ETRI-Knowledge Sharing Plaform



논문 검색
구분 SCI
연도 ~ 키워드


학술대회 Lightweight Object Recognizer for Edge Devices
Cited 0 time in scopus Download 5 time Share share facebook twitter linkedin kakaostory
홍승태, 김정시, 정영준
International Conference on Information and Communication Technology Convergence (ICTC) 2022, pp.1729-1731
22HS6200, 운동능력 강화 자율 소프트슈트 기술 개발, 이동우
Recently, as personal edge devices such as smartphones have become popular, real-time object recognizers are attracting attention. Edge devices not only have limited memory and storage, but also have various types of HW platforms depending on the purpose of use. In particular, depending on the GPU type of edge devices, whether to use a parallel processing framework essential for deep learning computation acceleration such as OpenCL or CUDA is determined. Therefore, for real-time object recognition in various edge devices, it is necessary to support deep learning computation acceleration optimized for the GPU structure of each edge device. To this end, in this paper, we propose a deep learning framework that analyzes the GPU structure of edge devices and automatically applies the optimal parallel processing technique (OpenCL or CUDA) according to the GPU type. In addition, based on the proposed deep learning framework, we implement a real-time object recognizer based on the field image on the edge device.
KSP 제안 키워드
Deep learning framework, Edge devices, Field image, Object Recognition, Object recognizer, Real-Time, based on the field, computation acceleration, deep learning(DL), parallel processing technique