Statistical modeling of video object's behavior for improved object tracking in visual surveillance

Abstract

This paper describes a post processing method for detected video objects to enhance the quality of detection. Starting with a basic set of video parameters (such as video frame and time, label of objects, the objects position in pixel, the width and height of object's bounding box) statistical parameters (such as arithmetic mean and standard deviation) about features are computed and with these parameters different statistical models are built. These models can be used to estimate the “normality” of an object's behavior.