Feature reduction for hand gesture classification: Sparse coding approach

Hand grasping patterns are highly complex and necessitate sophisticated hand kinematic models. To effectively investigate and study hand gestures in realistic and daily-life scenarios, it is crucial to reduce the dimensionality of hand kinematics. Many studies have proposed low-dimensional kinematic models using dimension reduction techniques, revealing that only a few dimensions of the kinematic model are significant for accurately recognizing hand gestures. In this paper, we propose a novel feature selection technique based on sparse coding to classify hand gestures, with a specific focus on grasping objects. Our technique outperforms Principal Component Analysis (PCA), which is a commonly used dimension reduction technique. By utilizing sparse coding, we are able to extract the most informative features from the kinematic data, resulting in a more precise and efficient classification of hand gestures. Our approach has significant potential for real-world applications in areas such as human-robot interaction, prosthetics, and virtual reality.