Real World Intelligence Technology

Theme and Laboratory Outline

Multi-modal Interface Ssytem for the CrossMediator
Multi-Modal Functions TRRC Laboratory
http://www.rwcp.or.jp/lab/mmtl/

Two kinds of interface components of the CrossMediator are presented. One is a gesture recognition system which uses a sequence of range image for both erasing background noise and distinguishing gestures of which categories are depending on depth features. The other one is an algorithm to self-organize a task model of user from word sequences using six basic rules and two types of background knowledge to adapt a new task to use a multi-modal interface system linked to the CrossMediator.

The CrossMediator
An Integrated multi-modal retrieval system
Information-Base Functions TRC Laboratory
http://www.rwcp.or.jp/lab/mmtl/

We have been developing a system called the CrossMediator which realizes a complex of retrieval systems which works among multi-media databases composing of still image, motion image, speech and text. The main body of the CrossMediator consists of retrieval engines and methods to obtain well-coded and well-organized databases. This paper shows two retrieval engines which have been developed for realizing realtime video and speech retrieval and a word set extraction from an unknown query still image to access the Internet.

Face and Facial Expression Recognition
Multi-Modal Functions KRDL
http://www.rwcp.or.jp/lab/mm-kdrl/face_recog/rwc_face.html

Face is the most important part of human body. Facial expression conveys a lot of KANSEI information, which is often very obvious to our instinct but we can seldom describe clearly. The application potential and the charm of the problem per se have attracted attention from computer vision researchers for many years. However, most of the methods proposed so far could not go beyond of the labs, due to the poor abilities to tackle real-world challenges in environment, pose and facial. KRDL lab thus took the challenge to develop robust technologies for face and facial expression recognition. This poster will present our novel and robust approach to the topic. Unlike conventional systems, our system is personalised. It recognises a person from cluttered background regardless of the lighting and facial changes, and then references a person-dependant expression library coded by HMM. We will show our system by interactive demonstrations on the spot.

Theme and Laboratory	Outline
Multi-modal Interface Ssytem for the CrossMediator Multi-Modal Functions TRRC Laboratory http://www.rwcp.or.jp/lab/mmtl/	Two kinds of interface components of the CrossMediator are presented. One is a gesture recognition system which uses a sequence of range image for both erasing background noise and distinguishing gestures of which categories are depending on depth features. The other one is an algorithm to self-organize a task model of user from word sequences using six basic rules and two types of background knowledge to adapt a new task to use a multi-modal interface system linked to the CrossMediator.
The CrossMediator An Integrated multi-modal retrieval system Information-Base Functions TRC Laboratory http://www.rwcp.or.jp/lab/mmtl/	We have been developing a system called the CrossMediator which realizes a complex of retrieval systems which works among multi-media databases composing of still image, motion image, speech and text. The main body of the CrossMediator consists of retrieval engines and methods to obtain well-coded and well-organized databases. This paper shows two retrieval engines which have been developed for realizing realtime video and speech retrieval and a word set extraction from an unknown query still image to access the Internet.
Face and Facial Expression Recognition Multi-Modal Functions KRDL http://www.rwcp.or.jp/lab/mm-kdrl/face_recog/rwc_face.html	Face is the most important part of human body. Facial expression conveys a lot of KANSEI information, which is often very obvious to our instinct but we can seldom describe clearly. The application potential and the charm of the problem per se have attracted attention from computer vision researchers for many years. However, most of the methods proposed so far could not go beyond of the labs, due to the poor abilities to tackle real-world challenges in environment, pose and facial. KRDL lab thus took the challenge to develop robust technologies for face and facial expression recognition. This poster will present our novel and robust approach to the topic. Unlike conventional systems, our system is personalised. It recognises a person from cluttered background regardless of the lighting and facial changes, and then references a person-dependant expression library coded by HMM. We will show our system by interactive demonstrations on the spot.