In this talk I will give an overview of recent work in my group
ranging from computer vision and scene context modeling to activity
recognition using wearable sensors.
In the first part of my talk I will cover four computer vision topics.
First I describe a new approach for people tracking by detection to
allow tracking of people across long occlusions. Second I summarize
our work using probabilistic topics models and a dense object
representation that is applicable to various tasks such as
unsupervised object class discovery and object detection. Third I
describe a hierarchical conditional random field model for object
detection. The last topic in computer vision is scene context modeling
by a dynamic conditional random field model that enables to jointly
label scene and objects in videos.
The focus of the second part of the talk is context modeling and
activity recognition for wearable computing. First I will describe how
we applied probabilistic topic modeling to discover activity routines
in sensor recordings of daily life. And second I describe a new method
to enable gesture recognition in continuous sensor streams using a
wrist-worn inertial sensor.