dc.description.abstract | 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. | en_US |