List of Research Projects
With the increasing number of wearable devices used by people in their
everyday lives, there is an equally increasing number of applications that
aim to grab the user's attention by various notifications. Be it arriving
e-mails or telephone calls, upcoming meetings, changes in the stock market
or navigation directions, the list of notifications on a wearable computer
that can happen anywhere at any time in any situation is increasing.
Clearly, there is a need to carefully handle and manage this increasing
number of notifications in order to prevent wearable devices to become
highly annoying. Importantly, management of notifications should take into
account that the value of receiving a notification varies depending on the
user's context. In this project we use context information extracted from
a set of body worn sensors, namely acceleration, audio, and location, to mediate notifications to the user of a
wearable device.
Contact: Nicky Kern,
Bernt Schiele
|
Object recognition has reached a level where we can identify a large
number of previously seen and known objects. However, the more challenging
and important task of categorizing previously unseen objects remains
largely unsolved. Traditionally, contour and shape based methods are
regarded most adequate for handling the generalization requirements needed
for this task. Appearance based methods, on the other hand, have been
successful in object identification and detection scenarios. Today little
work is done to systematically compare existing methods and characterize
their relative capabilities for categorizing objects. In order to compare
different methods we present a new database
specifically tailored to the task of object categorization. It contains
high-resolution color images of 80 objects from 8 different
categories, for a total of 3280 images. It is used to analyze the
performance of several appearance and contour based
methods. The best categorization result is obtained by an appropriate
combination of different methods.
Contact: Bastian
Leibe ,
Bernt Schiele
In professional downhill skiing competitions the results obtained by
elite athletes are very close to each other: A few hundredths of a
second can make the difference towards winning a race. The most
important feedback for the athlete is the trainer. He gives
instructions to the athlete about the optimal performance of the
sequence of motions. However, the perception of trainer and athlete are
always different: The athletes performs the technique and thereby he
has a certain feeling of his movements, whereas the trainer observes
the athlete and analyzes the movements due to his own experience and
the common training doctrines. We found that wearable sensors could
offer a new way for better matching the trainer's and athlete's view by
providing new information beyond the human (visual) senses. Contact: Florian Michahelles,
Bernt Schiele
Avalanches are one of the major threats to life in high mountain
terrain. Once buried by an avalanche, survival chances dramatically
drop from 92% after 15 minutes to only 30% after 35 minutes mostly
due to the lack of oxygen. It is therefore extremely important to
rescue any victims as fast as possible in order to maximize survival
chances. Today's technology, so called electronic avalanche beacons,
only allow to localize buried victims. We propose a novel avalanche
rescue system enhanced with wearable sensors. Those sensors provide
information about the vital state of the buried victims such as
heart rate, respiration activity, and blood oxygen saturation as
well as the orientation of the victim.
The proposed system will
help to maximize survival chances of buried victims by empowering
the rescuers to concentrate on the most urgent victims first.
Contact: Florian Michahelles,
Bernt Schiele
Tennenhouse coined the term proactive computing where humans get out of the interaction loop and
may be serviced specically according to their needs and current situation. In this paper we propose
a framework for proactive guidance which aims to overcome limitations of today's printed instructions.
By attaching computing devices and multi- ple sensors onto di erent parts of the assembly the system can
recognize the actions of the user and determine the current state of the assembly. The system can suggest
the next most appropriate action at any point in time. In an experimental case study with the IKEA PAX
wardrobe we show the feasibility of the proposed approach. At the end important issues are discussed and
future directions are outlined.
Contact: Stavros Antifakos
,
Florian Michahelles
,
Bernt Schiele
Imagine you could create an audio-visual record of your entire life.
Surprisingly this would only require 500 TB of data (assuming 100
years, 24h, 10 MB a minute). With current improvements in storage
technology this will be available to the average user in the
foreseeable future.
However, the retrieval of such data is not trivial. Humans do not
retrieve information by date and time, but rather associate items of
information with each other. This project addresses this issue by
not only recording audio and video, but also contextual information,
such as the users activity and the flow of discussion in a meeting. It
thus allows to distinguish different phases of a meeting, such as
discussion, presentation, or breaks or to find specific comments by
particular meeting participants.
Contact: Nicky Kern
,
Bernt Schiele
All
object models have their specific strengths and weaknesses depending on
context and environment dynamics. Since no single object model is
robust and general enough to cover all possible environmental
conditions we propose to combine different typed of
models using mutual information. The ultimate goal of the approach is
to overcome the limitations of the individual models by the combination
of multiple models on-the-fly using information theoretic concepts.
Contact: Hannes Kruppa
, Bernt
Schiele
The integration of multiple features and sensor modalities promises to
increase robustness of tracking. In this project a selforganizing
sensor integration scheme has been implemented.
Contact:
Martin
Spengler , Bernt
Schiele
The main problem with content-based image retrieval is the so-called
semantic gap between the human and the digital way to describe images.
While users query an image database on a high semantic level using
concepts/keywords (e.g. house, tree, people), the computer relies
mainly on low-level features such as color or texture. With the goal to
enable the user to use higher level concepts in his/her database query
we propose the framework of vocabulary-based image retrieval. Here, a
set of image detectors extract image regions that contain certain
concepts (= vocabulary). The image detectors are obtained through
hierarchical clustering and learning methods.
Contact: Julia
Vogel , Bernt
Schiele
The next generation of computers might be literally wearable. Besides size
and power, one important challenge is how to interact with wearable
computers. A promising direction is to make wearable computers more
aware of the situation the user is in and to model the user's context.
Sensors, such as cameras, mounted to the user's glasses, can recognize
what the user is looking at and might model what the the user is doing.
Contact: Nicky Kern ,
Bernt Schiele
We
have proposed an object representation based on the statistics of local
neighborhood operators which is capable to recognize in the order of
100 objects at a rate of 10Hz. Different extensions of the system have
been proposed including an active object recognition sheme based on
mutual information and object classification based on visual classes.
Contact: Bernt Schiele
An interesting problem is to learn a model of language from natural spoken
and visual input. Practical applications include adaptive speech
interfaces for information browsing, assistive technologies, education,
and entertainment. This is a project which has been done in
collaboration with Deb
Roy at the MIT Media Lab. Have a look at the project webpage at MIT.
Contact: Bernt
Schiele
|
