Low-power real-time CNN-based hand tracking hardware for mobile devices모바일 디바이스를 위한 CNN기반 저전력 실시간 핸드 트랙킹 하드웨어
A low-power 3D hand gesture tracking (HGT) processor is proposed for real-time user interaction in smart mobile devices. It features: 1) 1.80TOPS/W CNN-Stereo core for energy-efficient depth sensing; 2) Triple ping-pong buffers with workload balancing to reduce 23.9% processing time; and 3) Nearest neighbor searching processing-in-memory for hand tracking to achieve 2.8x energy efficiency. The proposed SoC facilitates real-time 3D HGT with 33.5mW and maximally 10.8mm error.
- Description
- 한국과학기술원 :전기및전자공학부,
- Publisher
- 한국과학기술원
- Issue Date
- 2019
- Identifier
- 325007
- Language
- eng
- Description
학위논문(박사) - 한국과학기술원 : 전기및전자공학부, 2019.8,[v, 97 p. :]
- Keywords
CNN processor▼a3D hand tracking▼aProcessing in memory▼aMemory efficient hardware▼aLow power; 컨볼루션 신경망 프로세서▼a3D 핸드 트래킹▼a컴퓨팅 메모리▼a메모리 효율적 하드웨어▼a저전력
- Appears in Collection
- EE-Theses_Ph.D.(박사논문)
- Files in This Item
- There are no files associated with this item.
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.