A Comprehensive Survey and Evaluation of MediaPipe Face Mesh for Human Emotion Recognition

Human Emotion Recognition (HER) has become a crucial application within computer vision and artificial intelligence, facilitating advancements in human-computer interaction, robotics, and other domains. Among the myriad approaches, Facial Expression Recognition (FER) systems stand out as a prominent method for automated emotion detection. In this paper, we conduct an extensive survey of the diverse techniques, modalities, and datasets employed for emotion recognition, with primary emphasis on non-verbal cues, and facial expressions. We delve into the significance of facial landmark detection, a critical task in numerous applications such as face recognition and emotion detection. Using real-time data frames, we focus on leveraging the capabilities of the open-source MediaPipe Face Mesh, a real-time face geometry framework capable of estimating 468 3D face landmarks, to enhance the accuracy and efficiency of landmark detection. We evaluate its performance with two other open-source models namely DLIB and OpenPose and check their efficacy. From the results, it is observed that MediaPipe Face Mesh offers real-time performance and high accuracy, making it suitable for real-world applications with stringent latency requirements.