Feature Extraction from Biological Motion of Human Gait Patterns for Emotion Discrimination

We study a method of a feature extraction to discriminate emotions of human from a sensing data of human gait patterns as ”Biological motion data”. We assume that the high–dimensional biological motion data are generated by low–dimensional features whose components are statistically independent. So we use a method of independent component analysis to extract the features. The extracted feature is evaluated by a discriminated result of the given biological motion data which identified five types of categories, ”Anger”, ”Grief”, ”Disgust”, ”Joy” and ”Fear”. We achieve 40% accuracy for 5–classes of emotion discrimination with 3 actors’ biological motion data.

• Type: Journal
• Year: 2007
• Where: MVA
• Authors: Hidenori Maruta, Masahiro Ishii
• Keywords: Biological Motion Data, Computer Vision, Human Gait Patterns, Independent Component Analysis, MVA 2007