Human Activity Recognition System: Using Improved Crossbreed Features and Artificial Neural Network

Abstract

In this paper, we present an intelligent method of human action recognition based on hybrid features. These features are calculated from the space-time cubes (interest points). The Blocks or cubes are derived from the difference of consecutive frames. These features include average number of blocks per frame, velocity of the blocks i.e. average angle and displacement, some of the kinematic features like divergence and vorticity, and hu features derived from motion energy images (MEI). Principal Component Analysis (PCA) has been used to reduce the dimensionality of the feature vector. For classification, we employ artificial neural networks (ANNs), in which each action video is represented by a bag of features. [Asmatullah Chaudhry, Javed Ullah, M. Arfan Jaffar, Jin Young Kim, Tran Anh Tuan. Human Activity Recognition System: Using Improved Crossbreed Features and Artificial Neural Network. Life Sci J 2012;9(4):5351-5356] (ISSN:1097-8135). http://www.lifesciencesite.com. 795

6 Figures and Tables

Cite this paper

@inproceedings{Chaudhry2012HumanAR, title={Human Activity Recognition System: Using Improved Crossbreed Features and Artificial Neural Network}, author={Asmatullah Chaudhry and Javed Ullah and M. Arfan Jaffar and Jin Young Kim and Tran Anh Tuan and Tran Nguyen Bao Anh}, year={2012} }