A YOLO-DRIVEN APPROACH TO DETECTING SUBTLE FACIAL EXPRESSIONS AND TEMPORAL PATTERNS IN PROFILE IMAGES

Abstract

Abstract:  Currently, facial recognition is getting ever more powerful in the development of artificial intelligence. Emotional recognition plays a very important part in interactive technology. Roughly one-third of the communication in interactive technology is verbal, whereas two-thirds are non-verbal. Facial expression recognition (FER) is a technique that detects facial expressions during interactions. It plays a crucial role in manifesting human emotions, expressing the inner feelings and state of mind or thoughts of a person. The paper here is a joint work of gender classification and age estimation in the identification of human emotions. Basic emotions include facial happiness, sadness, anger, fear, surprise, emotion, etc. Here, a facial recognition system has been developed based on You Look Only Once (YOLO) architecture coat version 2. The architecture of YOLO features a real-time goal tracking system; here, it has been used with the purpose of instant face detection and recognition. To enhance the accuracy, these images were taken with the help of an anchor box. The second architecture is the compression of the images, which is helpful in gender classification and age estimation. It can achieve very accurate detection of objects and high-level feature extraction that can help in getting the best performance concerning image classification and object detection. It can produce accurate outcomes as compared to other techniques because there are fewer hidden layers and cross-validation in the neural network.