Facial Expression Recognition and Positive Emotion Incentive System for Human-Robot Interaction

Facial Expression Recognition (FER) is a hot research topic currently, many efforts have been made on improving the recognition accuracy on certain datasets. Nevertheless, most of the existing works on FER are focused on verifying their algorithms on testing set, ignoring the practicability of their model in the real world. In this paper, more attention is addressed on improving the FER performance in the wild and the application of the FER model on robots. Firstly, a FER dataset is collected for training the model of facial expression recognition in the wild (FERW). Furthermore, a real-time positive emotion incentive system (PEIS) is developed for improving user experience of the robot. The proposed PEIS, which can recognize, record, analysis the emotion status of the users and give humanized feedback, consists of emotion recognition, emotion analysis and emotion feedback. Emotion recognition, the first as well as the most important part of this system, is realized by FERW based on deep learning and voting method. The PEIS is evaluated in two scenario, one is the accuracy of FERW in natural scene, and the other is the user experience of the robot employs the PEIS. Finally, experiments show that our FERW model can recognize facial expressions in real-life with an accuracy of 79%, which is practicable in the real world. Our robot XiaoBao, equipped with the PEIS, is able to enhance user experience.