Federated Learning for Real-Time Decentralized Smile Detection in Virtual Reality Environments

Facial expression detection is paramount in real-world applications, including patient monitoring, photo selection, and facial protection within the metaverse and virtual reality environments. Among various expressions, detecting smiles poses a significant challenge. Traditional methods relying on low-level face descriptors are being superseded by advancements leveraging Convolutional Neural Networks (CNNs) to enhance smile detection performance. In this research, we introduce and implement three CNN architectures tailored for smile detection from facial images. Our proposed model incorporates a fully convolutional neural network featuring two deep convolutional layers dedicated to smile detection. Additionally, pre-trained LeNet and DenseNet CNNs are employed for this task. The effectiveness of our approach is demonstrated through extensive evaluations of the Hromada dataset, encompassing images with and without smiles within the virtual reality environment. Moreover, we delve into detailed discussions on the practicality and efficacy of CNNs in facial expression detection, explicitly focusing on smile detection in real-time scenarios. Our experiments are conducted within a federated learning framework, enabling model inference and optimization for privacy preservation and scalability of the smile detection system in large-scale virtual reality applications. Through this decentralized approach, we address challenges related to real-time detection, model inference, and collaboration in the metaverse.