ng
let
he
le
Heikkinen, Jussi
2 IMAGE BLOCK CONSTRUCTION
In our imaging design the images will be taken in a very controlled way. The camera will not be attached to camera
tripod directly, only one end of a straight rod will be attached to it and the camera will be attached to the other end of
the rod (Figure 1). Images will be taken while turning the rod around the tripod during the turning sequence until a
whole circle is made. The subsequent images should have enough overlap since the individual image position will be
estimated in a freenet estimation based on tie point observations between subsequent images.
Figure 1 Imaging geometry in a circular image block
It is important that the circular image sequence; block; is closed since the co-ordinate system will be created on a site.
That means no control point network nor measurements for camera poses are required. The center point of revolution
will be determined as an origin for the measurements. Some scale measures may be though required. The restricted
imaging arrangement restrains the projection centers of each camera position to lie on the same plane and within the
same distance r from the revolution center.
For the 3D measuring point of view this circular image block is perhaps not perfect for the task, especially if the camera
is looking outwards from the revolution center and the consecutive camera positions will have diverging optical axes.
The solution is to turn the camera to the tangential direction of the circle (Figure 2).
Figure 2 Camera turned to tangential direction Figure 3 Two co-centric image blocks
By constructing two such circular image blocks on the same tripod position and changing the orientation of the
camera at the end of the turning rod, we will have two co-centered image blocks with same co-ordinate system
and image pairs with good imaging geometry (Figure 3). Images of an image pair will be at most two times as far as the
length of the rod from each other.
If video camera is used we can use all the images in a sequence to reconstruct the object instead of selecting the
images with best geometry. With this huge number of images we can substitute the lower quality of images to some
extent. In the object reconstruction part we can use parametric model estimation directly from image observations. By
involving parametric reconstruction method we can further reduce the effect of image noise on our results.
This kind of approach is applicable in such modeling tasks where you do not have any reference system nearby or the
camera constellation is difficult to design or build.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000. 359