摘要
视频智能分析目前已广泛应用于安防、交通等领域,目前识别人体、人脸、车辆、场景等信息的算法和技术已经成熟,但是针对基建场景下的危险源、不规范作业等识别技术还未普及。国网上海市电力公司基建视频监控中心具备1000路视频流实时接入能力,海量视频流无法通过传统人工盯屏方式实现基建安全管控,因此需要引入人工智能技术实现安全管控。文章基于深度神经网络技术,提出了基建视频智能分析系统,选择了一种适用于多路视频流实时分析的深度学习模型,实现了火灾、吸烟、人员着装、坑洞识别,在识别率和实时性上取得较好的结果。该系统能够实时识别出危害施工安全的行为,具有一定的推广和使用价值。
Video intelligence analysis has been widely used in security,transportation and other fields.At present,the algorithms and technologies for identifying human body,face,vehicle,scene information is getting matured,but the identification technology for dangerous sources and irregular operations under the infrastructure scene has not drawn great attention.State Grid Shanghai Electric Power Company infrastructure video surveillance center has 1000 channels of real-time access to video streams.The massive video cannot be achieved by traditional way of human eye screen construction safety control.Therefore,it is necessary to introduce artificial intelligence technology to achieve security control.Based on deep neural network technology,this paper proposes a real-time analysis framework of infrastructure video,and selects a deep learning model suitable for real-time analysis of multi-channel video streams,which realizes fire,smoking,personnel dressing,pothole identification,and obtains good results in real time and recognition rate.The system can identify behaviors that endanger construction safety in real time,and has certain promotion and application value.
作者
孙宇飞
顾书玉
李宾皑
李颖
SUN Yufei;GU Shuyu;LI Binai;LI Ying(Nanjing Chiebot Technologies Co.,Ltd.,Nanjing 211100,China;State Grid Shanghai Electric Power Company Construction Department,Shanghai 200120,China)
出处
《电力信息与通信技术》
2020年第2期69-74,共6页
Electric Power Information and Communication Technology
关键词
视频智能分析
基建
视频监控
深度神经网络
video intelligence analysis
infrastructure
video surveillance
deep neural network
