Multiple-View Video
for Traffic Surveillance (OIS)
Project-Page
3D Scene and Object
Representation and Interactive User Navigation
For visualization and control
purposes, a 3D-model from all input cameras of a crossroad will be generated.
In comparison to the 2D-case, where intermediate images were interpolated, an
interactive user navigation with arbitrary views will be possible. Since the
original camera views differ significantly in lighting condition, scene
creation will be carried out as follows: First, a segmentation algorithm wil be
applied to obtain static background information. The background images are then
mapped onto a ground plane in the 3D-environment. Foreground objects from
different views, that were extracted by segmentation will be represented by
mapping their original texture information onto simple artificial 3D-objects
from a database. The following example shows a 3D-scene with two artificial
objects on the background plane.
(Click to view scene in a
VRML-world, 1MB, please be patient, may take a while)
Scene Navigation on projected
background and artificial 3D-objects
In this example, only 3D-geometry information was used for the
3D-objects. Later, texture information will be added to such objects.
After the creation of static scenes, motion information will be added to
update the current position of each moving object. The appropriate procedure is
described in “3D Scene and Object Representation and Interactive User Navigation”.
People:
Michael
Droese
Karsten Mueller
Aljoscha Smolic
Last modified: May 27, 2002