Computer Vision and Multimodal Computing

Object class recognition is one of the most fundamental problems in computer vision
and I will discuss several approaches we have been developing. One of these approaches
is based on a shape-based model for object class detection that enables explicit knowledge
transfer between object classes. Another important problem in computer vision is
people detection and tracking. In our work we have aimed to integrate detection and
tracking into a common framework that has been recently extended to articulated pose estimation.
In the second part of the talk I highlight three approaches to enable activity and
gesture recognition in realistic scenarios using wearable sensors: the first address the
unsupervised discovery of daily routines; the second introduces a novel body-model based
human activity recognition approach;  and the third discusses an approach for mobile
gesture recognition.