An Efficient Approach for Video Action Classification Based on 3D Zernike Moments

摘要: Action recognition in video and still image is one of the most challenging research topics pattern computer vision. This paper proposes a new method for action classification based on 3D Zernike moments. These last ones aim to capturing both structural temporal information time varying sequence. The originality this approach consists represent actions sequences by three-dimension shape obtained from different silhouettes space-time volume. In fact, given segmented Then, are extracted images volumes moments computed video, volumes. Finally, least square version SVM (LSSVM) classifier with features used classify videos. To evaluate proposed approach, it was applied benchmark human dataset. experimentations evaluations show efficient results terms characterizations classification. Further more, presents several advantages such as simplicity respect silhouette movement progress guaranteed moment.