Overview of Applications of Digital CloseRange Photogrammetry

1
Overview of Applications of Digital Close-Range
Photogrammetry

• Clive S. Fraser
• Dept. of Geomatics

2
Applications areas of close-range digital
photogrammetry

• Industrial photogrammetry (vision metrology)
• Engineering measurement (e.g. civil
  geotechnical)
• Archaeological photogrammetry(Heritage
  recording)
• Architectural photogrammetry (Heritage
  recording)
• Traffic accident reconstruction
• Forensic photogrammetry
• 3D modelling for animators, the movie industry
  and virtual reality builders
• Biostereometrics and medical imaging
• Process plant documentation
• Underwater photogrammetry

3
Trends in close-range photogrammetry

• Higher resolution sensors leading to higher
  precision
• Improved computational models
• Full automation of the measuring process
  (real-time off-line)
• Enhanced modelling and visualisation
• Availability and suitability of low-cost
  digital cameras
• offering a broader range of applications in
  new fields

4
Operational framework a diverse range of
requirements

• Simple object Highly complex object
• Low-end instr. High-end system
• Simple survey Complete, precise 3D
  documentation
• Use by experts Suitability for general use
• Limited budget High cost
• No time factor Time constrained

5
Automation of the photogrammetric
measurement process

• Automated design not widely employed
• Automated image recording use of remotely
  controlled cameras
• Automated image measurement possibly within the
  digital camera
• Automatic orientation via resection or relative
  orientation
• Automatic point correspondence determination
  point triangulation
• Automated mesh generation and texturing where
  possible not common as yet

6
From High End
Off-line vision metrology
On-line, real-time vision metrology
7
Off-line photogrammetric object reconstruction
via image matching

• All network images first recorded, after which
  photogrammetric orientation may be manual or
  automatic and 3D point cloud extraction is
  automatic
• Close attention required to network geometry (low
  convergence angles required) image matching
  usually done via stereo pairs, but multi-image
  matching possible
• Image matching, especially least-squares
  matching, is much slower than point determination
  via target centroiding requires good image
  quality

8
higher sensor resolution leads to higher metric
accuracy
Triangulation accuracy is a function of imaging
scale, geometry, number of exposures image
measurement precision
sXYZ (q /
k1/2) S s sXYZ XYZ coordinate
standard error q empirical factor
(approx. 0.7) S scale number
(dist/focal length) s std. error of
image xy coords. k number of
images per station
Accuracy potential for digital cameras with a
350-500 field of view (pixel size of 9 mm, image
measurement accuracies of s1/30th pixel
s1/3th pixel)
targets non-targeted
9
Application diversity of industrial vision
metrology
Defence
Automotive
Engineering
Aerospace
Fabrication
Automated C-R Photogrammetry
Ship Building
Mining
Antennas
10
Coded targets allow automatic measurement

• Essential from a practical point of view for
  multi-image monoscopic convergent networksNot
  strictly essential for 3D model reconstruction
  from stereo imagery, but very useful for initial
  relative or exterior orientation

Coded targets
Retro-targets
EO Device

• Exterior orientation device (EO-Device) is a
  special coded target to establish both scale and
  an XYZ reference datum

11
Surface texture or projected patterns for surface
model extraction
Pattern projector
Natural surface texture
Artificial texture a pattern
12
System configurations for automatic C-R
photogrammetric measurement
On-line vision metrology system for measuring
arrays of projected targets

• Precise exterior orientation from coded targets
• Projected targets measured via feature-based
  matching, the projected targets being the
  features

13
Automatic camera calibration
A full metric modelling of interior orientation
and lens distortion for colour sensor is being
achieved to around 0.1 pixel in an operation
requiring only a few minutes
14
Example of automatic off-line vision metrology
11 stations 875 points measured in 10 seconds
2.5m Antenna Deformation Measurement of High Gain
Antenna
Note need for controlled illumination strobe
underexposed background
15
Can also automate the following
operations Absolute orientation via coded
targets Scaling the model via coded targets
16
Typical network for automated measurement Ship
block measurement
100 images, 1000 points
17
Industrial photogrammetry reverse engineering of
a tilt train

• Combined off-line real-time measurement
• Off-line survey 3000 strip target points
• Real-time survey 1000 probed points
• Images 130
• Survey duration 10 hours
• Survey accuracy better then 0.1mm
• 'Catia CAD used to generate 3D model

