Robust Approach for Emotion Classification Using Gait

The ability to gauge emotion via their gait is a key component in enabling machines to understand people in images, videos and in real-time. Emotion recognition using machinery has been done through various methods. Facial Expressions; necessitates an obstructed view of the face, Psychological Parameters; require constant and prolonged monitoring of the subject, and Sentiment Analysis; demands text be written by that individual. However, in real-time situations, these features are not readily available which makes it difficult to classify emotion. Thus, this paper proposes a technique for classification of emotions without having to compromise on the accuracy of the detection while at the same time utilising the most trivial of information; walking. A University of York dataset of videos segregated emotion-wise was chosen to perform pose estimation through which important features were extracted. These features were processed in multiple machine learning classification models like; Decision Tree, Naive Bayes, SVM, K-Nearest Neighbours and Artificial Neural Network, to classify into five emotions. It was observed that this approach demonstrates a higher accuracy than most theories proposed in the past.