Naïve Smart System for Real-time Human Emotion Recognition by AI Methods

Emotion recognition has emerged as a pivotal research domain at the intersection of computer vision, affective computing, and deep learning. The capacity to automatically perceive and interpret human emotions from facial expressions holds transformative potential across healthcare, human-computer interaction, education, automotive safety, and security surveillance. The system is designed to operate efficiently on standard computational hardware, achieving robust performance under varying lighting conditions, occlusions, and demographic diversity. The proposed architecture integrates a convolutional neural network (CNN) for facial feature extraction, a Haar Cascade-based or MediaPipe-driven facial detection module for precise face localisation from live video streams, and a softmax-equipped classification head that maps learned representations to seven discrete emotional categories: happiness, sadness, anger, fear, disgust, surprise, and neutral. The system is trained and evaluated on benchmark datasets including FER-2013, CK+, and AffectNet. A lightweight MobileNetV2-based backbone is employed to ensure real-time inference on consumer-grade hardware. The system achieves a classification accuracy of approximately 87.4% on the FER-2013 test set and processes video frames at 28–32 frames per second (FPS) on a standard GPU. The proposed system demonstrates that high-accuracy, computationally efficient, real-time emotion recognition is achievable using modern deep learning techniques, and the results confirm the viability of deploying such a system in practical, latency-sensitive applications.