From point cloud to rendered CAD model, with all
design changes via CAD
18
Deformation monitoring at Federation Square
North Atrium
Federation square
19
North atrium of Federation Square

• Atrium comprises two skins of in-plane frames
  separated by an average gap of approximately 1.5m
  consisting of 4 to 5 sided irregular polygons

Deflections expected to be as much as 50 mm
20
Photogrammetric network

• Vision Metrology System Networks
• V-STARS system with INCA camera
• 2cm retro-reflective targets along with coded
  targets on inside of both inside outside frames
• 32 basic camera station positions with 2-5 images
  per station ? 90-130 images per epoch
• RMS object point coordinate accuracy 0.15mm
• Time for photography 30 minutes
• Image measurement data processing 5 mins
  (fully automatic)
• 6 absolute datum points established

21
Deformation analysis for North Atrium of
Federation Square
Epoch 5 versus 1 After glazing, max.
deflection 22mm
Epoch 4 versus 1 After final de-propping, Max.
deflection 8.3mm
22
Photogrammetrically monitoring bridge beams - the
problem

• New bridge specifications call for increasing
  load capacities
• Over time, many existing bridges are falling
  below design specifications
• Insufficient capital expenditure on
  infrastructure to consider major new bridge
  building
• ? The need for bridge upgrading

23
Beam Strengthening via CFRP Stirrups
24
3D measurement requirements

• Beam displacements required at 1000-1500 surface
  points
• Measurements required at up to 15 load
  increments each increment is 20-60kN failure
  is at approx. 400kN
• Need to correlate 3D surface point measurements
  with displacement transducer (LVDT) data
• Accuracy of better than 0.1mm maximum
  displacement is approx. 3cm
  for a 6m beam

Single-sensor VM offers speed, accuracy,
reliability and process automation
25
VM/Photogrammetric Network

• 20-stn convergent network with set-back distance
  of approx. 6-7m
• Imaging scale of 1330 and 0.03 pixel measurement
  accuracy ? sXYZ 0.06mm
• Rapid data acquisition required (1 minute) shape
  invariance to be confirmed
• Automatic measurement utilising EO device and
  coded targets
• Full data processing completed prior to next
  measurement epoch ie within 4 minutes

Large data volume of approx. 200,000 surface
point measurements per beam, with computations
QA completed 3 minutes after final epoch (ie
after failure)
26
Targets comprised retro-reflective dots, coded
targets an EO device
27
(No Transcript)
28
(No Transcript)
29
measurement results
Comprehensive 3D beam displacement data (not
possible with LVDTs alone)
180 kN
380 kN
30
Deformation Measurement of a PC9 Trainer Aircraft

• Deflections result from an induced static load.
• Deformation vectors for fuselage twisting (from
  engine torque), lateral tail fin movement
  longitudinal bending reached 10 cm magnitude

31
Deformation Survey of the ADI Bushmaster
Aim To photogrammetrically measure to 0.5mm
accuracy the deformation to the vehicle caused by
a mine explosion
32
Deformation Survey of the ADI Bushmaster
Point displacements resulting from land mine
detonation
33
Cadia/Freeport Mine Projects

• Deformation Measurement of two of the Worlds
  Largest Electrical Motors

Rotor/Stator separation due to deformation
reached 3mm
34

• Dimensional Inspection and Deformation of a
  Rotary Kiln

Profiles checked for circularity/ linearity
also determine deformation 30 images, 400 pts,
0.5mm accuacy
35
Rudder skin of Boeing 777 / 300

• 777 Rudder
• Inspection
• Tooling hole coordination
• Reverse engineering
• 0.2mm accuracy

36
AUTOMATIC MEASUREMENT EXAMPLES Deformation
Monitoring of a Historic railway Bridge

• Survey carried out to 0.5mm accuracy using a
  consumer-grade camera (Nikon D100)
• Very inexpensive exercise, the main cost being
  simply applying the targets

