ResGER: Resnet-based Group Emotion Recognition for Crowd Evacuation

Emotion plays a vital role in crowd evacuation, e.g., the excessive panic emotion leads to pushing and stampede and thus causing heavy loss of life and property. Group emotion recognition can identify the emotions of the crowd and provide effective guidance for the emotional contagion of the crowd evacuation. However, the existing crowd emotional contagion models are mainly designed based on random initial conditions and specific hypothetical conditions, and cannot be applied to the actual crowd evacuation guidance. Deep learning based group emotion recognition is an efficient method which can capture real time features of crowd emotions from videos and apply them to the emotional contagion of the crowd evacuation. In this paper, we propose a resnet-based group emotion recognition (ResGER) method for crowd evacuation. This method is generally divided into three steps: The first step is to extract the expression features of each face obtained in the scene through the existing expression recognition network. The second step is to train a network that scores the importance of these characteristics. The third step is to fuse the feature vectors and output the classification results according to the results of the first two steps. Experimental results show that our method is effective and can provide great support for the practical emotional contagion simulation of crowd evacuation.