Exploiting inter-frame regional correlation for efficient action recognition

• We proposed a novel temporal feature extraction method for action recognition. The explores inter-frame correlation on the regional level. Our achieves state-of-the-art performance benchmark datasets. Temporal is an important issue in video-based Optical flow popular to extract feature, which produces excellent thanks its capacity of capturing pixel-level information between consecutive frames. However, such extracted at cost high computational complexity and large storage resource. In this paper, we propose method, Attentive Correlated Feature (ACTF), by exploring within certain region. ACTF exploits both bilinear linear correlations successive frames has advantage achieving comparable or better than optical flow-based methods while avoiding introduction flow. Experimental results demonstrate our competitive performances 96.3 % UCF101 76.3 HMDB51