Hand gesture recognition for user-defined textual inputs and gestures

Despite recent progress, hand gesture recognition, a highly regarded method of human computer interaction, still faces considerable challenges. In this paper, we address the problem of individual user style variation, which can significantly affect system performance. While previous work only supports the manual inclusion of customized hand gestures in the context of very specific application settings, here, an effective, adaptable graphical interface, supporting user-defined hand gestures is introduced. In our system, hand gestures are personalized by training a camera-based hand gesture recognition model for a particular user, using data just from that user. We employ a lightweight Multilayer Perceptron architecture based on contrastive learning, reducing the size of the data needed and the training timeframes compared to previous recognition models that require massive training datasets. Experimental results demonstrate rapid convergence and satisfactory accuracy of the recognition model, while a user study collects and analyses some initial user feedback on the system in deployment.