Special Issue Article
A Robust Method to Detect Human Actions by Fusing Hierarchically Filtered Motion with Stip Features
In recent years the human actions recognition in crowd video for security purpose in airport, highways or roads, metro line station, street etc. In this paper we are describing the different human actions in crowd video. In existing system, filtering parameters are sensitive in complex scenes and detected interest points are heavily affected by the cluttered background. So in order to handle cluttered background we propose a hierarchical filtered motion (HFM) with spatiotemporal interest point features to recognize more type of actions in crowd. The interest points are detected by 2-D Harris corner points detector and MHI Hierarchical motion filter is used to reduce the distracting motion and also find the characterize points. Finally the different human actions are classified by the GMM. The proposed method can detect three different human actions like boxing, hand waving, clapping. The proposed method is analyzed and validated by using KTH dataset and MSR action dataset II.