整合YOLOv11-Pose與ByteTrack於邊緣運算平台之老年照護自主跟隨機器人;Edge-Enabled Autonomous Follower Robot for Elderly Care via Integration of YOLOv11-Pose and ByteTrack

NCUIR > College of Electrical Engineering & Computer Science > Executive Master of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/98196

Please use this identifier to cite or link to this item: https://ir.lib.ncu.edu.tw/handle/987654321/98196

Title:	整合YOLOv11-Pose與ByteTrack於邊緣運算平台之老年照護自主跟隨機器人;Edge-Enabled Autonomous Follower Robot for Elderly Care via Integration of YOLOv11-Pose and ByteTrack
Authors:	高振耀;Kao, Chen-Yao
Contributors:	資訊工程學系在職專班
Keywords:	YOLO;ByteTrack;邊緣運算;機器人
Date:	2025-07-28
Issue Date:	2025-10-17 12:28:37 (UTC+8)
Publisher:	國立中央大學
Abstract:	全球高齡化使居家照護需求急升，傳統人力服務愈顯不足且難兼顧長者安全與自主。為解決此問題，本研究針對光線多變且狹小的動態家居環境，提出一款以邊緣運算為核心的自主跟隨機器人平台。系統融合改良之YOLOv11-Pose進行人體骨架聯合偵測，配合ByteTrack多目標追蹤以維持穩定ID，並透過RGB-D SLAM與OctoMap建構即時三維佔用地圖，以結合MIAT模組化架構。於NVIDIA Jetson Orin Nano實測，端到端閉迴路平均延遲85 ms，核心推論速率11.8 fps，追隨誤差±10 cm。結果證實，本方法能在低功耗裝置上提供近工業級精度與即時性，顯示深度視覺結合模組化設計可有效填補居家照護人力缺口，並為智慧照護機器人提供可擴充且具成本效益的實作範例。;The accelerating global aging trend is driving a surge in home-care demand, yet conventional, labor-intensive services can no longer balance the safety and autonomy of older adults. To bridge this gap, we present an edge-computing–centered autonomous follower-robot platform designed for dynamic, light-varying, and space-constrained residential settings. The system fuses an enhanced YOLOv11-Pose network for joint human-body and skeleton detection with ByteTrack multi-object tracking to maintain stable IDs, while RGB-D SLAM and OctoMap cooperate to construct a real-time 3D occupancy map. These perception and mapping modules are encapsulated within a MIAT modular architecture for efficient data flow. Deployed on an NVIDIA Jetson Orin Nano, the prototype achieves an end-to-end closed-loop latency of 85 ms, an overall closed loop rate of 11.8 fps, and a following error of ±10 cm. The results confirm that the proposed approach delivers near-industrial accuracy and real-time responsiveness on low-power hardware, demonstrating that deep-vision techniques married with modular design can effectively mitigate caregiver shortages in home settings while offering a scalable, cost-efficient blueprint for smart-care robots.
Appears in Collections:	[Executive Master of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	17	View/Open

社群 sharing

Loading...