Research On Improved YOLOv5 Model For Helmet Wearing Detection

The complexity of infrastructure scenarios has led to a sustained increase in the global number of worksites related fatalities and injuries.The head of the human body is the most important and fragile part of the body.Therefore,in the infrastructure scene with a large risk factor,correctly wearing a helmet can effectively avoid the occurrence of most safety accidents.However,due to the complex construction environments,it is challenging to precisely detect safety helmet wearing in real-time.Most of the current helmet detection algorithms have problems such as low detection accuracy,poor complex scene adaptability,and hard to achieve real-time detection.For the above difficulties,this paper is based on YOLOv5(You Only Look Once 5)object detection algorithm as the benchmark to conduct research on the safety helmet detection task.It optimizes the original model at both aspects of detection accuracy and inference speed.The main contributions of this article are as follows:(1)An enhanced YOLOv5 model is proposed to improve the accuracy of the detection model,where bi-directional feature pyramid network(Bi FPN),attention mechanism,and transfer learning are fully integrated.The Bi FPN is taken to replace the original feature pyramid network(FPN)via adding additional cross layer edges with adaptive connecting weights.Attention mechanism is added after the end of backbone and neck network to let the network pay more attention on the interested region.Transfer learning is adopted for model training.The model is pre-trained by a head detection database and then fine-tuned by the helmet database.The proposed enhanced YOLOv5 is tested on a public GDUT-HWD dataset,where both helmet and its color can be identified.This study achieves the accuracy at 93.3%,which is 4.8% higher than that of the original YOLOv5.(2)A lightweight YOLOv5 model is proposed to improve the inference speed.The original backbone network is reconstructed,and the whole network model is channel pruned.In this paper,the lightweight module of the latest YOLOv5 version is first adopted,and a scale of detection layer was clipped in combination with the characteristics of safety helmet data set,which greatly reduces the complexity of the model.Finally,the network model is pruned to further remove redundant information.(3)An Open VINO+Raspberry Pi detection system based on lightweight YOLOv5 model is proposed.In this paper,Raspberry Pi is used as the deployment platform,Open VINO framework and neural accelerator rod are adopted to accelerate inference.This paper realizes the wearing detection based on the safety helmet of Raspberry Pi,which makes it possible to run a large convolutional network model on the Raspberry Pi,and provides an idea and scheme for the landing and deployment of the target detection model.