School Violence Detection with CNN and LSTM

In Vietnam, the problem of school violence is now a growing concern. Accompanying the technology development, Artificial Intelligence is a good practice for detecting school violence. This paper proposes the work to examine whether ResNet-50 performs better than VGG19 in school violence detection. Moreover, this paper also presented an easy-to-manage school violence detection web application from real-time captures of CCTV connected to the hosting server. The research picked the transfer learning method for Keras’ pre-trained CNN (Convolutional Neural Network) models (ResNet-50 and VGG19) with LSTM (Long Short-Term Memory). Python was used to train and build the models and run the detection program. The web application was coded in JavaScript, HTML, and CSS. We used the model with the best performance and built the web application with it. Results suggested that the VGG19+ LSTM model had higher accuracy than the ResNet-50+ LSTM model in school violence detection. Finally, the web application had a friendly GUI and returned real-time detection results.