Reconhecimento de emoções em cenários competitivos de jogos usando Rede Neural Convolucional e Raspberry Pi

Abstract

Emotion recognition from facial expressions has gained prominence in competitive sports, where emotional monitoring can provide insights into players’ performance and mental state. This study aims to embed a convolutional neural network on a Raspberry Pi device to perform real-time emotion recognition in players’ facial expressions, offering insights into their emotional state without interfering with game performance. The research focuses on analyzing the performance of the approach, which considers the adaptation of a predefined model, trained with the FER-2013 dataset, and running on low-cost hardware. The work explores, as a case study, the feasibility of applying emotion recognition to a first-person shooter game. The results show that the emotion recognition system predominantly detected neutral expressions, with variations for happiness, anger, and sadness. Despite this, users reported feeling neutral or happy before and during the game rounds and happy and sad after the match - indicating that the system detected emotions not perceived by the players. Regarding the embedded approach, the application performed efficiently on the Raspberry Pi, performing real-time emotion processing and recognition despite the hardware processing limitations, demonstrating the feasibility of using intelligent models in low-cost devices.