The design of AVS has its roots in the modelling of the human visual system (HVS); an extremely challenging task that generations of researchers have attempted with limited success.

Vision is a very natural capability and it is commonly accepted that about 80% of what we perceive is vision-based. Vision’s highly intuitive nature makes it difficult for us to understand the myriad of problems associated with designing AVS, in contrast to sophisticated analytic tasks such as playing chess.Foto: Kristoffer Öfjäll

Thus AVS became a widely underestimated scientific problem, maybe one of the most underestimated problems of the past decades.

Many AI researchers believed that the real challenges were symbolic and analytic problems and visual perception was just a simple sub-problem, to be dealt with in a summer project, which obviously failed.

The truth is that computers are better than humans at playing chess, but even a small child has better generic vision capabilities than any artificial system.

My research aims at improving AVS capabilities substantially, driven by an HVS-inspired approach, as AVS are supposed to coexist with – and therefore predict actions of – humans.


Selected Publications

Highly Accurate Attitude Estimation via Horizon Detection

Authors: Bertil Grelsson, Michael Felsberg, Folke Isaksson

Publ 2016
Type Article in journal
To publication

Unbiased decoding of biologically motivated visual feature descriptors

Authors: Michael Felsberg, Kristoffer Öfjäll, Reiner Lenz

Publ 2015
Type Article in journal
To publication

Learning Spatially Regularized Correlation Filters for Visual Tracking

Authors: Martin Danelljan, Gustav Häger, Fahad Shahbaz Khan, Michael Felsberg

Publ 2015
Type Conference paper
To publication

More Publications



Staff at CVL

About the Division

About the Department