Introduction - If you have any usage issues, please Google them yourself
This paper proposes a powerful and robust local descriptor,called textureorientation
descriptor(TOD). TOD consists of two components: texture and
orientation. Considering that the human shape shows more short, fragmentary
and vertical edges, the proposed TOD descriptor encodes the spatial information
of pedestrian image. We combine the proposed TOD representation
model with mean shift framework to perform pedestrian tracking. Results
from experiments show the improved performance of the proposed algorithm,
in the scenarios of heavy occlusions, illumination variations and similar
objects interference, when compared with the other descriptors based trackers.