Emotion Recognition from Human Gait Using Machine Learning Algorithms

The analysis of human gait has been widely used in the clinical field, e.g., for the early diagnosis of some diseases. On the other hand, it is possible to associate movement patterns during gait with several human behaviors, such as emotions. The main objective of this work is to generate models to classify three discrete emotions: happy, sad, and angry, considering the neutral state as an additional class. A set of features were extracted from the 3D position of the human skeleton during walking sessions. A descriptive analysis of the data was performed in order to select the best subsets of joints for recognizing the emotions. The models were built with the algorithms: kNN, Random Forest, and a meta-classifier (boosting). The best results were obtained with boosting with a mAP of 0.77 for balanced data, and 0.79 for unbalanced data. The results were promising when using methods based on shallow machine learning, a deep learning approach is currently being worked on.