8-10 注目物体の検出に基づく一人称視点映像における自己行動識別(第8部門ヒューマンインフォメーション)

抄録

In this paper, we propose a classification method of human actions from first-person videos. Our proposed method detects visually salient objects and integrates saliency value and object likelihood into feature vectors. The experimental result shows that the proposed method can classify first-person actions more accurate than a previous method by 6.4 percentage points.