EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Optimizing Real-Time Object Detection in a Multi-Neural Processing Unit System

Sehyeon Oh iD, Yongin Kwon iD, Jemin Lee iD

Sensors · 24.02.2025 · 25 (5) · 1376

DOI: 10.3390/s25051376 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Sensors, online erschienen am 24.02.2025. Verfasst von Sehyeon Oh und 2 weiteren Autor:innen. Beteiligte Einrichtung: Department of Artificial Intelligence, University of Science and Technology, Daejeon 34113, Republic of Korea. Die Arbeit wurde bisher 6-mal zitiert (Crossref) und nennt 18 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: the Institute of Information \& Communications Technology Planning \& Evaluation(IITP), Institute of Information & Communications Technology Planning & Evaluation (IITP).

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Real-time object detection demands high throughput and low latency, necessitating the use of hardware accelerators. NPU is specialized hardware designed to accelerate the calculation of deep learning models, providing better energy efficiency and parallel processing performance than existing CPUs or GPUs. In particular, it plays an important role in reducing latency and improving processing speed in applications that require real-time processing. In this paper, we construct a real-time object detection system based on YOLOv3, utilizing Neubla’s Antara NPU, and propose two approaches for performance optimization. First, we ensure the continuity of NPU inference by allowing the CPU to process data in advance through double buffering. Second, in a multi-NPU environment, we distribute tasks among NPUs through queue-based processing and analyze the performance limits using Amdahl’s law. Experimental results demonstrate that compared to a CPU-only environment, applying the NPU in single buffering improved throughput by 2.13 times, double buffering by 3.35 times, and in a multi-NPU environment by 4.81 times. Latency decreased by 1.6 times in single and double buffering, and by 1.18 times in the multi-NPU environment. The accuracy remained consistent, with 31.4 mAP on the CPU and 31.8 mAP on the NPU.

Schlagwörter

Stichwörter der Autor:innen

YOLOv3double bufferingneural processing unitqueue-based processingreal-time object detection

Autor:innen und Institutionen

  • Department of Artificial Intelligence, University of Science and Technology, Daejeon 34113, Republic of Korea.
  • Electronics and Telecommunications Research Institute, Daejeon 34129, Republic of Korea.

ORCID-Profile: Sehyeon Oh, Yongin Kwon, Jemin Lee

Förderung

  • the Institute of Information \& Communications Technology Planning \& Evaluation(IITP)
  • Institute of Information & Communications Technology Planning & Evaluation (IITP)

Zitieren

APA

Oh, S., Kwon, Y., & Lee, J. (2025). Optimizing Real-Time Object Detection in a Multi-Neural Processing Unit System. Sensors, 25(5), 1376. https://doi.org/10.3390/s25051376

Vancouver

Oh S, Kwon Y, Lee J. Optimizing Real-Time Object Detection in a Multi-Neural Processing Unit System. Sensors. 2025;25(5):1376. doi:10.3390/s25051376

BibTeX
@article{Oh2025,
  author = {Oh, Sehyeon and Kwon, Yongin and Lee, Jemin},
  title = {Optimizing Real-Time Object Detection in a Multi-Neural Processing Unit System},
  journal = {Sensors},
  year = {2025},
  volume = {25},
  number = {5},
  pages = {1376},
  doi = {10.3390/s25051376}
}
Kostenlos starten →