|
Eventhough the aim of the study is to recognize the human beings in
the images from monocular camera without usual constraints,
initially the graph theory based methods for matching are analyzed
with new neighbour isomorphism. The robust Hausdorff method of
matching is extended to recognize the human beings with ample models
and modified distance measures. As the strategy to fuse different
algorithms to get better results despite occlusions is inherently
parallel, the parallel implementations on Cray T3E produced correct
results in appreciably shorter computation time. As industrial
applications in robotics, three example situations, one with known
environment, another with unknown environment and the third with
going around were discussed taking primitive models with human
beings as obstacles. Unless the problem of choosing a proper model
with the good indexing is available, the recognition of human beings
will continue to remain as one of the hard problems to be solved.
|