Emotion Recognition Methods In Intelligent Teaching

Currently, emotional recognition in intelligent teaching only works well under good conditions such as clear backgrounds with frontal faces. However, intelligent teaching is complicated and emotion missing issues significantly affect the learning efficiency. Therefore, it is important to solve the emotion missing issues to improve the intelligent education effectiveness. In this thesis, we investigate the facial expression recognition approaches in intelligent teaching. We proposed an improved Active Shape Model (ASM) to extract human facial features under different angles, and developed a pre-classification algorithm based on facial feature geometry characteristics, finally recognize human facial expression with support vector mechanism (SVM). This approach provides high effectiveness in solving emotion missing issues, which therefore shows great academic significance as well as industry prospective.The main contributions of this thesis are summarized as follows:1. Reviewed current modern remote teaching situation, analyzed the emotional missing problem in intelligent teaching. Based on the features and problems of learning emotions in internet teaching environment, this thesis proposed an method to automatically recognize human face expression with different angles.2. Built a human face model with different angel using the improved ASM method, which can be used in many other application fields such as intelligent detection, animation games and video retrieval.3. Proposed a pre-classification method for face expression recognition based on facial feature geometry characteristics and recognized the facial expressions with the SVM.4. Explored the SVM related core functions selection and parameters optimization. Based on image's geometry characteristics, conducted experiments to compare training time, parameter optimizing time, recognizing time and cross validation accuracy for frequently used linear core function, polynomial core function and radial core function. These conclusions can be extended to other data sets and characteristics.