Detecting Facial Emotion Expression to Mentor and Manage Crowds Using a Deep Convolutional Neural Network

Human interactions are heavily reliant on facial expressions. An individual’s emotional state can affect their safety and security in crowd places. Thus, monitoring and detecting emotions in facial expression in real time contributes to preventing potentially harmful situations. However, performing such tasks is difficult and complex and requires advanced computational methods. This study aimed to use a deep convolution neural network to detect and monitor facial emotion expressions in crowds. In addition, the effectiveness of transfer learning using VGG16, ResNet50 and Xception with DCNN models to improve accuracy was investigated. To achieve this, an FER2013 dataset with more than 35,000 images classified in terms of anger, fear, disgust, happiness, surprise, sadness and neutral was used. The results showed that transfer learning improved the accuracy and performance of an ensemble of the three models.