Research On The Application Of Safety Apparatus Detection Based On Edge Computing In Work Site

It is very important to detect the safety protective equipment on the construction site to protect the construction workers from damage.In the past,it was realized by manpower or traditional detection methods.With the rapid development of deep learning and computer vision technology,the target detection method based on deep learning gradually replaces the traditional target detection method.However,this method requires GPU or cloud server platform with a lot of computing power,There are many disadvantages,such as high computing cost,high transmission cost,large system delay and so on,so it is difficult to deploy in the Internet of things mobile devices.Therefore,in view of the above shortcomings of the current detection and methods of safety protection equipment,this paper takes the safety helmet detection as an example to study,the main work is as follows:(1)At present,it is difficult to deploy the safety helmet detection on the edge of Internet of things.This paper proposes a method based on edge computing,which uses low-cost Raspberry Pi 3B + as the edge computing device,Raspberry Pi camera(V2)as the video capture module and 7-inch display as the video display module(2)Aiming at the computational power requirements of deep learning target detection algorithm,the lightweight neural network is used to reduce the problem of insufficient computing power of edge computing equipment.The single-stage detection algorithm Tiny-YOLOV3 is selected as the algorithm research in this paper.At the same time,the algorithm is optimized.Six prior frames are obtained by K-means clustering to optimize the algorithm.At the same time,the moasic data enhancement method is used,It is used to prevent over fitting phenomenon of algorithm model,so as to strengthen the generalization ability of algorithm model;In order to improve the accuracy of the algorithm,residual structure and joint level maximum pooling feature fusion are added in the backbone network to ensure that the algorithm can be improved under the allowable computing power of edge computing devices.(3)In view of the limited computing power of edge computing devices,in order to achieve better detection results,this paper proposes to deploy OpenVINO ncs2 USB interface device as edge accelerator in Raspberry Pi 3B+.With OpenVINO tool configured,the model can be optimized and quantified,and the reasoning process of algorithm model can be further improved.The experimental results show that the improved Tiny-YOLOV3 algorithm has good detection accuracy and generalization ability in the real-time detection system of safety helmet wearing on construction site.The map reaches 83.7%,and the detection speed reaches 6fps.