 |
Email: Homepage: http://wwwhome.ewi.utwente.nl/~brugmani
|
Titel: | Multi-modal BrainPaint |
| Instituut: |
Universiteit Twente | | Plaats: | Enschede |
| Land: | Nederland |
| Begindatum: | 09-02-2011 |
| Voltooid: |
Nee |
| Begeleider: | Femke Nijboer |
|
Onderwerpen: | |
|
| Beschrijving: |
The goal of this Master Thesis is to create a BrainPaint application in which users can draw/paint on a virtual canvas using only their heads. To achieve this, the application will use multiple modalities for the user inputs and the research will focus on how the users experience this usage of multiple modalities. For some of the inputs, Brain Computer Interfacing techniques will be used, for which the Emotiv BCI headset will be used. The aim is to make the application usable with a relatively low amount of training and furthermore, the application should be fun and interesting.
At the University of Tübingen, a “Brain Painting” application was already developed which is specifically aimed at people with ALS (Amyotrophic Lateral Sclerosis), to give them the possibility to express themselves creatively. This application only uses P300-detection to control the application. The ALS patients greatly enjoyed using the “Brain Painting” application and were able to use it with the same accuracy as healthy subjects.
The innovative part of this Master Thesis will be the combining of multiple modalities to control the BrainPaint application. Possible modalities include the usage of P300-detection, gyro-sensors to track head-movement, eye blink detection and alpha-band power.
The goal is to test multiple combinations of these modalities after which I will try to find an answer to the research question: “How do people experience the usage of a multi-modal BrainPaint-application?”.
|
Titel: | Visuele Overweg Monitoring |
| Instituut: |
Strukton | | Plaats: | Hengelo |
| Land: | |
| Begindatum: | 03-11-2009 |
| Voltooid: |
Ja |
| Begeleider: | Betsy van Dijk |
| Begeleider extern: | Roel Westenberg | |
Onderwerpen: | Human factors |
|
| Beschrijving: |
De Visuele Overweg Monitoring is een systeem dat in staat is met behulp van één of twee camera’s spoorwegovergangen te monitoren. Software is hierbij in staat met behulp van camerabeelden onder andere de hoeken van slagbomen te bepalen, verkeer te tellen, de snelheid van het verkeer te meten, de snelheid van de treinen te meten en de werking van de signaallampen te controleren. Analyse van de beelden wordt ter plekke gedaan waardoor de beelden niet verstuurd hoeven te worden naar de servers van Strukton Systems, maar slechts de resultaten van de analyse (hoeken van de slagbomen etc), wat in de kosten scheelt.
Het doel van deze stageopdracht is het ontwikkelen van een webinterface welke de door het systeem verzamelde informatie op een overzichtelijke manier aanbiedt aan de diverse (potentiële) gebruikers van het systeem. Dit kunnen oa. onderhoudsorganisaties, spoorwegbeheerders en overheden zijn. Hierbij zal gelet moeten worden op de informatiebehoeften van de verschillende gebruikersgroepen, hun ervaring met soortgelijke systemen en praktische beperkingen of mogelijkheden van de te gebruiken technieken.
|
|
| Beschrijving: |
Imagine that you have a visual scene like the one below (a picture from the so-called TUNA corpus), showing several objects with different properties, and you want to describe one of these objects to somebody else. Which properties do you pick to describe the object that you have in mind, and why? For humans this is an easy task, but not for machines. For years, computational linguists have been working on the problem of selecting appropriate object properties to be included in a description, but a completely satisfactory solution (producing descriptions that are similar to what humans say) still has not been found.
As a further stimulus for this type of research, in 2007 and 2008 the first two "NLG Challenges" on Referring Expression Generation were organized. In 2007 the Challenge was to develop algorithms that can select the same object properties (attributes) as those found in a corpus of human object descriptions (for example, {colour = blue, type = chair}); in 2008 the Challenge also included generation of a word string ("a blue chair").
Theune et al. (2007) and Krahmer et al. (2008), see below, participated in the Challenges using the Graph-based framework of Krahmer et al. (2003). The graph-based algorithm assigns costs to the properties included in a description, and aims at generating the 'cheapest' description. Using a few relatively simple, stochastic cost functions and orderings we performed quite well in the 2008 Challenge, but many improvements are still possible, for example by taking more domain knowledge into account.
The goal of this Capita Selecta is to improve the referring expression algorithm as used in Krahmer et al. 2008. The TUNA corpus of object descriptions, used in both Challenges, will be made available as a resource. Also, the generation and evaluation software will be made available. (All implemented in Java.) We hope to use the results to participate in the 2009 Referring Expression Challenge.
Literature to start with:
* E. Krahmer, S. van Erk, A. Verleg, Graph-based Generation of Referring Expressions, Computational Linguistics, 29(1):53-72, 2003.
* M. Theune, P. Touset, J. Viethen and E. Krahmer. Cost-based attribute selection for GRE (GRAPH-SC/GRAPH-FP). Proceedings of the Workshop on Using Corpora for NLG: Language Generation and Machine Translation (UCNLG+MT), 11 September 2007, Copenhagen, Denmark.
* E. Krahmer, M. Theune, J. Viethen and I. Hendrickx. GRAPH: the costs of redundancy in referring expressions. In the Proceedings of the 5th International Natural Language Generation Conference (INLG 2008), June 12-14, 2008 Salt Fork, Ohio, USA, pp. 227-229.
|
|
|
|
