This paper presents a simple and competent approach for Human Activity Recognition based on the hypothesis that every action/activity has some kind of rotation information with translation. The average energy silhouette images give the structural and translation information integrated it with the rotational information. The shape information of the human activities is extracted using Edge distribution of Gradients and Directional pixels and orientation information obtained from –transform. The proposed method provides the advantage of merging the local and global features of the silhouette and thus provides the discriminative feature representation for human activity recognition. The combined information is classified by a multi-class SVM classifier. Experimental results on the two publically available datasets i.e. Weizmann and KTH show the superior performance and accuracy of our method.
展开▼
