Presenter

Mr Lubos Omelina - ETRO, Vrije Universiteit Brussel [Email]

Abstract

As the smart home concept begins to take shape, alongside accessible biometric systems more broadly, an integrated and reliable identification may transform our home to a stream of personalized data. Modern sensors are able to provide posture information next to the color image for identifying and tracking multiple people in a home environment. Since the sensors are already deployed, new applications are emerging for instance in gaming in automation related to the hybrid broadcast broadband TV and related services or health, like exergames, serious games and tele-rehabilitation.
In this thesis we focus on human identification, in particular on face and iris recognition which offer the least intrusive identification, hence can be used within a smart home without discomfort. We contribute by proposing a new method (based on local binary patterns) which offers trade-off between recognition accuracy and performance, hence is suitable for in home use. We simplified the training procedure by (i) improving the accuracy of the method when only small number of samples (or even a single sample) is available, (ii) selecting appropriate samples (using a clustering algorithm) from camera stream to avoid redundancies in the training set. We complement face recognition with recognition based on the iris (which currently offers superior recognition rates). We improved iris coding stage the de-facto standard iris recognition method. In addition we have proven that iris can be recognized also from the color cameras of standard mobile devices.
We integrate the recognition methods in a novel concept of a platform designed particularly for serious games used in physical rehabilitation. This platform is unique middleware that decouples sensor devices, games and data analysis. Interconnection of different parts can be done through a novel interface that allows therapists to re-program the exercises rapidly without unnecessary knowledge of technical details. Part of the platform is also a proposed method for interaction detection (incorporating also biometric identification) between the patient and the therapist. This method help with the objective and automated assessment of the patient's state during the therapy.

Short CV

Master in Software Engineering, Slovak University of Technology Bratislava, 2009