A SURF-Based Spatio-Temporal Feature for Feature-Fusion-Based Action Recognition

In this paper, we propose a novel spatio-temporal feature which is useful for feature-fusion-based action recognition with Multiple Kernel Learning (MKL). The proposed spatio-temporal feature is based on moving SURF interest points grouped by Delaunay triangulation and on their motion over time. Since this local spatio-temporal feature has different characteristics from holistic appearance features and motion features, it can boost action recognition performance for both controlled videos such as the KTH dataset and uncontrolled videos such as Youtube datasets, by combining it with visual and motion features with MKL. In the experiments, we evaluate our method using KTH dataset, and Youtube dataset. As a result, we obtain 94.5% as a classification rate for in KTH dataset which is almost equivalent to state-of-art, and 80.4% for Youtube dataset which outperforms state-of-the-art greatly.