No abstract

No abstract

The problems of dense stereo reconstruction and object class segmentation can both be formulated as Random Field labeling problems, in which every pixel in the image is assigned a label corresponding to either its disparity, or an object class such as road or building. While these two problems are mutually informative, no attempt has been made to jointly optimize their labelings. In this work we provide a flexible framework configured via cross-validation that unifies the two problems and demonstrate that, by resolving ambiguities, which would be present in real world data if the two problems were considered separately, joint optimization of the two problems substantially improves performance. To evaluate our method, we augment the Leu-L. Ladický ( )

The problems of dense stereo reconstruction and object class segmentation can both be formulated as Conditional Random Field based labelling problems, in which every pixel in the image is assigned a label corresponding to either its disparity, or an object class such as road or building. While these two problems are mutually informative, no attempt has been made to jointly optimise their labellings. In this work we provide a principled energy minimisation framework that unifies the two problems and demonstrate that, by resolving ambiguities in real world data, joint optimisation of the two problems substantially improves performance. To evaluate our method, we augment the street view Leuven data set, producing 70 hand labelled object class and disparity maps. We hope that the release of these annotations will stimulate further work in the challenging domain of street-view analysis.

I present here two results from investigations of a computational theory of how the human visual system may obtain information about the physical environment directly from optically sensed velocity fields (optical flow). Previous work has shown how optical flow arises when an observer moves through the static environment. The inverse problem is investigated here: how to recover intrinsic characteristics of the environment from optical flow patterns. The first result is a method for computing the local slant of surfaces relative to the moving observer. The second result is a method for detecting and discriminating among the five types of edges that form the boundaries of surfaces. The results have been formalised as mathematical equations, a physiological model, and a computer program. Properties and predictions resulting from these formal models are discussed with regard to experimental findings in psychophysics.