Cornelia Fermüller

Bio-inspired Machine Vision


I received an M.S.  from the University of Technology, Graz and a Ph.D. from the Technical University of Vienna, Austria in Applied Mathematics. I am Associate Research Scientist at the Center for Automation Research at the Institute for Advanced Computer Studies at UMD.  I  cofounded the Autonomy Cognition and Robotics (ARC)  Lab and co-lead the Perception and Robotics Group at UMD. I am the PI of an NSF sponsored  Science of Learning Center Network for Neuromporphic Engineering and co-organize the Neuromorphic Engineering and Cognition Workshop.

My research is in the areas of Computer Vision, Robotics, and Human Vision, focusing on biological-inspired solutions for active vision systems. I have modeled perception  problems using tools from geometry, statistics and signal processing and developed software in the areas of multiple view geometry, motion, navigation, shape, texture, and action recognition. I have also combined computational modeling with psychophysical experiments to gain insights into human motion and low-level feature perception.   
My current work is on robot vision in the  following two areas: 
1) Integrating perception, action, and high-level reasoning to interpret human manipulation actions with the ultimate goal of advancing collaborative robots and creating robots that visually learn from humans. 
2) Motion processing for fast active robots (such as drones) using as input bio-inspired event-based sensors.



Manipulation Actions

By integrating cognitive processes with perception and action execution, we investigate ways of structuring representations of events at multiple time spans, that allow generalization of actions. The main application of this work is to create robots that visually learn from humans.

Read more
Event-based Vision

Dynamic vision sensors, because of their high temporal resolution, low latency, high dynamic range, and high compression, hold promises for autonomous Robotics. We study the advantages of this data for fundamental navigation processes of egomotion, segmentation. and image motion.

Read more
Mid-level Vision

Between low-level image processing and high-level reasoning are grouping mechanism that implement principles of Gestalt, such as closure, or symmetry. We have developed new 2D image-processing operators and 3D operators that implemet these principles for attention, segmentation and recognition.

Read more
Motion Analysis

Active vision system compute from video essential information about their environment's  spatio-temporal geometry for navigation. In a series of studies, I investigated the recovery 3D motion and scene structure from image motion and implementation in efficient algorithms.

Read more

Taking advantage of the mathematical properties of fractal geometry, we designed texture descriptors that are inavriant to changes in viewpoint, geometric deformation , and illumination, and  which encode with very low dimension the essential structure of textures. 

Read more
Optical Illusions

Optical illusions can provide insight into the mechanism of vision. Using geometry and statistics I have uncovered a number of principles explaining limitations in what we can recover from images, and these principles can explain different optical illusions and were used to create new ones.

Read more


Grant Project Pages 

Our work on Vision and Robotics has been sponsored by the following NSF grants in the Cyberphysical program.


Currently I serve for the folllowing journals and conferences


email —
phone — 301 405 1768
office — AV Williams Building 4459

ADDRESS (for shipping)
Cornelia Fermüller
Computer Vision Lab, UMIACS
4459, AV Williams Building
8223 Paint Branch Dr, College Park, MD 20742


My research is made possible by the generous support of the following organizations.