Human action recognition in video

Abstract

In the world of automation every event needs some kind of auto response. Automatic responses can only be made when events are perceived automatically. With camera as a source of visual sensing, some intelligent system fitted with camera can make automatic visual perception possible for any event. Recognizing human activities for some automated response can be one challenge under this problem domain. In this paper, motion feature of a moving object is used for recognizing human action/activity. Histogram of Oriented Gradient (HOG) features with Support Vector Machine (SVM) classifier is used for classifying the human actions into 5 basic categories i.e. bending, boxing, handclapping, jogging and jumping. Pre-Processing involves Lucas-Kanade Algorithm to extract the human silhouette and Skeletonization operation to generate human skeletons. Skeletons are secondary features which are made input to SVM for activity classification. Experiments are conducted on KTH database and Weizmann database for accuracy calculation.