基於 YOLO 之輕量化色球辨識物件偵測模型於超低功耗邊緣裝置之應用;YOLO-Based Lightweight Object Detection for Color Ball Recognition on Ultra-Low Power Edge Devices

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Electrical Engineering > Electronic Thesis & Dissertation > Item 987654321/98547

Please use this identifier to cite or link to this item: https://ir.lib.ncu.edu.tw/handle/987654321/98547

Title:	基於 YOLO 之輕量化色球辨識物件偵測模型於超低功耗邊緣裝置之應用;YOLO-Based Lightweight Object Detection for Color Ball Recognition on Ultra-Low Power Edge Devices
Authors:	陳旻盛;Chen, Min-Sheng
Contributors:	電機工程學系
Keywords:	邊緣裝置;物件偵測;微控制器;深度學習
Date:	2025-08-20
Issue Date:	2025-10-17 12:54:56 (UTC+8)
Publisher:	國立中央大學
Abstract:	隨著人工智慧技術在各領域迅速發展，如何將 AI 功能有效整合至終端裝置，成為智慧系統設計的重要挑戰。傳統雲端運算雖具備強大計算能力，但在即時性要求高的場景中，常因網路延遲、頻寬限制與資料隱私風險而無法滿足需求。相較之下，邊緣運算（Edge AI）能在本地裝置上完成推論與決策，降低資料傳輸延遲、提升反應速度，並避免敏感資訊上傳，有助於保障使用者隱私與降低能耗，因此成為取代雲端的理想解決方案。本研究提出一套輕量化的色球偵測系統，基於簡化後的 YOLO 架構，並針對資源受限的邊緣平台進行最佳化。系統部署於 STM32L4、STM32H7、i.MX 8M Plus 以及 MAX78000 等低功耗邊緣裝置上，展現出良好的即時處理能力與整體效能。此系統特別適合應用於智慧玩具、互動式遊戲機與可攜式設備等情境，不依賴外部伺服器即可完成影像辨識與回應。綜合而言，本研究展示了邊緣 AI 在即時性、低功耗與資料安全等面向的應用潛力，為未來智慧終端設備提供了可行且具擴展性的解決方案。 ;With the rapid advancement of artificial intelligence across various domains, the effective integration of AI capabilities into end devices has become a critical challenge in intelligent system design. While traditional cloud computing provides substantial computational power, it often falls short in latency-sensitive scenarios due to network delays, bandwidth constraints, and data privacy concerns. In contrast, edge computing (Edge AI) enables on-device inference and decision-making, thereby reducing communication latency, enhancing responsiveness, and mitigating the risk of transmitting sensitive data. As a result, Edge AI has emerged as a compelling alternative to cloud-based approaches, particularly for applications requiring real-time performance, energy efficiency, and data confidentiality. This study presents a lightweight color ball detection system based on a streamlined YOLO architecture, specifically optimized for deployment on resource-constrained edge platforms. The proposed system is implemented on low-power hardware platforms, both with and without integrated neural network accelerators, such as the MAX78000, STM32L4, STM32H7, and i.MX 8M Plus, demonstrating robust real-time processing capabilities and efficient operational performance. Owing to its compact design and independence from external servers, the system is well-suited for applications in smart toys, interactive gaming machines, and portable embedded devices. Overall, this work substantiates the feasibility of deploying deep learning models on edge platforms, and highlights the potential of Edge AI in delivering scalable, low-latency, and privacy-preserving solutions for next-generation intelligent systems.
Appears in Collections:	[Graduate Institute of Electrical Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	8	View/Open

社群 sharing

Loading...