A Facial Sentiment Recognition System based on Smart Edge Device

How to establish a facial sentiment recognition system based on Computer Vision is a hot research issue of affective computing but still remains an open problem. This paper first introduced the architecture of our facial sentiment recognition system and the control flow in detail. We then used a self-built dataset containing more than 6000 facial expression images to train the convolutional neural network model, which is the core classifier of our sentiment recognition system. The hardware in the study adopts Yahboom K210-Deve1oper-Kit which is a low-power, low-computing power, low-cost development board, in the meanwhile programs and our six-layer neural network model are installed on the development board. Experimental results show that on edge devices with limited computational resources, our system is able to recognise four facial expressions (i.e. smiling, neutral, angry and surprised) with over 80% accuracy.