Abstract
Hand gesture recognition is a crucial aspect of how we interact with computers, allowing machines to understand and respond to our movements in a natural way. This paper explores an advanced method for recognising hand gestures using a 2D convolutional neural network (2D CNN), a type of deep learning technology. We built and tested our model on a specialised dataset, evaluating its performance through several metrics, including loss function, accuracy, precision, recall, root mean squared error (RMSE), mean absolute error (MAE), and mean squared error (MSE). We compared two different data splitting ratios, 80:20 and 70:30, to determine which one performed better. Our approach effectively captures the complexities of hand gestures, thanks to the spatiotemporal features in the input data. To enhance the model's robustness against varying conditions, such as lighting, backgrounds, and hand positions, we also employed data augmentation techniques. Our findings indicate that the model achieved an impressive accuracy of 80% and a precision of 85% on the training data, along with solid results on the testing data, including an accuracy of 79%. Ultimately, the 80:20 data splitting ratio showed the best overall performance, highlighting its effectiveness for our gesture recognition task.