Human Action Recognition under Partial Occlusions

摘要

This paper addresses the problem of human activity recognition under partial occlusions and viewpoint changes. We tackle with these problems by means of a novel temporal template denoted as Histogram of Normalized Optical Flow (HoNOF) and a set of HMM-based classifiers. Experiments have been conducted on the IXMAS data set for training and the EPFL-IXMAS data set for testing, showing the good behavior of this approach.