ETRI-Knowledge Sharing Plaform



논문 검색
구분 SCI
연도 ~ 키워드


학술대회 Accelerating Face De-identification System for Real-time Video Surveillance Services
Cited 1 time in scopus Download 0 time Share share facebook twitter linkedin kakaostory
김량수, 유학, 류지형, 김성창
International Conference on Information and Communication Technology Convergence (ICTC) 2022, pp.1477-1479
22PK1700, 건물 분산사업장 대상 클라우드 에너지 관리시스템 핵심 기술 개발 및 실증 연구, 유학
With an increased interest in personal information protection, there are numerous efforts to reduce personal information leakage caused by existing video surveillance services that continuously monitor and analyze multiple video streams at a surveillance control center. In particular, the face images included in the video stream are the most vulnerable information to easily identify specific personal information. In this paper, we propose a real-time face de-identification system that finds the locations of faces through a face detection deep learning model and modifies the corresponding pixels to make the faces unrecognizable. In order to accelerate the overall process, we apply TVM-based deep learning inference optimization and task-level pipeline parallelism to the proposed face de-identification system. Throughout experimental results, we show the effectiveness of the proposed system for providing face de-identification service in real-time.
KSP 제안 키워드
Face Image, Face detection, Identification system, Information Leakage, Learning model, Real-time video, control center, deep learning(DL), face de-identification, multiple video streams, overall process