Video-based Skeletal Feature Extraction for Hand Gesture Recognition

Hand gesture recognition is a hot topic and a central key for different types of application. As applications of computers and intelligent systems are growing in our daily life, facilitating natural human computer interaction becomes more important. In this paper, we focus on video-based approach on hand gesture recognition integrated with 3-D hand skeletal features to construct the raw video sequences, retaining the key video frames, extracting spatial temporal data and feeding them into a Support Vector Machine model for 2-D hand sign classification. Our novel method integrates hand skeletal descriptor into video sequence to retain the spatial temporal information which will be extracted as vectors for classification task. As oppose to conventional method of requiring a well placed pair of cameras or depth detection hardware, our method only require only one camera. The proposed approach outperforms state-of-the-art static hand gesture recognition methods, achieving almost 100% accuracy among 24 classes.