DESCRIPTION OF MOVING OBJECTS BASED ON PARAMETERIZED REGION EXTRACTION.

A sequence of TV-frames is recorded from a real-world scene. In order to determine motion attributes of objects the segmentation of these (digitized) TV-frames has to provide a reliable symbolic description for each image. For this purpose each grayvalue frame is transformed into a vector field, the vector components describing small local lanes fitted to patches of the image. Regions are extracted from these fields by means of three parameterized filters; the region features reflect a considerable amount of region internal structure. By comparing their features a similarity value is computed between regions in (time-) adjacent frames.