ystemic
model
d non-
neously
id only
tions of
ape and
on 3d
nension
yriented
els and
gration
cted as
centres,
xis of a
axis 18
ft-hand
f inner
cquired
ung or
? points
aken as
of the
photo).
al with
e and
oint a2
S
International Archives of the Photogrammetry,
P,
XX
t (X)
Fig.2-1
(X,Y. Z) and (xy Y.Z) are corresponding photo pair
coordinates of baseline coordinate system.
2.3 Connection of Free Network Models
The elements of relative orientation of respect photo pair under
baseline coordinate system can be acquired after each photo
pair relative orientation. For providing initial values to block
adjustment, the elements of respect model relative orientation
and point’s coordinates must calculate under universal
coordinate system, namely connection of models.
After the relative orientation and the solution for the model
points of the second image pair, model connection can be
carried out with respect to the first model based upon the points
in the overlap area of the deferent models. This same procedure
is to be carried out for the other succeeding models. As to the
first image pair, the model scale is arbitrary; therefore, the scale
of the first model determines the free scale of the whole close-
range strip. From movement, rotation and scaling, the
independent models can be connected a whole close-range strip.
The SI is origin of first model baseline coordinate system. The
analogous transformation is:
X Yl IE
Flim bY] +7 (3)
Z 24:32.
i y 4j
where [x ¥ 7 L- any model point vector; M Jm model
rotation matrix; á scale coefficient; x, y Z. f =No. ]
model coordinate vector under first baseline coordinate system:
lu Y 2 = No. j model any 1 point coordinate vector
under corresponding baseline coordinate system. Based on first
model, the close-range independent models can be connected as
whole strip (see Fig. 2-2).
7/7 d
Fig.2-2 close-range strip
te
=
269
Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004
2.4 Solution of Connection of Big Angular Orientation
There are connections of big angular problems when taking
photo around building in convergent photography mode (see
Fig.2-3 Big angle photography
Fig.2-3). When the case of the independent image pair method
is adopted, the coordinate system is right hand system, and axis
of independent model coordinate is not parallel. The
circumstances of approximate vertical angular occur in the
change of direction, it need transform the model coordinate
system as follows: transform the former model tie points to next
model coordinate system. Through rotating 90 degree
anticlockwise solve the coordinates in current independent
model. The solution process is as above. Simultaneously,
considering starting. values of exterior orientation clement
transfer between two models. The connection of angular
orientation is depend on the results of relative orientation
/ / ! .
calculated elements (p, K,p , 0 , K and left photo exterior
orientation elements (p, , ,, K, , the right photo orientation
.,0., K,., the primary relation is:
p? y? + p y
7
A E Room, s Re (4)
it can be written as:
p T pT =
RAR, = B R R. ; R R, (5)
T.V Represent direction. angular of photography
baseline.
Kok, = RR eoe
Known Rs, cansobets solved: -,.(0,, Ko The
independent model initial values of left image €p,, 0, , K, arc
0. Then the results can be taken into baseline coordinate system.
3. FREE NETWORK BUNDLE ADJUSTMENT
Under non-control point condition, only utilize relative control
condition solution spatial position of free network, which
involving not enough initiative data to carry out adjustment
calculation within part network, can result in rank defect. How
to eliminate the rank defect, is base of free network solution.
Using least squares operation can solve the free network
adjustment determine relative position of net figure, and
smallest norm condition give Helmert conversion under given
approximate coordinate system so that determine absolution
position of network, and build centroid as relative reference
coordinate system. Because normal equation matrix is not
complete rank, rank defect is 7, which include the seven
freedoms of basis of adjustment calculation, must be ensured to
solve rank defect problems.