基于深度学习的危险行为识别系统设计

Design of Dangerous Behavior Recognition System Based on Deep Learning

随着人们安全意识的不断增强,传统的人工监控系统已不能适应社会发展的需要,在数量巨大的监控视频数据面前,单纯依靠人来处理监视内容难以构成真正安全的系统。将计算机视觉技术应用到监控系统中,使计算机对监视视频中的内容进行自动理解、识别出危险行为并发出报警,已成为新一代的智能监控技术。针对YOLOv4目标检测算法方便部署与监控系统的特点,本文提出一种基于深度学习的典型危险行为识别系统设计方案,基于YOLOv4算法提出一种有优化网络结构的方法,提高了算法检测的准确度。实验测试表明,本文设计的系统能提高监控系统对危险行为的识别准确度,同时在公共场合发生危险行为时也可发出警报进行警示。

With the gradual improvement of people’s safety awareness,the traditional manual monitoring system can not meet the needs of social development.In front of a huge number of monitoring video data,it is difficult to rely solely on people to deal with the monitoring content to form a truly safe system.The application of computer vision technology to the monitoring system enables the computer to automatically understand the content in the monitoring video,identify dangerous behaviors and give an alarm,which has become a new generation of intelligent monitoring technology.According to the characteristics of easy deployment and monitoring system of YOLOv4 target detection algorithm,this paper proposes a design scheme of typical dangerous behavior identification system based on deep learning,and proposes a method to optimize the network structure based on YOLOv4 algorithm,which improves the accuracy of algorithm detection.The experimental test shows that the system designed in this paper can improve the recognition accuracy of the monitoring system for dangerous behavior,and can also send an alarm for warning when dangerous behavior occurs in public.