Real-Time facial emotion recognition techniques: An exploratory survey on requirements, challenges, and opportunities

Recognizing facial emotions in real time is absolutely paramount in various application domains such as healthcare, robotic automation, education, and so forth, for enhancing decisive solutions that balance high accuracy, low latency, resource efficiency, and privacy preservation. Practitioners’ or researchers are perplexed in streamlining machine learning models and edge-assisted deployment strategies for Facial Emotion Recognition (FER). This article surveys the advancements in real-time FER, focusing on optimal and scalable deployment strategies. It examines the requirements for employing advanced deep learning techniques to enhance feature extraction, reduce computational complexity, and improve responsiveness and efficiency. It categorizes deployment strategies to address varying needs, from local device implementations for simple tasks to distributed edge-cloud frameworks for scalable and complex applications. Additionally, it expresses the importance of evaluating performance metrics such as frames per second (FPS), latency, and power consumption while ensuring resilience to real-world conditions, including occlusions and pose variations. To our knowledge, this is the first survey to jointly address algorithm-level optimization and deployment level feasibility study for real-time FER that offers actionable guidance for practitioners and system designers.