Dynamic scene analysis from the polar form of image flowrepresentations

摘要

A basic area of research in digital video processing is the frameto frame low level motion estimation of image sequences. The motioninformation obtained is usually displayed as a 2-D distribution ofdisplacement vectors, called the optical flow field or image flow. Howto use that information, in order to precisely estimate the location ofobjects in the scene, remains a relevant and very hard problem incomputer vision. In previous schemes, the discontinuities appearing inthe motion field have been studied in order to use them to discriminatebetween the various objects in the scene. In this work, the polar formof the image flow representation is used to enable a new approach to thehandling of motion field discontinuities in dynamic scene analysis