A Hybrid Speed and Radial Distance Feature Descriptor Using Optical Flow Approach in HAR

Abstract

Video surveillance is ubiquitous in the public area of the world now. Human activity recognition based on video surveillance automatically identifies and analyses the human activity of continuous frame or video sequence data from various sensors. Human activity such as walking, standing, running, sitting, jogging, interaction and so on is a good representation. The human activity contains such inter-class differences, intra-class similarities and so forth challenges to tackle. In this paper, recognizing human actions based on the Optical Flow approach in video sequences has been addressed. The descriptor with Optical Flow vector of boundaries of activity performance includes silhouette speed on several frames, different areas and the eight different angle’s radial distance. The proposed descriptor with a multi-class Support Vector Machine classifier has achieved good accuracy for each action of the Weizmann datasets.