Automatic Recognition of Affective Laughter from Body Movements

Laughter is often associated with happiness but recent studies show that there are actually five types of Filipino laughter and these are happiness, giddiness, excitement, embarrassment, and hurtful laughter. Facial expressions and vocalization related to Filipino laughter have been the focus in many studies, however body movements with regards to laughter are still left unexplored. The main focus of this research would be determining what body movements are associated to specific types of laughter. This research will make use of the Microsoft Kinect for capturing the different body points of the participants that will result to the low-level features. High-level feature extraction will be applied in order to translate the points gathered through the Microsoft Kinect into a more understandable data. This research aims to classify these types of laughter with the use of one modality, which is the body movement. The researchers will determine which specific body parts are important to track in order to determine the type of laughter through feature selection. With the use of different machine learning techniques, this research aims to build a model that would classify which type of laughter is being performed. The machine learning algorithms used are Bayesian Networks, k-Nearest Neighbor for k=3 and k=5, and J48 decision tree. Based on preliminary results, J48 is the recurring algorithm that holds the best accuracy results, although the kappa is still low for all experiments.