Hand Gesture Recognition for Virtual Reality: An LSTM-Based Approach for Multi-Class Classification

The integration of physical and virtual reality is gaining increasing relevance, especially in contexts where the interaction between users and digital environments is essential. Virtual reality (VR) enables the simulation of sensory experiences, opening up new opportunities in fields such as industrial training. The quality of simulations depends on how accurately the virtual world replicates the behavior of real objects and the tactile feedback provided to users. Haptic gloves, which deliver tactile sensations during interaction with virtual objects, are among the most advanced tools in this field. This work aims to develop an artificial neural network for hand gesture recognition, focusing on three specific gestures performed with the right hand: Swipe Sx (swipe left), Swipe Dx (swipe right), and Twist (90° clockwise rotation of the wrist). The core of the model is based on an LSTM (Long Short-Term Memory) layer, which allows capturing long-term temporal dependencies in sequential data. The problem addressed is a multi-class classification (with three classes), within the context of supervised learning, where the network is trained on pre-labeled data to learn how to classify new observations. This work explores the applicability of neural networks, specifically LSTMs, for handling complex patterns in time series.