BROWSE
Detail
Conference Paper
비동기 큐 파이프라인 기법을 통한 임베디드 환경에서 추론 처리량 개선
Cited - time in 
Share 
Share
- Issue Date
- 2026-05
- Citation
- IEMEK Symposium on Embedded Technology (ISET) 2026, pp.1-3
- Publisher
- 대한임베디드공학회
- Language
- Korean
- Type
- Conference Paper
- Abstract
- This paper proposes a CPU-NPU pipelined inference architecture for real-time object detection on embedded platforms. Four execution structures, CPU-only, sequential, synchronized, and asynchronous queue-based, are evaluated on a Raspberry Pi 5 equipped with a Hailo-10H NPU. The proposed asynchronous pipeline achieves 41.28 FPS, representing a 12.9× improvement over the CPU-only baseline and 38.8% over the synchronized pipeline.
- Keyword
- Edge AI, Neural Processing Unit (NPU), Asynchronous Queue, Heterogeneous Computing
- KSP Keywords
- Asynchronous pipeline, Embedded platform, Neural processing, Processing unit, Raspberry PI, heterogeneous computing, real time