Human Action Recognition Using Optical Flow Accumulated Local Histograms

摘要: This paper addresses the human action recognition task from optical flow. We develop a non-parametric motion model using only image region surrounding actor making action. For every two consecutive frames, local descriptor is calculated flow orientation histograms collected overlapping regions inside bounding box of actor. An built by weighting and aggregating estimated along temporal axis. obtain promising trade-off between complexity performance compared with state-of-the-art approaches. Experimental results show that proposed method equals or improves on approaches these databases.