A Real-Time Hand Gesture Recognition System Using Image Processing

This study presents a systematic approach to hand gesture recognition in images, focusing on extracting meaningful gestures for communication purposes. The three main phases of the workflow are recognition, training, and labelling. During the labelling stage, hands are separated from photos with different skin areas by using MATLAB's Image Labeller to interactively generate region-of-interest (ROI) labels. By supporting both labelled and unlabelled data, this tool increases the diversity of datasets. The next step, called training, is taking the gathered dataset and extracting fully linked and classification layers for every move that has been described. The Stochastic Gradient Descent (SGD) approach is then used to train the model, optimizing with an initial learning rate of 0.01 and 20 epochs and 64 mini-batch sizes. This research contributes to a practical hand gesture recognition system with potential applications in communication for individuals with diverse needs.