Enhanced Hand Gesture Recognition using Optimized Preprocessing and VGG16-Based Deep Learning Model

This research paper highlights advancements in hand gesture recognition, particularly addressing challenges related to adaptability to dynamic gestures. A meticulous approach is adopted, focusing on comprehensive preprocessing tasks such as image resizing, numerical array conversion, and pixel normalization. Leveraging the pre-trained VGG16 model from ImageNet via transfer learning enhances the model's pattern recognition capabilities. An attention mechanism is introduced post-VGG16 to further augment the model's ability to capture intricate details, emphasizing crucial features for classification. The sequential model built upon the VGG16 base, enriched with custom layers like flattening, ReLU activation, dropout, and SoftMax activation, is central to our methodology. The novelty of our research lies in the integration of innovative preprocessing techniques, model architecture, and attention mechanisms, ensuring robustness in recognizing intricate hand gestures. Evaluation of the Hand Gesture Recognition Image (HaGRID) dataset yields an accuracy of 85.13%, validating the efficacy of our approach in real-world scenarios. This research significantly contributes to the field by offering a refined model tailored for accurate hand gesture classification, promising advancements in human-computer interaction and related domains.