Prof. Dr. Bernt Schiele

"Perceptual and Sensory Augmented Computing"

Perceptual computing is concerned with the processing of sensor data such as audio, video, and other sensory data. The growing interest and the increase in the amount of sensor data calls for efficient techniques to index, search, and structure those data. In the future, both the number of sensors as well as the range of applications may increase substantially since many types of sensors become very cheap such that computing and applications have access to and can be enhanced by ubiquitous sensors. Ubiquitous sensors and sensing have the potential to fundamentally change the way we think about human-computer interaction and how machines perceive the world. For example wearable sensors can perceive the human and the environment of the human from a first-person perspective. Or sensors attached to objects can perceive events and actions performed with those objects. I call this area of research 'sensory augmented computing'.

The talk will touch upon three areas of research. The first area is concerned with the robust recognition and categorization of objects in real world scenes using computer vision techniques. In the second area of content-based image retrieval we will discuss a novel image representation that addresses the semantic gap between the image understanding of humans and the low-level representations based on features. The third and last area then deals with the recognition of human activities using wearable sensing based on supervised as well as unsupervised learning techniques. In this context we explore the possibility to estimate interruptability of users from wearable sensors.