Application Research Of Improved Small Object Detection Method In Helmet Wearing Detection

In the safety management of the construction site,it is very important to supervise the construction personnel’s helmet wearing,but the efficiency of manual supervision is often relatively low.Therefore,it is very necessary to use intelligent technology to detect helmet wearing automatically.However,the current helmet wearing detection algorithms have many problems,such as low detection accuracy of small targets and poor adaptability in complex scenes.Therefore,aiming at the problem of low detection accuracy of small targets,this paper proposes an improved small target detection method for helmet wearing detection by combining with deep learning technology.The work and achievements of this paper are as follows:(1)The principle,advantages and disadvantages of the existing clustering methods were analyzed.In view of the defects existing in the method,the K-means clustering method was optimized.The optimized clustering method was applied to the helmet data set,and the size of the helmet detection box was obtained through clustering experiments.(2)A small target detection method with enhanced feature fusion is proposed and applied to helmet wearing detection.Aiming at the problem of low detection accuracy of small and medium-sized targets in the current helmet wearing detection algorithm,a model was built by the fusion of deep features and shallow features,the detection output of shallow network was increased,and the classification and location of small targets such as helmet detection were enhanced.Through the comparison of the helmet wearing testing experiments of the three models,it is proved that the improved method has a great improvement in the detection accuracy,and the model obtained can effectively detect the helmet.(3)In order to ensure the real-time performance of helmet wearing detection under video surveillance,the model in this paper was accelerated.The safety helmet wearing detection model is processed by compression clipping,and the parameters that need to be clipped are screened out.The threshold value and clipping ratio are used to cut the model,so as to improve the detection speed.In order to verify the effectiveness of the acceleration model,a clipping experiment was carried out on the helmet wearing detection model according to 11 clipping ratios,and the optimal model was selected through data analysis.The experiment proves that the accelerated helmet wearing detection model FPS exceeds the frame rate of traditional monitoring,and the model can realize helmet wearing detection more effectively after improving the accuracy and speed.