Until recently a video sequence was considered
to be of only one type: either a video mode sequence or a film mode sequence.
However, more and more broadcasts consist of a mixture of both types. Existing
motion estimation methods cannot handle those hybrid sequences, and therefore
new algorithms are required. In this work, an object based motion estimation
algorithm for hybrid video sequences is designed. The desired output is
a segmentation of each image of the sequence into objects, with a description
of the motion and a local detection of the type of material for each object.
For this, an object based motion estimator, which segments the image into
regions with similar motion as well as possible, is developed first. Then,
the resulting algorithm is extended to support hybrid sequences, and the
performance of this algorithm is compared with a simple block based motion
estimator for hybrid sequences, i.e. an extension of the 3-D Recursive
Search block matching algorithm, which is designed in this work too.
The developed object based motion estimation
algorithm is well capable of segmenting the images of a video sequence
into regions, or objects, with similar motion. The block based motion estimator
for hybrid sequences performs better than expected, and the output of this
algorithm is well suited for the use in temporal upconversion methods.
The use of object segmentation in a hybrid motion estimation algorithm
was not able to improve those results.