37
Dynamic monitoring example automatic tracking of
a targetted parachute
Approx. 100 points tracked by 6 cameras as
parachute falls, frequency 30 Hz
38
Dynamic monitoring example automatic tracking of
a targetted parachute
39
Accident Reconstruction surveys with close-range
photogrammetry
Efficient automatic orientation followed by
semi-automatic and manual feature point extraction
40
Automatic network orientation with manual curve
extraction
41
Traffic Accident Reconstruction Difficult
Geometry
Camera is a 7 mpixel Olympus C7070Wz zoomed fully
out (f5.6mm)
Expected Accuracy Mean sigma of XYZ coords. 5
cm or 12100 of size RMS of image residuals
1.1 pixels
42
First iWitness example
43
(No Transcript)
44
(No Transcript)
45
(No Transcript)
46
(No Transcript)
47
AR scenes can be complex with poor geometry
48
AR scenes can be complex with poor geometry
49
Texturing of 3D Models via iWitness
Rectified image on planar surface in object space
Texturing in this case achieved via plane
rectification of image patches
50
Building the textured model 1) point cloud or 3D
curves, 2) wireframe model 3) texture mapping
via plane rectified image patches

• Higher definition can be achieved through smaller
  image patches (even single pixels) in the texture
  mapping, but at a cost of time and effort.
  Automation of the process is feasible.

51
Photogrammetry for heritage recording

• Mapping of monuments sites
• 3D reconstruction of objects
• Documentation
• Visualization and presentation

52
(Patias, 2003)
Photogrammetric measurement outcomes

• 2D vector reconstructions
• Planar texture maps
• 3D vector reconstructions
• 3D texture representations

53
Example 1 Recording visualization of BET
GIORGIS, Ethiopia
54
Model building, rendering visualization of
Bet Giorgis, Ethiopia
55
(Patias, 2003)
Example 2 Church interior recording
reconstruction
Purpose Import to GIS Product
2D 3D vector, textured Methodology Multi
photo arrangement Emphasis Visualization

56
Example 3 computer reconstruction of artifacts
57
courtesy ETH Zurich
Example 4 Bayon Temple
Automated 3-D reconstruction of a complex
Buddhist tower of the Bayon Temple, Angkor Thom,
Cambodia
Project Aim Automated derivation of a texture
mapped 3-D model of a very complex object using
tourist-type terrestrial images.
58
Bayon Temple

• The Angkor Site in Cambodia Hindu and Buddhist
  monuments listed in the UNESCO World Heritage
  List
• Project goal Image-based reconstruction of one
  of the many complex Buddha-faced towers of Bayon
  Temple in Angkor Thom

Image acquisition
59
Procedures and results
Semi-automated phototriangulation
Automated surface reconstruction, editing and
triangulation
Visualization
View-dependent texture mapping
60
(No Transcript)
61
The Great Buddha of Bamiyan
3-D reconstruction of the Great Buddha statue of
Bamiyan, Afghanistan

• Project Aim
• Reconstruction of a 3D model of the Great Buddha
  of Bamiyan
• The 3-D model could serve as the basis for future
  physical reconstruction

62
The Bamiyan Valley, Afghanistan

• 200 km N-W of Kabul
• 2500 m altitude
• Center of silk road
• Major Buddhist area (gt100 statues, 5000 monks)
• 3 larger Buddha statues cut out of the cliff (ca
  200 AD)

ca 900 m
63
Ikonos image, 1 m
64
The Great Buddha of Bamiyan ca 200 AD - March
2001

• 53 m high - the tallest representation of a
  standing Buddha
• Cave was covered by frescos and paintings
• Statue was covered with mud and straw to model
  face folds
• Statue was probably painted in gold and colors
  and decorated

65
Photogrammetric Reconstruction Three Data Sets
(in parallel)
2. Amateur images provided by a tourist - no
information available
Research work to test algorithms
Research contribution to the possible
reconstruction project
66
The Great Buddha of Bamiyan
Automated reconstruction with Multi-photo
Geometrically Constrained Least Squares Matching
Manual reconstruction
67
The Great Buddha of Bamiyan
68
The Great Buddha of Bamiyan
69
Conclusion and Outlook for Automated Close-Range
Photogrammetry

• More automation ? greater ease of use
• Growing demand across broader applications
  domains for more flexible and robust low- to
  moderate-accuracy systems
• More demand for low-cost systems
• Demand for better accommodation of difficult
  image geometry
• Better modelling visualisation capabilities
  especially 3D
• Integration with other measurement systems (eg
  TLS) data fusion
• Greater integration of derived 2D 3D
  information into information systems (eg GIS)
• leading to a very healthy future!