School bullying is widely recognized as a one of the most serious problems among teenagers, which affects teenagers both in mentality and physique. Because of school bullying, thousands of teenagers are afraid to go to school every day, so it is of gr...
School bullying is widely recognized as a one of the most serious problems among teenagers, which affects teenagers both in mentality and physique. Because of school bullying, thousands of teenagers are afraid to go to school every day, so it is of great importance to find a preventative measure to prevent or reduce the incidence of school bullying. This paper proposed a school bullying detection method based on action recognition and speech emotion recognition. This paper used movement sensor, to specific, accelerometer and gyroscope on Arduino nano 33 BLE Sense, to gather data for action recognition and used a micro recorder for recording audio samples. As for speech emotion recognition, this paper extracted pitch and its 5 statistical movements, intensity, Mel-Scale Frequency Cepstral Coefficients (MFCC) and its 1st and 2nd derivatives. As for action recognition, this paper extracted 9 features movement features, 5 from accelerometer and 4 from gyroscope. A KNN is trained to distinguish bullying action from normal action and a combination of random forest and KNN are trained to recognize audio emotions. Validation results show that the combined audio emotion recognition and action recognition to generate a combination result outperforms either of them only, achieving an accuracy of 91%. Finally, in order to deal with the situations, where bullying actions and bullying emotions do not occur at the same time, this paper proposed a new detection method based on action recognition and audio emotion recognition.