MODEST: Multimedia objects descriptors extraction from surveillance tapes

• Research unit: EU RFP
• Project number: 98.0179

Key words (English)

Image analysis; vidéo surveillance; MPEG.4 coding; MPEG-7 indexing; FIPA (intelligence physical agents); segmentation; tracking

Alternative project number (English)

EU project number: AC304

Research programs (English)

EU-programme: 4. Frame Research Programme - 1.2 Communications technologies

Short description (English)

See abstract

Partners and International Organizations (English)

Coordinator: Université Catholique de Louvain (B)

Abstract (English)

MODEST (Multimedia Object Descriptors Extraction from Surveillance Tapes) defined and developed a framework for the analysis of video sequences aiming at high-level semantic scene interpretation. The approach was based on the segmentation, tracking and indexing of moving objects in video scenes by using Intelligent Physical Agents (IP A). The work was performed in the scope of the MPEG-4 and MPEG- 7 standards.
MODEST produced as result a generic, standard-based and scalable architecture. Moving object (MPEG- 7) descriptors are obtained through an original statistical-based segmentation determining a change detection masks. This step is followed by a 3-D characterization of the objects (3-D bounding boxes, speed, trajectories), based on the a-priori knowledge of the monitored situation (cars are rigid objects driving on a surface) and camera automatic calibration. The MPEG-7 descriptors are used by a distributed intelligent agents system able to model the scene. MODEST has developed a specific FIPA compliant platform, with specific Agent Communication Language and MPEG-7-based ontology, that we have called MONITORIX. MONITOR IX integrates two different worlds: a set of image processing algorithms (VK, Video Kernel) and a multi-agent system. The VK algorithms extract a feature-based description from the image and send it to the multi-agent system (MAS). The multi- agent system is made up by a collection of intelligent autonomous agents each of which performs specific tasks on the description received from the VK. One of the agents in the MAS system is called Proxy. The Proxy is the agent that represents the camera. It receives the description produced by the VK and delivers it to the agents that request it. From the point of view of an agent, the only access to the image is through the Proxy.ln the other end of the system, there is another agent called User Agent. The UserAgent from the point of view of all agents represents the user. The User Agent asks information to the other agents as required to satisfy the user desires and presents it through a generic graphical browser to the user.
The main obtained value of MODEST relies in an original approach to scene interpretation based on multi-camera sources. The proposed approach has led to a prototype, namely MONITORIX. This prototype integrates new signal processing modules for moving objects characterisation and a FIPA- based distributed artificial intelligence system for road scenes interpretation. The architecture of MONITORIX will be delivered as a free-ware on the World-Wide-Web. It opens new prospects in the processing of video signals for scene interpretation. Though a lot of work has been devoted to search in databases both in MPEG- 7 and FIP A, MONITORIX offers new prospects for using MEPG- 7 and FIPA for the processing of real-time and real-life video signals.
The end of the project has been devoted to dissemination of theresults. (1) The organisation of a workshop grouping researchers from thevideo signal processing community and people from the distributed artificial intelligence community ended the project. (2) The transfer of the results to new 1ST projects (ART .LIVE and SEFIRA) has been achieved. (3) The industrial partner of MODEST (Philips) has developed a specific navigation system based on MODEST descriptors, which is particularly well suited for retrieving scenes in recorded sport events. (4) Another major contribution of MODEST is the dataflow multi-agent system developed, which will serve as a basis for further interpretation systems based on multi-camera viewing of scenes.

References in databases (English)

Swiss Database: Euro-DB of the
State Secretariat for Education and Research
Hallwylstrasse 4
CH-3003 Berne, Switzerland
Tel. [phone]
Swiss Project-Number: 98.0179