|
Numero |
3
|
Veckoblad om forskning, undervisning och |
29
jan. 1999
Årgång 29 |
![]() |
Doktorandkurs i Uppsala: "Topos Theory"På onsdag (3 feb.) börjar kursen med ovanstående namn. Mer information på webbadressen <http://www.math.uu.se/ ~palmgren/topos.html>. |
Kalendarium
Institutionskollegium...äger rum nästa måndag (8 feb.) kl 15.15 precis, i sal E3. På programmet står
Ingrid Exjobbsseminarium i datalogiTorsdagen den 4 februari 1999 kl 15.15 i rum 4523 Peter Gustafsson, D94, 20 poäng Per Nystedt, E94, 20 poäng Joachim Johansson, E94, 20 poäng Kerstin F.
|
![]() |
CAS-seminariumTorsdag 4 februari kl 15.30 i sal BB2, Lill-Janshuset (Fiskartorpsvägen 15). Incremental Speech Recognition Combining Statistical and Declarative Language ModelsDr. Gernot Fink, Applied Informatics School of Technical Sciences, University of Bielefelt Germany For the development of effective multi-modal man-machine interfaces the incremental generation of speech recognition results is an important prerequisite. On the one hand the system can instantaneously react to the user's intention even while the user is speaking. On the other hand the results of speech recognition can be used to assist the processing of other modalities, e.g. the interpretation of gestures. In this talk an incremental speech recognition system will be presented. The system is able to incrementally produce word hypotheses in real time. Thus, utterances of arbitrary length can be processed and direct interactions with other modalities are possible. In order to enhance the overall recognition quality most speech recognizers apply statistical language models. Frequently, however, the speech corpora available are not sufficiently large to estimate such models. Therefore the proposed recognizer can make use of expert knowledge about about the expected utterances given as context-free grammars. The additional restrictions are robustly combined with the statistical acoustic and language models. Experiments show that significant reductions of word error rate can be achieved by using both a statistical language model and a declarative grammar. Informationsdagar om EU:s femte ramprogram...äger rum 16-18 febr. någonstans på KTH - kontakta Astrid Borg, enheten för externa relationer. |
CVAP-seminariumTisdag 9 februari kl 10.00 i sal BB2, Lill-Janshuset (Fiskartorpsvägen 15). Video-Based Perception in Robotic Projects on Exploration, Telepresence, Micro-Manipulation and Hand-Eye-SystemsChristof Eberst, Technische Universitaet Muenchen Lab. for Process Control and Real-Time Systems (LPR). Chair: Prof. Dr.-Ing. G. Faerber. Robust and affordable perception is a key to autonomy in robotics.This talk will focus on vision systems of different robotic-projects at the LPR. The aspects of recognition, environmental modeling, and reliable obstacle detection in unknown indoor environments will be emphasized. These topics will be presented for the following projects. Micro-manipulation: A system will be presented that allows for micro-manipulation with standard industrial robots. Jitter and pose-errors of the manipulator are compensated by a piezo-actuated adapter. The position-control gets its input from accelerometers and a video-camera. Hand-Eye Systems: The goal of this project is to obtain a better understanding of the human hand-eye-systems that may help to promote the development of their technical counterparts. While the clinical part focuses on human grasping, the technical part aims at the synthesis of its robotic counterpart. Autonomous exploration: For video-based mobile robots that operate in unknown environments it is imperative to perceive the environment fast and robust. By the example of the mobile platform Marvin, a low cost robust exploration and navigation system will be presented, focusing on video-based environmental reconstruction, object-recognition, obstacle detection, and motion estimation. Telepresence: For run-time compensation of the visual channel in telerobotics a model-based scene prediction is enriched by mapping original textures to the models in order to achieve photorealism. The applied computer-graphics techniques improve the performance of the operator, its acceptance, and compensate for modeling errors. Since the teleoperator operates in unknown environments, the remote environment is explored and updated with video sensors. Reliable video-based exploration: This project aims to build up a highly reliable, vision-based exploration system. The main focus of this work is the automatic, general, and flexible integration of actuators, sensors, perception- and interpretation modules in a complex mobile robot system. Thereto a system-specification and system architecture that allows for efficient integration of existing and future modules is in development.
|