An Expression Recognition Method on Robots Based on Mobilenet V2-SSD

This paper adopts a combination of MobileNet and SSD method to recognize facial expressions, and then apply it to Nao robots. MobileNet is a lightweight deep network model for mobile devices. The standard convolution is decomposed by depthwise separable convolution to decompose the calculation and reduce the amount of computation. The SSD model evolves from the VGG16 model and maintains excellent object detection performance despite a sharp decrease in the number of parameters. This paper combines MobileNet V2 with SSD for expression recognition, which not only meets the real-time requirements, but also keeps high recognition accuracy. Since the robot's processor performance is limited, while the deep neural network can automatically extract the image feature for accurate classification, it is of great significance to use lightweight deep neural network to apply the real-time detection and recognition of the Nao robot.