在積極部署5G及物聯網普及的智慧環境中，邊緣運算與機器學習成為提升智慧環境效能的關鍵技術。本研究探討「雲—邊—端」架構在智慧物聯網（AIoT）會議室中的應用，評估其在環境控制、設備管理及人機互動的成效。傳統雲端運算雖提供強大計算能力，但頻繁數據傳輸易造成延遲與隱私風險。因此，本研究將機器學習訓練部署於雲端，推理則在邊緣設備執行，以提升即時性與效率。AIoT 會議室包含兩大功能：（1）基於邊緣運算的環境控制與設備管理平台：採用工業級可編程邊緣控制器（EPIC），整合照明、空調及空氣品質監測。（2）具備機器學習能力的智慧物聯網環境：透過邊緣設備執行人臉辨識（FR）與情緒辨識（FER），即時分析與會者身份與情緒，並調整環境。本研究透過「雲—邊—端」架構，建置具機器學習能力的 AIoT 會議室，證實邊緣運算能降低延遲並提升效能，雲端則負責模型訓練與數據分析。未來可探索設備橫向聯動或動態運算分流，優化雲端與邊緣資源配置。此架構亦可應用於企業辦公與智慧教室，提升智慧環境的適應性與價值。

In the context of actively deploying 5G and the widespread adoption of the Internet of Things (IoT) in smart environments, edge computing and machine learning have emerged as critical technologies for enhancing performance. This study investigates the application of the "cloud-edge-terminal" architecture in an Artificial Intelligence of Things (AIoT) conference room, evaluating its effectiveness in environmental control, equipment management, and human-machine interaction. While traditional cloud computing offers robust computational power, frequent data transmission often leads to latency and privacy risks. To address this, the study deploys machine learning training on the cloud, with inference executed on edge devices to improve real-time performance and efficiency. The AIoT conference room encompasses two core functions: （1）Edge Computing-Based Environmental Control and Equipment Management Platform: Utilizing an industrial-grade Edge Programmable Industrial Controller (EPIC), this platform integrates lighting, air conditioning, and air quality monitoring.（2）Smart IoT Environment with Machine Learning Capabilities: Edge devices perform real-time face recognition (FR) and facial expression recognition (FER) to analyze attendees’ identities and emotions, adjusting the environment accordingly. Through the "cloud-edge-terminal" architecture, this study establishes an AIoT conference room with machine learning capabilities, demonstrating that edge computing reduces latency and enhances performance, while the cloud handles model training and data analysis. Future research could explore horizontal device integration (e.g., real-time interaction between emotion recognition and environmental adjustments) or dynamic computation offloading to optimize resource allocation between the cloud and edge. This framework is also applicable to enterprise offices and smart classrooms, enhancing the adaptability and value of intelligent environments.