Research And Implementation Of The Detection System Of Safety Helmet Wearing Based On Video Surveillance

With the rapid development of pattern recognition and computer hardware technology and the wide application of artificial intelligence in all walks of life,intelligent analysis of industrial video surveillance has become one of the hot topics in the field of computer vision.Safety helmet wearing is an important part of the standard wear on the industrial site,which can effectively protect the life safety of workers.Therefore,it is of practical significance to detect whether the helmet is worn or not and to identify those who do not wear the helmet under video surveillance.Based on the above considerations,on the basis of the existing deep learning theory,this paper studies the detection of helmet wearing in the complex environment,and realizes face recognition for the person without helmet.In order to solve the problem that the detection background of helmet wearing is complex and the detection target is small in the real scene,this paper introduces SSD algorithm based on deep learning to realize helmet wearing detection.The algorithm recognizes whether the detection target is wearing a helmet.If the detection target is wearing a helmet,the face and the helmet are positioned as a whole in the image.Otherwise,only the face is positioned.Experiments show that the detection algorithm based on SSD has high detection accuracy and can effectively implement the detection of safety helmet wearing in a variety of application scenarios.However,the method has low detection speed,high computational complexity,high memory consumption and high hardware performance requirements.In order to solve the shortcomings of SSD detection algorithm,this paper introduces Mobile Net to replace the basic network VGG in SSD,and makes a theoretical analysis of Mobile Net and explores the reason why it can accelerate the detection speed.Finally,experiments show that under the same experimental platform and data,Mobile Net-SSD greatly improves the detection speed of helmet wearing with a small sacrifice of accuracy.After the helmet wearing detection,it is also necessary to carry out face recognition for the target without the helmet.In this paper,a Siamese network with simple structure and high robustness for face recognition under non-limited conditions is designed.At the same time,inorder to accelerate the convergence of the model,reduce the training time and simplify the training process,the cyclical learning rate strategy was introduced.Experiments show that the algorithm has high recognition accuracy and is robust to complex environments.Finally,in order to flexibly implement the helmet wearing detection and face recognition functions in the actual scene and can no longer rely on Caffe framework,based on the research of detection algorithm and recognition algorithm,this paper develops a helmet wearing detection and face recognition system.The test results show that the system has high real-time performance and can achieve fast detection and recognition on both CPU and GPU.