Proceedings of the 13th EAI International Conference on Mobile Multimedia Communications, Mobimedia 2020, 27-28 August 2020, Cyberspace

Research Article

Light Deep Learning based Edge Safety Surveillance

Download465 downloads
  • @INPROCEEDINGS{10.4108/eai.27-8-2020.2295046,
        author={Yimo  Lou and Wengang  Cao and Zhimin  He and Guan  Gui},
        title={Light Deep Learning based Edge Safety Surveillance},
        proceedings={Proceedings of the 13th EAI International Conference on Mobile Multimedia Communications, Mobimedia 2020, 27-28 August 2020, Cyberspace},
        publisher={EAI},
        proceedings_a={MOBIMEDIA},
        year={2020},
        month={11},
        keywords={intelligence safety surveillance centernet mobilenet-v2 marginal devices},
        doi={10.4108/eai.27-8-2020.2295046}
    }
    
  • Yimo Lou
    Wengang Cao
    Zhimin He
    Guan Gui
    Year: 2020
    Light Deep Learning based Edge Safety Surveillance
    MOBIMEDIA
    EAI
    DOI: 10.4108/eai.27-8-2020.2295046
Yimo Lou1, Wengang Cao1, Zhimin He1, Guan Gui2,*
  • 1: College of Telecommunication and Information Engineering, Nanjing University of Posts and Telecommunications, Nanjing, 210003 China
  • 2: Nanjing University of Posts and Telecommunications
*Contact email: guiguan@njupt.edu.cn

Abstract

Safety is considered as the first important factor in many industries such as construction sites. Hence, artificial intelligence based safety surveillance techniques have been received strong attentions in recent years. Conventional surveillance systems for monitoring whether the workers wearing helmets are not easy to install and carry, and the largest trouble is that the system needs considerable computation, which is not that simple to satisfy the requirement of hardware. Considering the characteristic about construction sites, in this paper, we proposed a new system based on CenterNet with MobileNet-V2 as backbone. It has a video camera, a marginal device embedded with Jetson TX2 and wireless communication routers to ensure real-time transmission about live-scene about construction sites. After inspection, the light-weight network we proposed can be run in portable marginal device smoothly and stably with slight loss of average precision.