Dynamic emotion recognition using histogram of oriented gradients

Abstract

A dynamic emotion recognition model is being developed for the use in a closed loop feedback system to support the emotional development of children with autism spectrum disorder. A neural network model is designed to learn the patterns in the acquired data. The model takes as input the sequence of features extracted from the traditional feature extraction technique of histogram of oriented gradients. The model performance is assessed by the classification accuracies of the OULU-CASIA database validation set. Accuracies of up to 43.64% were achieved. This preliminary model marks the first step in developing a robust dynamic emotion recognition platform.