ENTERFACE’08 Multimodal high-level data integration Project 2 1.

eNTERFACE’08

Multimodal high-level data integration

Project 2

1

TeamOlga Vybornova (Université catholique de

Louvain, UCL-TELE, Belgium)Hildeberto Mendonça (Université

catholique de Louvain, UCL-TELE, Belgium)

Ao Shen (University of Birmingham, UK)

Daniel Neiberg (TMH/CTT, KTH Royal Institute of Technology, Sweden)

David Antonio Gomez Jauregui (TELECOM and Management SudParis, France)

Project objectivesto augment and improve the previous

work, look for new methods of data fusionto resolve the problem and implement

a/the technique distinguishing between the data from different modalities that should be fused and the data that should not be fused but analyzed separately

to explore and employ a context-aware cognitive architecture for decision-making purposes.

3

4

A set of variables describing states of the world (user’s input, an object, an event, behavior, etc.) represented in different media and through different information channels.

GOAL OF DATA FUSION: The result of the fusion (merging semantic content from multiple streams) should give an efficient joint interpretation of the multimodal behavior of the user(s) – to provide effective and advanced interaction

Background - Multimodality

Cognitive behavior is goal-oriented, it takes place in a rich, complex, detailed environment, so…

the system should: o acquire and process a large amount of

knowledge,o be flexible and be a function of the

environment, o be capable of learning from the

environment and experience.

Requirements

behavior = architecture + content

6

Types of contextDomain context (prior knowledge of the

domain, semantic frames with predefined action patterns, user s profiles, situation modelling, a priori developed and dynamically updated ontology defining subjects, objects, activities and relations between them for a particular person)

Video context (capturing the users’ actions in the observation scene)

Linguistic context (derived from natural language semantic analysis)

Example scenario

7

Knowledge-based approach

Restricted-domain ontology – structure and its instantiation

Pattern situations (semantic frames)User profile - a priori collected

information about users - preferences, social relationships information, etc. - and dynamically obtained data

8

Audio Stream Video Stream

SpeechRecognizer

Video Analyzer

Sound Waves

SyntacticAnalyzer

Recognized String

Sequence ofImages

SemanticAnalyzer

Syntactic Triple

KnowledgeBase

Fusion Mechanism

Human BehaviorAnalyzer

KnowledgeBase

Movements Coordinates

Movements Meanings

Advise PeopleLinguistic meanings

Tooling / Implementation

Speech recognition: Sphinx-4Syntactic Analysis: C&C Parser + semantic analyzer (

http://svn.ask.it.usyd.edu.au/trac/candc/wiki)Semantic Analysis

Ontology construction and instatiation: Protégé (http://protege.stanford.edu/)

Analysis: Soar (http://sitemaker.umich.edu/soar/home)

Video Analysis: Visual Hull (developed by Diego Ruiz, UCL-TELE)

Human Behavior Analysis: SoarOntology: Protégé

Fusion Mechanism: SoarIntegration: OpenInterface (www.openinterface.org)

Challenges

Unrestricted natural languageFree natural behavior within

home/office environment

11

Why do we need Soar ? CAPABILITIES:manages a full range of tasks expected of an intelligent

agent, from routine to extremely difficult, open-ended problems

represents and uses different forms of knowledge – procedural (rules), semantic (declarative, facts), episodic

employs various problem solving methods interacts with the outside worldintegrates reaction, deliberation, planning, meta-reasoning

dynamically switching between them has integrated learning (chunking, reinforcement learning,

episodic & semantic learning)is useful in cognitive modeling + taking into account

emotions, feeling and moodis easy to integrate with other systems and environments

(SML – Soar Markup Language – efficiently supports many languages) 12

Soar architecture

13

Procedural Semantic Episodic

Symbolic short-term memory

Reinforcement learning Chunking

Semantic learning

Episodic learning

Decision procedure

Appraisal detector

Clustering

Perception Visual imagery Action

Body

Project schedule

14

WP 1Pre-workshop preparation 

Task 1.1 - Identify and investigate the necessary multimodal components and systems to use during the workshop;

Task 1.2 - define the system architecture taking advantage of the previously accumulated experience and existing results;

Task 1.3 - describe precisely the scenario to work on it during the workshop

Task 1.4 - make the video sequences15

WP 2Integration of multimodal components

and systems (1st week) 

Task 2.1 - implement the architecture, putting all multimodal components and systems work together.

Task 2.2 - explore and select the most suitable method(s) for action-speech multimodal fusion.

Task 2.3 - investigate the fusion implementation

16

WP 3Multimodal fusion implementation (2nd

week) 

Task 3.1 - fusion algorithms implementation

Task 3.2 - fusion algorithms testing in a distributed computer environment.

17

WP 4Scenario implementation and reporting

(3rd and 4th weeks) 

Task 4.1 – integrate and test of all the components and systems on the OpenInterface platform

Task 4.2 - prepare a presentation and reports about the results

Task 4.3 - demonstrate the results

18