anbul 2004
800 (nm) 859
and field
for 30)
racy. The
etric tests
adiometric
The field
€ camera
ion
lay-to-day
variety of
te. before
3. DATA PROCESSING
Unlike with frame-based photography, the three-line geometry
is characterized by a nearly parallel projection in the flight
direction and perspective projection perpendicular to that. The
sensor model for the TLS images is based on modified
collinearity equations and uses different forms of trajectory
models. These models are used for the improvement of the
measured exterior orientation parameters for each scan line by a
modified photogrammetric bundle adjustment procedure, called
TLS-
LAB (Linear Array Bundle adjustment). This is part of a
comprehensive package of new methods and the related
software for the processing of TLS imagery, which is described
in this chapter .
3.1 TLS Digital Photogrammetry System
The outline of the TLS data processing chain is shown in Figure
9 (Gruen et al., 2003).
À i E et
i GOP Information |
|
| Oniginal TLS | GPS amd INS “Came
{ Tae Data i Parauiclers
i ; E ?
y x 1 Y
rl
Y Y
Se : tadine TLS e E Er
ae TE Reading TLS (mages Interpotaling Position & Attitude
Image Database idus © Dafa for ench kmage Scanine
^ TLS Image Pre Froccesing le eU e eerte e etae
and Kahancement
Y 1
Tisneuhtion (Tic Point Measurement, =
pt M tm
f ! if 1
j Refined Position. | | Die Points
& Aitilude Dës | | {DSN A
| Semi - Y Y Y
| Automated i ;
1
Matching | NN
Se |
] Fully Automatic fmage Mulching
i |
im EL date —— 77
Paint / Feature Measurement: Frame arrete f= cr
c 1 P, A 'ex te: 1 ez m
In Mono or Stereo Mode 3 f es Vector |! Mass Points
a 1,
! Y
| TIN DIM Generation
VIN ( Ond Based D8M
DTMs, M Models
AD festahase ie jo
Images, TMs, Vector
EM ee] Inching Qu pipol
Ortho-Imagc Generation
Figure 9. TLS data processing chain
The processing modules include:
1)
User interface and measurement system: The user interface
allows the display, manipulation and measurement of
images. It includes the mono and stereo measurement
modules in manual and semi-automated mode. It employs
large-size image roaming techniques to display the TLS
forward, nadir, and backward (plus other channels if
available) view direction images simultaneously.
Triangulation: This module consists of two stages. At the
first stage, the directly measured GPS/IMU data is taken as
input and the exterior orientation elements for each scan-
line are calculated/interpolated for the time of image
es
Capture. The output of this procedure is called "raw
4)
5)
6)
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part Bl. Istanbul 2004
orientation data”. The raw orientation parameters are
already of fairly good quality and may be used in some
applications right away. For high accuracy applications
however we recommend a triangulation. The related
software is a modified bundle adjustment called TLS-LAB.
The software includes a special TLS camera model and
offers three different trajectory models: DGR (Direct
Georeferencing Model), PPM (Piecewise Polynomial
Model) and LIM (Lagrange Interpolation Model). The
self-calibration — technique for systematic error
compensation is currently implemented (Gruen, Zhang,
2003). The triangulation module also covers new methods
for semi- and fully-automated pass and tie point
measurement. Tie points in mult-strip/cross-strip
configuration, with different image scales and image
directions can be measured semi- or fully automatically
through least squares matching.
Image Rectification: Here the raw level image data is
transformed into quasi-epipolar form in order to reduce the
large y-parallaxes caused by high frequency variations of
the parameters of exterior orientation. This is absolutely
necessary for smooth stereo viewing. Rectification comes
in two modes. The coarse version just uses the orientation
elements as given (or already derived from triangulation)
and projects the raw images onto a pre-defined horizontal
object plane. The refined version uses an existing
DTM/DSM (of whatever quality) in replacement of the
object plane. This latter method reduces the remaining y-
parallaxes substantially. It should be realized that the
notion of “epipolar” images does not exist for this kind of
linear array-based images. Therefore these transformed
images can be regarded as “quasi-epipolar”.
DSM/DTM generation: A new matching strategy is
devised and implemented for the automatic generation of
Digital Surface Models (Zhang, Gruen, 2004), from which
Digital Terrain Models may be derived. This strategy
consists of a number of matching components (cross-
correlation, least squares matching, multi-image matching,
geometrical constraints, edge matching, relational
matching, multi-patch matching with continuity constraints,
etc.), which are combined in particular ways in order to
respond to divers image contents (e.g. feature points, edges,
textureless areas, etc.). The matching mo^le can extract
large numbers of mass points by using m -.-images. Even
in non-texture image areas reasonable matching results can
be achieved by enforcing the local smoothness constraints.
Ortho-image generation: This is a special solution for fast
derivation of ortho-images given the sensor geometry,
orientation and the DTM/DSM.
3D object extraction and modelling: For 3D object
extraction two different semi-automated modes have been
realized: (a) Point-based object extraction: This is
achieved by interfacing the TLS software with CyberCity
Modeler. For details see Gruen et al., 2003. The key points
are manually measured in stereomode, producing a 3D
point-cloud. The 3D model is generated automatically. (b)
Line-based object extraction: The lines of interest are
measured manually and monoscopically in the nadir image,
while the corresponding lines in the forward and after
images are then matched automatically.
