Real-time Facial Expression Recognition on Robot for Healthcare

Abstract

Facial expression is important indicator to human health status. This paper is devoted to improving the safety monitor and healthcare for the old through facial expression recognition (FER). A novel convolutional neural network (CNN) architecture which is able to accelerate training process was proposed to deal with FER problems. In order to improve the FER performance in real life, a new dataset was collected to ease the data imbalance of FER2013 dataset. Further more, the method of moving average is ultilized to make up for the drawbacks of still image-based approaches, which is efficient for smoothing the real-time FER results. To monitor the safety and health status of the old, a digital healthcare (DHc) framework was proposed for better healthcare. As a result, the proposed model achieved compariable performance to the state-of-the-art methods both on FER2013 and NVIE datasets. The robot, equipped with the DHc framework, can recognize facial expression in real-life with high performance, and achieved substantial improvement for safety monitor and healthcare of the old.