangerous areas applicable to binocular stereoscopic vision technology,and solve the problem that safety personnel on construction site can’t be on duty in the whole area all day long.To timely warn personnel close to the source of danger,to avoid the occurrence of safety accidents,the main research content is as follows:(1)In response of the problem of binocular camera recognition and positioning,the basic principle of relevant target recognition and positioning are compiled,as well as the existing problems in engineering applications and solutions;(2)In the open construction site,when the existing YOLOV3 algorithm is used to identify human-machine,the pixels occupied by the personnel are overlapped by multiple objects to be ignored,resulting in a distance calculation and early warning.By giving full play to the residual block structure of YOLOv3 network structure,combined with self-built training set and testing feature enhancement model,YOLOv3 can improve its recognition ability to small targets.The results show that the accuracy and loss error of the improved algorithm are better than those of the improved algorithm,which improves the applicability of the algorithm in the open area of the construction site.(3)The traditional monocular industrial camera on the construction site is transformed and upgraded into a relatively independent binocular camera set,so as to realize the efficient measurement of man-machine distance on the construction site.The camera calibration module in MATLAB was recompiled with Python language to realize camera calibration,stereo matching and three-dimensional reconstruction in Python environment.According to the calibration parameters of the camera,the image is processed to eliminate the distortion,and the binocular stereo vision distance measurement model is completed according to the imaging principle.The effectiveness and real-time performance of the model parameters and target ranging are further verified.(4)Taking the construction machinery excavator commonly used in the construction site as an example,the dynamic classification rules of dangerous areas are established,and different dangerous areas are established according to different excavator models.The man-machine distance measured by the binocular camera is compared,so as to judge the safety state of construction personnel.