Research And Implementation Of Helmet Wearing Detection System Based On Improved YOLOv

Safety operations are currently of concern to various industries and most safety measures still rely on manual monitoring.With the development and maturity of deep learning technology,the corresponding object detection algorithms have been studied more and more,and helmet wear detection systems have also become important research content.Deep learning-based helmet wear detection system solves the problems of low detection accuracy,low rate and poor generalization in traditional helmet wear detection system.The existing detection algorithms used in helmet wearing detection systems based on deep learning still have problems with the same IoU values but different intercept targets,problems with the target function and evaluation metrics not being uniform,and problems with uneven positive and negative samples that make model training more difficult.In this thesis,a new computational method for the overlap degree of two boundary boxes is proposed,based on the idea of the new computational method,the improved YOLOv3 algorithm is implemented,the detection accuracy is improved,and finally the research and implementation of the helmet wearing detection system based on the improved YOLOv3 algorithm is completed.The specific research and realization of this thesis is as follows.First,there is the problem of incomplete target interception for the same IoU value.The problem of incomplete target interception is mitigated by a new method of calculating the degree of overlap between the two boundary boxes,which is proposed after a study of the GIoU calculation method.Second,to address the problem of inconsistency between the objective function and the evaluation indicator in the YOLOv3 algorithm.The problem causes the same error value and different IoU values,making the model less stable.The objective function of the YOLOv3 algorithm is improved based on a new method for calculating the degree of overlap between two boundary boxes.The experimental results of the public dataset VOC2007 showed that the improved objective function improved the stability of the model compared to the YOLOv3 algorithm,and the mAP value of the model improved by 1.24%.Thirdly,addressing the problem of uneven positive and negative samples in the YOLOv3 algorithm.This problem causes the training model to be more difficult.The method for calculating the ignore mask in the YOLOv3 algorithm is improved based on a new method for calculating the degree of overlap between two boundary boxes.The experimental results of the public dataset VOC2007 showed that the mAP value of the model improved by 1.03%compared to the YOLOv3 algorithm.Finally,the improved YOLOv3 algorithm was implemented based on the improved objective function as well as the improved ignore mask,and the experimental results of the public dataset VOC2007 showed a 2.07%improvement in the mAP value of the improved YOLOv3 algorithm compared to the YOLOv3 algorithm.Based on the improved YOLOv3 algorithm,the helmet wear detection system is designed and implemented according to the actual requirements of helmet wear detection,the main functions include real-time detection of helmet wear,detection reporting and processing,problem management and statistics.The research and development of this system provides a new research direction for the effective identification and object detection algorithm for helmet wear in the field,and lays the foundation for the establishment of a helmet wear detection system and development platform with independent intellectual property right,which is of great significance and popular application value.