Investigation on Generative Autoencoders for Hand Gesture Recognition
N. Bharanidharan, N. Mugunthan, B. Nitish, V. Puneeth, Kumar V. Vinoth · Advances in computational intelligence and robotics book series · 2025
As a natural interaction method, hand gesture recognition has gained increased attention, particularly for human-computer interfaces. The goal of this research work is to develop a hand gesture detection system using machine learning techniques. Rather than making direct gesture predictions, this system compares the accuracy of several models in identifying hand gestures from dataset, thereby enabling real-time gesture-based interactions. To achieve this, we employed a combination of machine learning methods, neural networks, PCA for feature reduction and data pre-processing techniques. Algorithms like K-Nearest Neighbors, Decision Trees, SVM, and Autoencoders are used for classification. The dataset used in this work is the LeapGestRecog collection, which has 20,000 grayscale images of ten distinct hand gestures. Our best-performing model, SVM with an RBF kernel and auto-encoder, attained a remarkable accuracy of 99.98% on the test dataset, indicating its robustness in classifying hand gestures accurately.