Motion Based Video Classification for SPRITE Generation

Abstract

In this paper we address the problem of video classification for sprite generation based on various features along with the global and local motion present in the video. Our feature set consists of features such as global (or camera) motion, cumulative global motion, local motion (motion of objects in the video), duration of the video, number of objects in motion, number of macro-blocks in motion and presence of objects at the borders of the image. These features are analyzed together to classify the video into one of the six pre-defined classes. The main focus of our approach is to analyze the number of frames that are processed in order to extract the feature set from the video. We perform experiments on a variety of videos by varying the number of frames being processed and analyze the outcome while calculating the accuracy of our approach.