The objective of MODEST is to define and develop a framework for analysing video sequences in order to extract high-level semantic scene interpretation. This will be based on the segmentation, tracking and indexing of moving objects in video scenes and the index interpretation by the use of Intelligent Physical Agents (IPA). The work will be undertaken within the context of the MPEG-4, MPEG-7 and FIPA standards. The results will be demonstrated by means of a video surveillance application.
Cameras capture raw image sequences. A first step extracts objects from these sequences. A second step uses indexing to extract features from the objects. Finally, content analysis, interpretation and decisions are performed by the camera assistants, standing alone or working together with the application assistant.
The results of the first steps can help to encode the sequences in MPEG-4 format and send them to the user for verification purposes.
To take decisions from video scenes, MODEST uses the following hierarchical approach:
1. Object extraction
a. Physical objects extraction (and transmission) - link with MPEG-4
b. Decomposition of images into regions (low-level segmentation, tracking and layering) - link with COST 211quat
c. High-level grouping of regions into objects with the help of external knowledge
2. Content extraction - link with MPEG-7
a. Automatic extraction of primitive features
b. Semantic features inferred (e.g. through fuzzy reasoning)
3. Content analysis, interpretation and decisions - link with FIPA
a. Reasoning at the camera level, basic scene interpretation
b. Reasoning at the application level: decision, activation of alarms and user interface
Summary of Trial
A test bed will demonstrate the validity of the MODEST approach. The trial will take place in two steps.
-An initial off-line trial will demonstrate the results of the overall MODEST application. This will be based on indoor (building) surveillance, and outdoor (road) monitoring videotapes.
-A later on-line trial will use a camera network to demonstrate the real-time execution of a subset of the MODEST application excluding, for example, the MPEG-4 encoding.
Demonstration of a system for remote video surveillance, extension for Internet delivery and the emergence of reference tools for image analysis.
The demonstration of the system is expected to have an impact on standardisation bodies (MPEG, FIPA) and also on evolution towards integrated solutions for concrete applications.
Main contributions to the programme objectives:
Multi-camera intelligent surveillance system tested in a real road-surveillance application
Contribution to the programme
Demonstration of leading-edge image analysis and artificial intelligence systems - standardisation of agent technologies
The goal of MODEST is to define a global hierarchical framework for image analysis and scene interpretation. However, this goal will be limited by focusing on a very specific topic: surveillance applications. MODEST aims to set the following contributions and result exploitation:
-Demonstration of the object-oriented coding capabilities of MPEG-4,
-Development, possibly in collaboration with other ACTS projects and with COST 211quat, of a low-level Analysis Model for video sequences,
-Contributions to MPEG-7 on surveillance applications and exploitation of the MPEG-7 recommendations,
-Exploitation of Agent Technology (FIPA) for the autonomous and distributed reasoning system required for high-level scene interpretation and understanding requests from human operators.
-Contributions to the FIPA application field of Audio-Visual Entertainment and Broadcasting.