Human Action Detection and Recognition Using SIFT and SVM

Human action detection and recognition is the most trending research topic in applications like surveillance of videos, analysis of sports videos and many applications which involve human computer interaction. Many researchers are working on different algorithms to improve the accuracy of human detection. Identifying the actions of human from the given video is a challenging task. In the proposed paper combination of two different techniques is applied i.e. SVM and SIFT techniques are used to identify and recognize the human actions in a given video or image. To extract local features of the given video SIFT based technique is used. In this techniques initially we extract features based on the interest points at a particular point or frames, Mainly SIFT techniques involves 4 basic steps Scale-space extreme detection, Key-point localization, Orientation assignment and Key-point descriptor. Once the key features are extracted they are further classified using SVM classifier. In the results and discussion we perform the comparative analysis of these two techniques on a standard KTH dataset with running and hand clapping actions. The experimental results determine the overall accuracy of 82% for the actions: running and hand clapping actions.