Video - Based Detection and Classi cation of Real - World

In this paper we propose a method for classifying objects moving in front of a video camera. Consecutive frames are grabbed from the video camera and the diierence image from consecutive frames is use to detect moving objects. Border tracing and tting of quadratic forms along the border of the objects are performed on the diierence image. Classiication is achieved by calculating an entropy value from the set of sequences of quadratic forms. The method is tested on domestic images such as people, curtains blown by wind and external events such as tree branches.