手勢作為一種直觀的交流和溝通的方式，在人機互動中一直是探索的焦點。然而，目前手勢控制的相關產品只允許使用者使用特定的手勢進行操作，使不同文化習慣或不同地域的使用者增加學習成本，損失原本手勢自然、直觀且低學習成本的優勢。 本論文利用深度學習影像辨識技術，創建一個自定義手勢系統，讓使用者能夠根據個人偏好設置手勢指令，避免固有產品的局限。此系統的靈活性可解決因文化和地區差異帶來的手勢理解和使用上的混淆。 透過MIAT方法論的應用，我們模組化地設計了手勢辨識、指令設置和人機互動等功能的整合系統。利用深度學習模型替代傳統影像處理，提高了在複雜環境下的準確性和適應性。除了靜態手勢的辨識外，我們還實現了動態的手勢組合與追蹤，使得使用者能夠更自由地定義手勢指令，創造出數十種適合自己的手勢指令，最後我們實現了含有這些功能的圖形介面系統應用程式。 ;Gestures have long been a focal point in human-computer interaction, serving as an intuitive form of communication. However, current gesture-controlled products restrict users to specific gestures, amplifying the learning curve for individuals from diverse cultural or geographical backgrounds. Consequently, this limitation undermines the inherent advantages of natural, intuitive, and easily learned gestures. This thesis capitalizes on deep learning image recognition technology to establish a customizable gesture system. This system enables users to configure gesture commands based on personal preferences, effectively bypassing the limitations of existing products. Its flexibility resolves confusion in gesture interpretation and usage arising from cultural and regional differences. Utilizing the MIAT methodology, we modularly designed an integrated system that encompasses gesture recognition, command configuration, and human-computer interaction functionalities. By replacing traditional image processing with deep learning models, we have significantly enhanced accuracy and adaptability, particularly in complex environments. In addition to recognizing static gestures, we have achieved dynamic gesture combinations and tracking, empowering users to freely define gesture commands and create numerous personalized gestures. Finally, we created a graphical interface system application that incorporates these functionalities.
