Learning sparse representation for dynamic gesture recogniton

His Gesture recognition is an important task for gesture-based Human Computer Interaction. A novel gesture recognition model based on sparse representation is proposed in this paper. The model mainly consists of the following four stages: firstly, the spatial-temporal interest points are detected from the video sequences; secondly, a cuboid is founded around each spatial-temporal interest point and the 3D SIFT features are extracted based on the cuboids; thirdly, we encode local 3D SIFT features within the sparse coding framework. In so doing, each local 3D SIFT is transformed to a linear combination of a few atoms in a pre-trained dictionary. Finally, we employ an max pooling strategy to get the final representation of a video and we use multi-class linear SVM to accomplish the classification task. We test our model in the video dataset made by ourselves and get a good performance.