Motion Estimation in Color Image Sequences


To develop robust and efficient algorithms applicable for robotic and vehicle guidance where the task is to detect and track moving objects in the scene.




Feature-based matching-technique using color regions, resulting from a prior color segmentation (e.g. using the CSC), as the matching primitives.

Takes into account the unavoidable variations in the segmentation by the extension of the matching model to multi matches.

Combines matching based an region attributes (feature-based matching) with matching on pixel level (iconic color region matching) in order to provide extended trajectories.




The system was applied to a motion segmentation task in vehicle guidance. Experiments on more than 1000 natural outdoor images, taken from a moving car, show promising results. About 89% of all regions appearing in these images were successfully matched, these regions covered about 99% of the whole image area.

The average processing time per image (720x280 pixels) was 800 msec on a SUN ULTRA SPARC I (166 MHz).




For further information please consult our publications-page.


Example Movies


We have provided several example sequences in quicktime-format.
These are example sequences of our realtime motion analysis algorithm. A fast approximation of the convex hull of a motion segment is plotted in green. The trajectory of a motion segment is drawn in yellow. To better visualize the results, only one manually chosen motion segment is shown per image.

Click to download video.




Example Images

Click to enlarge.