David Suter - Research Project

Image Motion Estimation

Motion is a primary visual cue for humans. Motion alerts our attention mechanisms (towards potentially interesting or dangerous moving objects). Motion also allows us to segment (break up) the work into meaningful parts (for example, consider a soldier hiding but who becomes visible when moving). Motion also allows us to recover 3-D shape from 2-D images (a primary way to illustrate 3-D data on a 2-D computer screen is to animate the data so that it rotates - the human viewer can then reconstruct the 3-D shape that produces the images).

We (mainly with Ali Bab-Hadiashar) have developed the most accurate optic flow (motion from images) technique known so far.

We are also using the motion to segment images - purely on the way the objects move - not uing any other visual cues.

The work also involves motion model selection and segmentation. See Konrad's Homepage

Lately, we have started looking at the possibility of using image motion estimation for landing robotic aircraft.

Click on the images to see movie files

Calculation of Optic Flow

Motion Segmentation