Campus Event Calendar
Event Entry
What and Who
The lifetime of an object – long-term monitoring of objects
Dima Damen
University of Bristol
Talk
Dima Damen is a lecturer (assistant professor) at the Department of Computer Science, University of Bristol. Her latest work is focused on egocentric video understanding.
https://www.cs.bris.ac.uk/~damen
AG 1, AG 2, AG 3, AG 4, AG 5, RG1, SWS, MMCI
MPI Audience
English
Note: We use this to send email in the morning.
Date, Time and Location
Friday, 23 October 2015
10:00
60 Minutes
E1 4
633 (sixth floor rotunda)
Saarbrücken
Abstract
As opposed to the traditional notion of actions and activities in
computer vision, where the motion (e.g. jumping) or the goal (e.g.
cooking) is the focus, this talk will argue for object-centred
understanding of actions and activities. The number of viable usages
of one object is several orders of magnitude smaller than the number
of activities in which the same object can be used. Automatic
understanding of the various `modes of interaction’ with objects in an
environment could be achieved by long-term monitoring of an object,
potentially throughout the object’s lifetime.
The talk will start with the bicycle theft problem as one of the early
works on long-term monitoring of objects.
The bulk of the talk will focus on object-level interactions from
first-person videos. The egeocentric view of the world offers a unique
perspective on daily object-level interactions. When monitoring
multiple people interacting with an environment over time,
task-relevant objects can be discovered. More importantly, when
harvesting interactions with the same object from multiple operators,
the object’s various statuses and modes of interaction can be
understood. This can lead to fully unsupervised object-based guidance
where knowledge on how objects can be used is provided to a novice
user. I will present an Android prototype of this unsupervised
approach that runs on Google Glass.
computer vision, where the motion (e.g. jumping) or the goal (e.g.
cooking) is the focus, this talk will argue for object-centred
understanding of actions and activities. The number of viable usages
of one object is several orders of magnitude smaller than the number
of activities in which the same object can be used. Automatic
understanding of the various `modes of interaction’ with objects in an
environment could be achieved by long-term monitoring of an object,
potentially throughout the object’s lifetime.
The talk will start with the bicycle theft problem as one of the early
works on long-term monitoring of objects.
The bulk of the talk will focus on object-level interactions from
first-person videos. The egeocentric view of the world offers a unique
perspective on daily object-level interactions. When monitoring
multiple people interacting with an environment over time,
task-relevant objects can be discovered. More importantly, when
harvesting interactions with the same object from multiple operators,
the object’s various statuses and modes of interaction can be
understood. This can lead to fully unsupervised object-based guidance
where knowledge on how objects can be used is provided to a novice
user. I will present an Android prototype of this unsupervised
approach that runs on Google Glass.
Contact
Rodrigo Benenson
--email hidden
passcode not visible
logged in users only
Rodrigo Benenson, 10/21/2015 08:35 -- Created document.