Research Of Safety Helmet Wearing Recognition Method Based On Video Surveillance

The construction industry is a labor-intensive,complex working environment,safety accident prone industry.Brain injury death caused by falling objects is a common typical accident in the construction industry.As an effective safety protection equipment,safety helmet can block the impact energy of falling objects and reduce head shock injury,which is widely used in the construction sites.Aiming at the common problems of false detection and missing detection in the existing safety helmet wearing recognition system,this thesis designs a safety helmet wearing recognition system based on video monitoring through in-depth study of image processing and target detection technology,which can be used to replace the traditional manual monitoring.When workers who do not wear safety helmet correctly enter the video monitoring area,they will give an alarm to prevent accidents,this thesis has important advanced significance and practical application value in protecting people’s life and property safety.The main work of this thesis is as follow:(1)Thesis completed the collection of video images,data set production and data enhancement.The video is processed into images,and the recognition targets are divided into five categories: person,red helmet,yellow helmet,white helmet,and blue helmet according to the research content.At the same time,use data augmentation to expand the data set to increase the amount of data for training models.(2)Thesis completed the method of identifying wearing helmets based on the YOLOv3 algorithm.This thesis has deeply studied the model principle of YOLOv3,trained the helmet wearing recognition model in the environment of Ubuntu16.04+Python 3.6+Py Torch1.4,and used the test set to test the model,recorded the experimental structure and planned to display some of the test results,calculate its detection parameters to evaluate the model.(3)Thesis completed the Center Net-based identification method for wearing helmets.Through in-depth research on the principle of the Center Net model,this thesis trains the helmet wearing recognition model in the environment of Ubuntu16.04+Python 3.6+Py Torch1.4,and tests the model on the same test set,and displays part of the test results.And compare with the detection results of YOLOv3 in the previous article.(4)Thesis completed the target detection method based on the improved Center Net model.Improve the calculation method of intersection and ratio to realize the local optimization of Io U.On this basis,train an improved Center Net target detection model,and test the model to compare with the detection results of YOLOv3 and Center Net models.(5)Thesis completed the helmet wearing recognition system based on video surveillance,and built the hardware and software platform of the helmet wearing recognition system.The research results show that the target image set in this subject can realize the recognition function of wearing a helmet.This thesis has researched and completed a safety helmet wearing recognition system based on video surveillance.Compared with traditional manual recognition,this helmet wearing recognition system makes up for the traditional manual recognition waste of manpower and material resources.It can not only perform monitoring,but also design alarms.Function,this thesis solves the problem of false detection and leakage detection of helmet wearing recognition by improving the IOU calculation method.The research results show that it can correctly identify target images in five categories: person,red helmet,yellow helmet,white helmet,and blue helmet.The test results show that the accuracy and recall rates are higher than92%,and there is no false detection and leakage detection.