Video Object Tracking using foreground models

Abstract

This Master Thesis present an approach to Video Object Tracking segmentation using foreground models. For the video sequences analysed, the foreground and the background have been modelled using Spatial Colour Gaussian Mixture Models (SCGMMs). SCGMMs are Gaussian Models which describes the foreground and the background using five components in colour and spatial domains. In order to have a better result in the segmentation process, the Gaussian Models computed for each frame are passed to the next frame using a tacking technique that helps in the individuation of the object in foreground alone the sequence. Using the location provided by the tracking, the Gaussian Mixture Model for the background is computed only in the close region around the object in foreground allowing in this way a better modelling of the region. The Thesis is structure as follows: after a presentation of the study of the State of the Art where the techniques for tracking and segmentation are presented, there is the presentation of the method proposed. At the end there is a Chapter that describes the results obtained and some conclusions and a Chapter which presents some future developments.