A Method Of Worker’s Helmet Detection And Person Identification Based On Computer Vision And Deep Learning

Safety helmet detection and identity recognition based on computer vision have always been hot topics in construction on-site safety monitoring research.At present,many studies can effectively identify workers without safety helmets,but it is difficult to determine the identity information of workers.Therefore,in order to implement safety management more effectively,this paper proposes a comprehensive method of safety helmet detection and identity recognition based on computer vision and deep learning,which can identify the identity of construction workers while carrying out safety helmet detection.This paper adopts the neural network algorithm based on YOLOV3 for safety helmet detection.The data are derived from the public dataset SHWD(Safety Helmet Wearing Detect Dataset)and 3000 images taken from on-site construction surveillance video.Firstly,YOLOV3 algorithm was pre-trained with SHWD dataset to improve the generalization ability.Then,the performance test of helmet detection was carried out using the images captured by on-site construction video.Light CNN(convolutional neural network)algorithm is adopted for identity recognition of construction workers in this paper.Considering that the environment onsite and video features are easy to generate noise pollution to images,the method in this paper meets real-time and rapid detection requirements.Firstly,the construction workers are identified by face feature extraction,and then the data set of field screenshots is used as the training and testing set of Light CNN.In this paper,the above two methods are integrated to solve safety helmet detection and identity recognition.Firstly,through the YOLOV3 deep learning model,the workers’ head and helmet features are extracted from the input images to detect whether the helmet is worn.Then the image is input into Light CNN for face recognition.In the face recognition process,the face in the image is detected through the Haar cascade classifier,and the detected face is input into Light CNN for feature extraction and identification recognition.In order to verify the feasibility and effectiveness of this method,this study intercepted 3,000 images under different lighting conditions from the video on-site to test the performance of the YOLOV3 detection method.Similarly,the performance detection of the Light CNN face recognition method is mainly tested and compared in the public database.The experimental results show that YOLOV3 has the advantages of high detection accuracy,fast speed,high recall rate and strong robustness.Light CNN has an identity recognition rate of 77.53%,which can effectively identify workers.