Customized Neural Network for Hand Gesture Recognition Design and Implementation

Abstract

This thesis presents a hand gesture recognition system that combines hand-landmark extrac tion with compact neural classification. The study focuses on building an efficient solution for real-time gesture understanding through a comparison of different landmark-based and image-based approaches. The findings show that a landmark-driven representation, when paired with an appropriate classifier, offers a strong balance between accuracy, efficiency, and practical deployability in interactive applications. The thesis also shows how this approach can be integrated into a modern application architecture for real-world hand gesture recognition use.