Dynamic Adaptation in Deep Learning for Enhanced Hand Gesture Recognition

Abstract

The field of Human-Computer Interaction (HCI) is progressing quickly with the incorporation of gesture recognition, which requires advanced systems capable of comprehending intricate human movements. This study introduces a new Dynamic Adaptation Convolutional Neural Network (DACNN) that can adjust to different human hand shapes, orientations, and sizes. This allows for more accurate identification of hand gestures over a wide range of variations. The proposed model includes a thorough process of collecting and preparing data from the Sign Language MNIST dataset. This is followed by a strong data augmentation procedure that provides a wide variety of realistic variations. The architecture utilizes sophisticated convolutional layers to leverage the capabilities of deep learning to extract and synthesize essential gesture features. A rigorous training procedure, supplemented with a ReduceLROnPlateau callback, was used to assure the model's generalization and efficiency. The experimental findings provide remarkable results, showing a substantial accuracy of 99% in categorizing a wide range of hand movements. This study makes a significant contribution to the field of hand gesture recognition by introducing morphological operations, thus enriching input data quality and expanding the model's applicability in diverse HCI environments.