The 1st COE Postdoctoral and Doctoral Researchers
Technical Presentation

Date: Thursday, April. 27, 2006
Time: 13:30 - 15:30
Place: L1 Lecture Room
Language: English (Oral Presentation), English/Japanese (Question)
Chairperson: Toshihiro Asai (Vision and Media Computing Lab. : D3),
Kazuo Hara (Computational Linguistics Lab. : D3)

Program (20 mins each: 15 mins presentation and 5 mins discussion)

  1. "Indoor user location system for ubiquitous computing "
    β“c @”V ( ‘œξ•ρˆ—Šwuΐ : COE“Α”C•Žθ )
    Muneyuki Sakata ( Image Processing Lab. : COE assistant professor )

    [Abstract]
    In ubiquitous computing environment, user's location is one of the most important information. GPS is most powerful tool for outdoors and it is widely spread, but indoor location systems have individual strengths and limitations. We has proposed ALTAIR (Automatic user location system using active IR-tag), user location system in indoors. ALTAIR uses IR-tag, wide-angle camera and wireless network, and it realizes high positioning accuracy and identification of user. In this presentation, I will introduce some related works about location systems and introduce our systems.

    Top
  2. "Trigger-Based Language Model Adaptation for Automatic Transcription of Panel Discussions "
    ƒJƒ‹ƒƒX ƒgƒƒ“ƒR[ƒ\ ƒAƒ‰ƒ‹ƒR[ƒ“ ( ‰Ήξ•ρˆ—Šwuΐ : COE“Α”C•Žθ )
    Carlos Troncoso Alarcon ( Speech and Acoustics Laboratory : COE assistant professor )

    [Abstract]
    We present a novel trigger-based language model adaptation method oriented to the transcription of meetings. In meetings, the topic is focused and consistent throughout the whole session, therefore keywords can be correlated over long distances. The trigger-based language model is designed to capture such long-distance dependencies, but it is typically constructed from a large corpus, which is usually too general to derive task-dependent trigger pairs. In the proposed method, we make use of the initial speech recognition results to extract task-dependent trigger pairs and to estimate their statistics. Moreover, we introduce a back-off scheme that also exploits the statistics estimated from a large corpus. The proposed model reduced the test-set perplexity considerably more than the typical trigger-based language model constructed from a large corpus, and achieved a remarkable perplexity reduction of 44% over the baseline when combined with an adapted trigram language model. In addition, a reduction in word error rate was obtained when using the proposed language model to rescore word graphs.

    Top
  3. "Modeling driversf cognitive information processing to construct a driving support system "
    ’|–{@‰λŒ› ( ƒVƒXƒeƒ€§ŒδEŠΗ—uΐ : COE“Α”C•Žθ )
    Masanori Takemoto ( Systems and Control Lab. : COE assistant professor )

    [Abstract]
    In order to construct a system supporting our driving, it is necessary to examine how human beings drive a car. Based on the results of experiments with a driving simulator which are conducted by the systems and control laboratory, we will make a model of driversf cognitive information processing in entering an intersection. In the model, we regard driversf emotion, consciousness of dangerous situations, past experiences in driving, and so on, as PSF (Performance Shaping Factor) which is essential thought in Human Factors and which disturbs human action. And this model includes PSF which effects on the flow of cognitive information processing of cognition, judgment, and action. A simulation based on the model of the cognitive information processing will clarify latent problems of driversf behavior which can cause serious troubles. In future, we will construct a driving support system to solve the latent problems using the ubiquitous computing system.

    Top
  4. "Modeling human behavior of fault detection and identification in plant operations "
    —« Šσ–’ (ƒVƒXƒeƒ€§ŒδEŠΗ—uΐ : D3)
    Xiwei Liu (Systems and Control Lab. : D3)

    [Abstract]
    In a chemical process system, a few of user interfaces supply intuitive visual information of process variables, alarm messages, trend graphs, etc., based on which a human operator detects abnormalities and identify failure causes. These user interfaces are safety-critical and influence the operator's performance of fault detection and identification (FDI). In order to construct an effective, efficient, and satisfactory user interface, it is important to evaluate the interface by investigation of human operator's FDI behaviors in an emergency. However, a human has good capabilities of adaptation and learning, and her/his performance is unstable. Building an operator model is a flexible approach to analyze the FDI performance for various user interface systems. In this research, we proposed an operator model as a virtual subject with knowledge bases and procedures for FDI. Applying the operator model to a boiler plant simulator, we obtained FDI tracks for a user interface system. Case studies show that this model-based approach can be used to evaluate user interface.

    Top
  5. "Networked Active Camera System for Monitoring Applications"
    ƒLƒAƒbƒg ƒAƒuƒfƒ‰ƒYƒBƒY (ƒƒ{ƒeƒBƒNƒXuΐ : D3)
    Khiat Abdelaziz (Robotics Lab. : D3)

    [Abstract]
    Nowadays, security issues are becoming a major concern in our society. For this reason, we propose a networked active camera system to improve monitoring applications in public spaces. The proposed system allows the capture of high resolution images of people's faces going through specific areas. The system is a combination of stereo and active cameras. In this presentation I will give details of the first implementation, which has been demonstrated during the last COE festival. I will also explain the internal process and talk about future developments.

    Top

21st Century COE Program
NAIST Graduate School of Information Science