The COST 211quat Analysis Model


The current focus of the COST 211quat project is the investigation of tools and algorithms for image sequence analysis in order to provide emerging multimedia applications with the means of availing of the functionalities offered by MPEG-4 and MPEG-7. The objective is to define an Analysis Model (AM). The AM is a framework for image analysis and includes a complete specification of the analysis tools used within the framework. It consists of both a textual description and a software simulator.

The textual description of the AM is updated during each COST 211quat Simulation Subgroup meeting. At meetings members may propose new tools or algorithms for inclusion in the AM. This proposal takes the form of a textual specification of the proposed tool/algorithm and a demonstration of the performance of the AM with this tool/algorithm incorporated. This demonstration consists of a set of experimental results obtained using the AM software simulator. Proposals are accepted/rejected based on evaluation of experimental results by the entire group. Members whose proposals are successful (and thus merit the right to have their image analysis tool/algorithm integrated into the AM) are obliged to provide a software implementation of their technique in order to update the software simulator.

The first version of the software simulation of the AM was collaboratively implemented by the members of COST 211quat. The simulator is implemented in the ANSI C programming language using a common software platform adopted by all members involved in the collaborative implementation. The platform is a robust stable development environment incorporating strict programming rules to ensure portability and ease of use. Furthermore, the software platform is compatible with that adopted by the ACTS-MoMuSys project, which was used in the development of the MoMuSys MPEG-4 Video Verification Model (VM).

Version 1.0 of the COST211ter Analysis Model was defined at the COST 211 meeting in Ankara, Turkey, in October 1996. This initial specification is now undergoing continual refinement and improvement. The most recent version of the AM is based on the work described in the following references:


Structure of the AM

The AM is a collection of image analysis tools which can be flexibly combined to achieve fully automatic segmentation and tracking of moving objects in a video sequence. Both scenes with static textured background and scenes where the background can be described by global motion parameters are considered. With minor modifications, the AM can also be used to perform semi-automatic object segmentation and tracking. In the future the AM will be extended to incorporate tools which will allow feature extraction and subsequent video content description. All of the above features (and more) of the AM are encapsulated in the following block diagram which presents a generic structure through which it is possible to achieve the required functionalities of automatic/supervised object detection and tracking, and feature extraction/classification. A more detailed description of this structure is contained in the following reference:

Illustrative Results:

Some illustrative results are presented below which indicate the performance of the AM. The results are in the form of image segmentation masks for various well known video test sequences. These results were generated using the software simulation of the AM. The AM algorithm was run over the entire test sequence in each case. The detected and tracked moving object at a particular time instant in each sequence is illustrated in the images below.

Moving object from the Akiyo test sequence

Moving object from the Mother and Daughter test sequence

Moving object from the Table Tennis test sequence

Moving object from the Coast-guard test sequence

Moving object from the Container test sequence

Moving object from the Hall Monitor test sequence