BROWSE
Detail
Conference Paper
Implementation of deep learning based intelligent image analysis on an edge AI platform using heterogeneous AI accelerators
Cited 2 time in 
Share 
Share
- Authors
- Ryangsoo Kim, Jaein Kim, Hark Yoo, Sung Chang Kim
- Issue Date
- 2023-10
- Citation
- International Conference on Information and Communication Technology Convergence (ICTC) 2023, pp.1347-1349
- Publisher
- IEEE
- Language
- English
- Type
- Conference Paper
- Abstract
- Recent advancements in artificial intelligence (AI) technology have spurred efforts to implement deep learning-based intelligent image analysis in various video surveillance applications. However, traditional approaches to using remote cloud computing platforms results in increased network latency and potential breaches of personal information. To address these issues, the adoption of edge AI, utilizing deep learning-based data analysis directly on the devices where image data is collected, has emerged as a promising solution. In this paper, we introduce a practical edge AI platform that enables real-time deep learning-based image analysis, including object detection and multi-person pose estimation. The platform is built on an embedded board equipped with heterogeneous AI accelerators, facilitating parallel inference of multiple deep learning-based image analysis models. In addition, the platform applies task-level pipeline parallelism to maximize the utilization of computing resources, which leads to a reduction in overall image analysis latency. Experimental results demonstrate the effectiveness of our edge AI platform in providing real-time intelligent video analysis services.
- KSP Keywords
- Analysis Model, Analysis services, Artificial intelligence (AI) technology, Cloud computing platforms, Computing resources, Data analysis, Embedded board, Image analysis, Image data, Intelligent Video Analysis, Learning-based