YOLO-Based Real Time Face Detection and Expression Recognition

Abstract

Face detection and facial expression recognition are hot topics in machine learning area. We propose a system that YOLO-based real time face detection and facial expression recognition. This system achieved detecting faces and classifying expression in real time. The face detection module is based on YOLO algorithm, it achieved successful result in face detection. Where the expression recognition module is implemented by an alternative way based on Action Units (AU), an unusual way for facial expression classification. In this project, we proposed a method called Improved Self-defined Patches Support Vector Machine (Improved SP-SVM) to improve conventional hand-crafted features method on AU detection. Experiment shows this method get significant improvement even the detection result is better than some convolutional neural networks with simple architecture. Also, to classify facial expression through AUs, we combine YOLO-based face detection model and state of the art AU detection model, which can be used in wild images or webcam for AU detection and expression classification at real time.