sesa 20(23): e3

Research Article

Minor Privacy Protection by Real-time Children Identification and Face Scrambling at the Edge

Download63 downloads
  • @ARTICLE{10.4108/eai.13-7-2018.164560,
        author={Alem Fitwi and Meng Yuan and Seyed Yahya Nikouei and Yu Chen},
        title={Minor Privacy Protection by Real-time Children Identification and Face Scrambling at the Edge},
        journal={EAI Endorsed Transactions on Security and Safety},
        volume={7},
        number={23},
        publisher={EAI},
        journal_a={SESA},
        year={2020},
        month={5},
        keywords={Child Detection, Minor Privacy Protection, Smart Surveillance, Video Feature Extraction, Decentralization},
        doi={10.4108/eai.13-7-2018.164560}
    }
    
  • Alem Fitwi
    Meng Yuan
    Seyed Yahya Nikouei
    Yu Chen
    Year: 2020
    Minor Privacy Protection by Real-time Children Identification and Face Scrambling at the Edge
    SESA
    EAI
    DOI: 10.4108/eai.13-7-2018.164560
Alem Fitwi1, Meng Yuan1, Seyed Yahya Nikouei1, Yu Chen1,*
  • 1: Dept. of Electrical and Computer Engineering, Binghamton University, Binghamton, NY 13902, USA
*Contact email: ychen@binghamton.edu

Abstract

The collection of personal information about individuals, including the minor members of a family, by closed circuit television (CCTV) cameras creates a lot of privacy concerns. Revealing children’s identifications or activities may compromise their well-being. In this paper, we propose a novel Minor Privacy protection solution using Real-time video processing at the Edge (MiPRE). It is refined to be feasible and accurate to identify minors and apply appropriate privacy-preserving measures accordingly. State of the art deep learning architectures are modified and repurposed to maximize the accuracy of MiPRE. A pipeline extracts face from the input frames and identify minors. Then, a lightweight algorithm scrambles the faces of the minors to anonymize them. Over 20,000 labeled sample points collected from open sources are used for classification. The quantitative experimental results show the superiority of MiPRE with an accuracy of 92.1% with near real-time performance.