Facial Expression Recognition Method of Educational Ro-Bot Based on Non-Overlap Region

Facial expression recognition is one of key factors of understanding the learning information about students for educational robot. In the process of facial expression recognition, recognition regions usually overlap, resulting in the recognition of adjacent regions where facial landmarks are located in the same region, which may not be able to convey accurate expression change information. In order to solve this problem, a facial expression classification method based on three-stage parallel-series region non-overlapping cascaded convolution neural network is proposed. The algorithm model of this method consists of three parts: segmented convolution neural network model, parallel-series shallow convolution neural network model and key point aggregation. Finally, the recognition accuracy of this method is compared with the commonly used 3DCNN and TCDCN cascade convolution algorithms in the database and the results show that the facial expression recognition accuracy of this method is significantly higher than that of the two cascade convolution neural networks.