Research And Implementation Of Helmet Wearing Recognition System Based On Deep Learning

In the 21 st century,artificial intelligence products have gradually been applied to various fields of people’s lives.It is common for employees who do not wear safety helmets to die due to falling objects at construction sites.Therefore,designing and developing a safety helmet wearing recognition system based on deep learning technology for scenarios such as construction site entrances and exits is beneficial for helping relevant personnel develop the habit of wearing safety helmets and protect their own safety.However,existing helmet wearing recognition algorithms cannot perform live detection on individuals in video streams.When individuals without helmets use photos or videos of wearing helmets to forge identification targets,they can still enter the construction site through access control;Secondly,the current helmet wearing recognition algorithm has certain shortcomings in recognition speed and accuracy,and does not have the ability to recognize helmet wearing in complex environments.In response to the above issues,this article conducts in-depth research on helmet wearing recognition technology based on deep learning.The main research content of this article is as follows:(1)A helmet wearing recognition algorithm has been studied and designed.This algorithm mainly consists of three modules,namely live body detection-blink detection,head posture estimation,and helmet wearing recognition.Firstly,the blink detection module is based on the Eye Aspect Ratio Support Vector Machines method(EAR SVM)to detect the blinking of individuals to be detected in the video stream.Secondly,in response to the issue of using videos wearing safety helmets for camouflage,the head pose estimation module has designed and adopted a convolutional neural network with a four layer network structure to estimate and recognize the head pose of the subject to be detected.In addition,the use of Leaky Re LU activation function can effectively prevent the Re LU function from having a neuron gradient of 0 in the negative value interval when training with gradient descent method.Finally,in view of the insufficient recognition speed and accuracy of existing helmet wearing recognition algorithms under complex background,this paper improves the network structure of YOLOv5 s,introduces SE attention mechanism into its backbone network,enhances the ability of YOLOv5 s network to aggregate global context information,and improves the accuracy of helmet wearing recognition;At the same time,replace the CIo U in the original YOLOv5 s with alpha-Io U to improve the speed and accuracy of the deep neural network boundary box regression.(2)In response to practical application problems,a deep learning based helmet wearing recognition system was designed and implemented.Using the Client/Server architecture,the server determines whether to enable access control by detecting whether the detected person is wearing a helmet in the real-time video stream collected by the client.The system client consists of a nanopi m4 main control module,USB camera,and other hardware components,and the server is a personal computer.Develop a registration and login module,a live body detection blink detection module,a head posture estimation module,and a helmet wearing recognition module using Python and deep learning technology,in order to achieve the expected goal of individual helmet wearing recognition.(3)Conduct experiments on the safety helmet wearing recognition system based on deep learning technology designed and implemented in this article in practical application scenarios to test its average accuracy.Through experimental results,it can be found that the accuracy of its system basically meets the requirements of practical applications in specific environments,thus proving the effectiveness of the method and corresponding system described in this article at both theoretical and practical levels.