Study On The Algorithms Of Safety Helmet Wearing Detectionl And System Implementation Based On Umanned Aerial Vehicle And YOLOv4

In housing construction and municipal engineering,safety accidents lead to an increasing number of casualties.Wearing safety helmets is the most basic preventive measure to protect the life safety of construction workers in the safety production regulations.However,in the actual supervision process,it is difficult to rely on individual management personnel to inspect the safety helmet wearing conditions of the entire construction site in a timely manner.China’s 14th Five-Year Plan clearly points out the industry development goal of"digitalization".How to promote the intelligent transformation of construction technology and promote the integrated development of the construction industry and contemporary information technology is the frontier research focus in the current construction field.Based on this research background,this paper uses unmanned aerial vehicle(UAV)as detection equipment and deep learning technology as detection method,and proposes a helmet detection technology based on UAV and YOLOv4.The specific work is as follows:(1)Most of the existing helmet data sets are obtained from crawlers and construction site surveillance videos,which are not suitable for aerial photography helmet detection tasks.Therefore,this paper uses drones and Labelimg labeling tools to construct a set of aerial photography helmet data sets,and uses data enhancement technology to expand the data set.Finally,the effect of data enhancement technology on model detection is analyzed through comparative experiments.(2)Since the original YOLOv4 has a low recall rate and serious missed detection on the aerial helmet data set,the backbone network and feature fusion network of YOLOv4 are optimized.In order to verify the effectiveness of the improved YOLOv4 network in the aerial helmet detection task,three groups of experiments were designed,of which two groups of ablation experiments were used to verify the influence of the backbone network and other optimization techniques on the model detection effect.The third set of comparative experiments is used to verify the overall detection advantage of the improved YOLOv4 compared to other networks.(3)In order to give full play to the practical value of this research work,an Androidbased aerial helmet detection system is constructed.After running the system on a mobile phone with Android system,you can control the UAV and use YOLOv4 algorithm to complete the automatic supervision of the helmet.Finally,the detection accuracy of the system in practical application is verified by testing 200 helmet detection images at different construction stages.In terms of theoretical research,the improved YOLOv4 algorithm proposed in this paper can not only effectively deal with the task of helmet detection under aerial photography conditions,but its improved strategy also has certain reference value for the detection of small objects in remote sensing images.In terms of practical application,the aerial photography safety helmet detection system constructed in this paper can be widely used in the safety helmet supervision operation of housing construction and municipal engineering,which promotes the effective implementation of safety production regulations and the scientific and technological progress of safety production supervision mechanism.