Study on the Features of Public Space for Children's Emotional Healing Based on Facial Emotion Recognition Technology

Abstract

This paper uses facial emotion recognition and machine learning techniques to explore the influence of spatial features of children's activity areas on children's emotions. Through the children's evaluation experiment on the emotional quality of public spaces, we obtained the facial expression pictures and the questionnaire data of the emotional self-assessment scale (SAM) when children used the public space. Then use MEGVII facial emotion recognition API to recognize children's facial expressions, and form a facial emotion feature variable dataset. Then, a public space emotional quality model was established using decision tree (DT), neural network (NN), and random forest (RF) classifiers. Finally, the model's performance was evaluated through the confusion matrix, and five groups of data reserved in the original dataset were used for external validation of the model. The results show that: 1) In the binary classification model, the classification accuracy of the emotional valence of the NN classifier was 94.44%, and the classification accuracy of arousal was 84.62%. These two results outperform the models built by the DT and RF classifiers. In the three-class model, the classification accuracy of the emotional valence of the DF classifier is 71.88%, and the accuracy of arousal is 84.62%. These two results outperformed the models built by NN and RF classifiers. 2) The analysis results of the correlation between spatial features and emotional quality showed that the width of the bordering tree, width-to-height-ratio of the bordering tree (w2/h2), the number of layers of the border, the continuity of the spatial border, the proportion of facilities, the color of the space, the number of types and pavement color types were usually correlated with the spatial emotional valence.