Development of a GPU-Based Human Emotion Recognition Robot Eye for Service Robot by Using Convolutional Neural Network

Abstract

Service robots can be used widely to assist elderly and disable population due to the lack of caregivers in future. Real-time human tracking, detection, focusing and implementing various algorithms are a wide range of application in emotion recognition service robots. Therefore service robots must have a properly designed robot eye model to be human-friendly with accurate human-robot interaction. Developed robot eye can be recognized the human emotional states by using well trained deep convolutional neural networks (ConvNet). This paper describes graphics processing units (GPUs) based human emotion recognition robot eye by using ConvNet. Mainly, the robot eye performs two processes in the intelligent systems. They are the robot eye focus to the human face and head by using pre-trained haar cascade classifier and recognizes the human emotional states probability with percentages as happy, sad or relaxes by using pre-trained ConvNet. The developed robot eye was implemented and tested by using different people successfully and the results of them are presented. According to the results, the emotions are detected more than 85% of overall accuracy for each person.