Abstract
<jats:p>In the era of rapid digital transformation and the growing prevalence of artificial intelligence, enabling natural, seamless, and contactless human-computer interaction has become a critical priority across various domains. This paper presents a novel deep learning-based model for virtual mouse control using hand gestures, termed CLVM (CNN-LSTM Virtual Mouse). The proposed system introduces a hybrid architecture that integrates three powerful components: (1) MediaPipe for efficient and real-time hand landmark detection; (2) a Convolutional Neural Network (CNN) for spatial feature extraction; and (3) a Long Short-Term Memory (LSTM) network for temporal dynamics modeling, enhancing the system’s ability to recognize gestures continuously and accurately over time. Unlike traditional models, CLVM is designed to maintain robust performance in real-world environments, particularly under conditions of inconsistent lighting and cluttered backgrounds. The system also provides low latency and high responsiveness and can be deployed effectively on resource-constrained devices, making it practical for widespread adoption. Experimental results demonstrate that CLVM achieves a high accuracy (99.88%) while reducing the loss to 0.38, significantly outperforming conventional gesture recognition methods. These findings highlight CLVM’s potential to serve as a reliable, scalable, and efficient solution for natural gesture-based interaction. It offers a valuable step forward in the development of intelligent, user-friendly interfaces for contactless control applications.</jats:p>