本研究旨在開發一套結合即時視覺辨識與二軸控制系統的手勢追蹤平台，利用 YOLOv8 模型進行手掌目標辨識，並整合 OpenCV 處理流程與 WeMos D1R3 控制模組，驅動兩顆 SG90 伺服馬達以實現紅外線雷射同步標記。透過自建資料集與多場景擴增機制，各模組的應用原理與整合邏輯，從影像辨識、運動控制到資料標記皆經實作與評估驗證，本系統在遮蔽、偏光與快速移動等極端條件下仍保持 mAP@0.5 達 91.7% 的穩定辨識能力，同時引入 MPA（Mean Perceived Accuracy）作為應用導向指標，量化控制誤差與人眼容忍度之間的關聯。 實驗結果顯示系統具備良好的追蹤準確性與即時性，本研究在智慧控制與人機互動上的應用潛力，未來可應用於人機互動並延伸應用至協作型機器手臂、醫療輔助、智慧場域控制、擴增實境間的交互與教育導引……等領域，具高度擴展潛力。 ;This study presents the development of a gesture-tracking platform that integrates real-time visual recognition with a dual-axis control system. The platform leverages the YOLOv8 deep learning model for palm detection, combined with an OpenCV-based processing pipeline and a WeMos D1R3 control module to drive two SG90 motors, enabling synchronized infrared laser marking.A custom dataset and multi-scenario augmentation strategy were employed to enhance robustness across varying environmental conditions. The system architecture—from image recognition and motion control to data labeling—was implemented and validated through practical experiments. Under challenging scenarios such as occlusion, glare, and rapid movement, the system maintained a stable detection performance with an mAP@0.5 of 91.7%. Furthermore, Mean Perceived Accuracy (MPA) was introduced as an application-oriented metric to quantify the correlation between control error and human perceptual tolerance. Experimental results demonstrate high tracking accuracy and real-time responsiveness, indicating the system’s strong potential in smart control and human-computer interaction. Future applications include collaborative robotic arms, medical assistance, intelligent space control, augmented reality interfaces, and educational guidance systems, reflecting its high scalability and adaptability.
