Human Action Recognition Using Kinect

Abstract

In this paper, a general approach for human action recognition is applied for classifying human movements into action classes. The propose method uses Kinect for capturing depth stream. The system performs preprocessing on depth information for reducing noisy pixels and getting depth information in appropriate format. The background subtraction method is used for extracting region of interest i.e. human. The system extracts contours of person. The Hu moments are extracted from contours of person for training action classifier. The Support Vector Machine (SVM) is used for classifying human activities.