Research On Helmet Detection Algorithm Based On Deep Learning

With the rapid development of science and technology and the increasing emphasis on modern industrial security,video surveillance has begun to shift from manual monitoring to intelligent monitoring.As the most important protective equipment for building construction personnel to ensure their own safety,helmets are required to be worn during the operation of the construction site.Helmet identification detection is an important part of intelligent monitoring of building sites,and it is also a research area of computer vision practice.For traditional digital image processing algorithms,illumination problems,target object occlusion,and shape and size problems are challenging for detection accuracy and target location.The lack of robustness of traditional digital image processing algorithms under various challenges is an important reason why monitoring intelligence has not been realized before.With the continuous discussion of neural networks by researchers,convolutional neural networks can theoretically fit arbitrary functions by virtue of their nonlinear transformation characteristics,and their generalization ability is also excellent,so they are on object recognition and detection tasks.It is widely used,which also enables deep learning recognition detection technology to be applied in intelligent monitoring.This paper takes the intelligent monitoring of building sites as the research background,studies the detection and positioning technology of helmets,and combines the theory of computer vision and deep learning to study how to improve the overall accuracy and robustness of helmet detection and positioning.The theme mainly carries out the following work:(1)The traditional digital image processing algorithm and deep learning detection algorithm are used to study the identification and detection of the helmet.Firstly,the color space of the construction site is studied.The Cr variable and Cb variable of YCr Cb color space are used as the reference for the image front and back segmentation of skin color detection.Then the ellipse detection algorithm is used to locate the human head from the segmented foreground.In the region,the color characteristics of the upper part of the human head are used to determine the wearing state of the helmet.The experimental results show that the detection effect of the traditional digital image processing algorithm is barely qualified under different illumination and helmet occlusion.Aiming at the situation of target missed detection in the traditional method,a safe helmet detection algorithm based on Faster R-CNN is proposed.Under the same circumstances,the detection effect is obviously improved.(2)In the framework of Faster R-CNN detection algorithm,the recognition accuracy of small target objects is low and the accuracy of target position detection is not ideal.This paper proposes a helmet detection algorithm based on residual network model and position sensitive layer.By using data augmentation operations to enhance the diversity of data types,it is verified that data diversity can improve the robustness of the network model.The improved helmet detection algorithm and Faster R-CNN algorithm are compared.The experimental results show that the residual network model and the position sensitive layer can improve the detection accuracy and positioning ability.