Efficient Emotion Recognition based on Hybrid Emotion Recognition Neural Network

The practical application of computer vision on robots such as emotion, age, gender recognition can improve the interactive experience between robots and users. This paper uses a webcam to capture the image as a visual system input. Then, facial image is obtained through high-performance face detect neural network. Facial landmarks is used to correct the face. After that, we input facial image into the multi-person emotion recognition system. In order to improve the accuracy of emotion recognition, a hybrid emotion recognition is proposed based on Convolutional Neural Network. Taking facial points and facial image as input, training hybrid neural network to convergence and outputting five home common emotion, neutral, happy, surprise, sad and angry. The other hand, the Microsoft Azure API is used for age and gender recognition. Finally, the experimental result shows that the accuracy of emotion recognition is as high as 86.14%. In practical applications, the system can recognize the emotions, age and gender up to thousands of people at the same time.