| Abstract | Many reverse engineering and inspection applications re-quire generation of point clouds representing faces of physical
objects. This paper describes a computational framework for
constructing point clouds using digital projection patterns. The
basic principle behind the approach is to project known patterns
on the object using a digital projector. A digital camera is then
used to take images of the object with the known projection pat-
terns imposed on it. Due to the presence of 3-D faces of the ob-
ject, the projection patterns appear distorted in the images. The
images are analyzed to construct the 3-D point cloud that is ca-
pable of introducing the observed distortions in the images. The
approach described in this paper presents three advances over
the previously developed approaches. First, it is capable of work-
ing with the projection patterns that have variable fringe widths
and curved fringes and hence can provide improved accuracy.
Second, our algorithm minimizes the number of images needed
for creating the 3-D point cloud. Finally, we use a hybrid ap-
proach that uses a combination of reference plane images and es-
timated system parameters to construct the point cloud. This ap-
proach provides good run-time computational performance and
simplifies the system calibration.
|
