Detecting and classifying visual objects is useful in many ways. The attention of visual systems can be drawn to interesting parts of a scene, robots can learn about their environment, dangerous objects can be detected by surveillance systems or information can be provided by augmented reality applications. While humans are able to distinguish 30.000 different objects modern object classification systems are only able to categorize a few hundreds without solving the detection task. For detection and classification the number of objects is even lower. Hence, object recognition is currently a very active and challenging field of research.
Grzeszick, R., Rothacker, L. and Fink, G. A., Bag-of-Features Representations using Spatial Visual Vocabularies for Object Classification, ICIP 2013
Nasse, F. and Fink, G. A., A Bottom-up Approach for Learning Visual Object Detection Models from Unreliable Sources, DAGM 2012
The task of detecting faces of humans in videos or still images and verifying their identity has obvious relevance in the fields of surveillance and automatic access control. Furthermore, the possibility to visually identify people opens interesting options in human-computer interaction, for example by offering personalized interfaces and services. While the problem can be considered more or less solved given controlled environments and cooperative users, face recognition in unconstrained scenarios remains a challenging research topic. While Face Recognition is not the focus of our research it is still part of several applications (e.g. person tracking and identification within a smart environment).
Stein, S. and Fink, G. A., A New Method for Combined Face Detection And Identification Using Interest Point Descriptors, FGR 2011
Nasse, F., Thurau, C. and Fink, G. A., Face Detection Using GPU-based Convolutional Neural Networks, CAIP 2009
With the ever growing functionality offered by computational devices, they became common in every part of our life. Consequently, these devices have to be operated and accessed by users that are untrained or not familiar with operating them. Their complexity often overstrains potential users instead of making their lives easier. In this context, conventional mechanical and graphical interfaces have reached the limits of their capabilities. Consequently, lots of effort has been made in realizing alternative man-machine interfaces that utilize natural modes of inter-human communication and, therefore, are more intuitive to use. Besides speech, the most important of those modes is gestures. We are working on user and view-independent efficient gesture recognition methods that enable a user to interact with the services of a smart environment in a natural and intuitive way.
Richarz, J., Fink, G. A. Visual recognition of 3D emblematic gestures in an HMM framework, Journal of Ambient Intelligence and Smart Environments, 2011
Richarz, J. and Fink, G. A. Feature Representations for the Recognition of 3D Emblematic Gestures, ICPR HBU 2010
Richarz, J., Plötz, T. and Fink, G. A. Real-time Detection and Interpretation of 3D Deictic Gestures for Interaction With an Intelligent Environment, ICPR 2008