Research On Target Detection Algorithm Based On R2-yolov5

The outbreak of the epidemic has made people realize that wearing masks can effectively reduce the spread of viruses,and likewise wearing masks in some special scenarios can protect people’s health and safety.Applying the target detection algorithm to mask detection can effectively supervise the relevant personnel to wear masks and reduce the risk of inhaling and spreading harmful substances.With the development of society,the aging problem is a major problem faced by China.Elderly people may have unstable walking and abnormal gait in the process of walking,and if they fall,it may lead to fracture or even paralysis,which seriously affects the quality of life and life safety of the elderly.In order to solve the above problems,this thesis applies the target detection algorithm to mask detection and elderly helper detection,and the main works are:(1)For the problems of poor detection and false detection in the process of mask detection due to low light and difficulty in small target detection,this thesis proposes a mask detection method based on R2C-yolov5 network.In the yolov5 network,the Bottleneck structure of C3 module is replaced with Res2 Net to form R2-C3 module,and combined with CBAM attention mechanism module to obtain R2C-yolov5 network.The experimental results show that the MAP of R2C-yolov5 network on the mask dataset is improved by 2.4% compared with yolov5 network,which can better perform the mask detection task.(2)To address the problems of poor detection and false detection due to different shooting angles and the presence of occlusion in elderly helper detection,this thesis proposes an elderly helper detection method based on R2F-yolov5 network.Based on the yolov5 network,a new feature fusion method is proposed,and the R2-C3 module and the feature fusion method are combined to obtain the R2F-yolov5 network.The experimental results show that the MAP of R2F-yolov5 network on the elderly helper dataset is improved by 2.2% compared with yolov5 network,and the MAP in Support class is improved by 6.5%,which can better accomplish the task of elderly helper detection.