terrestrialphotogrammetry and laserscanning 2011
DESCRIPTION
geomaticsTRANSCRIPT
-
Terrestrial Photogrammetry &
Laser Scanning
-Mapping Science Overview -
Stuart Robson [email protected]
-
Introduction to Terrestrial Photogrammetry
Principles of terrestrial photogrammetry The digital image and small format cameras
Coordinate systems, resection, intersection and bundle adjustment
Uncertainty and camera calibration
Example applications Stereo, multi-photo, panoramic
Photogrammetry with Targets Coded targets and example applications
-
From Leonardo to Laussedat
da Vinci c.1480 The appearance of points and lines to
the eye perspective geometry
First used in mid 19th century from balloons (Laussedat - 1885) and for surveys of
buildings (Meydenbauer)
-
VEXCEL ULTRACAM D 8 sensors delivering an image of 11,500 x 7,500 pixels
Flight over UCL, 4cm pixel foot print with Applanix 510 IMU Microsoft 3D model the worlds largest 3000 cities in the next 5 years
.. to today
-
Measuring with light
-
Kodak DS460 (~1995) 6MP
Hasselblad H3D
(~2007) 39MP pixels
Some close range imaging systems
GIS INCA 3 metric RolleiMetric6008
Digital 39MP
The GSI
ProSpot
projection
system
AXIOS 3D CamBar
-
Light photons
incident on the
sensor
material are
collected to
produce an
electrical
signal at each
pixel
Analog image
voltage and
timing signals
produced by
reading the
signal
produced at
each pixel in
turn.
Analog image
signal
quantised into
individual
pixels by
analogue to
digital
converter
Digital image
data in
computer
readable form
Distance
I A/D
Converter
(typically 8 bit, but
10, 12 16 and 32
bits possible) Distance
gv
0
2 5 5
Analog signal Digital representation
Digital Image Acquisition
-
Two principal types of image sensor
Interline transfer, is derived from the TV and broadcast standards where the
array produces an interlaced image to
minimise the quantity of data
transmitted whilst maintaining the 25
frames per second necessary to avoid
perceptible image flicker.
Method is limited in that the odd and even lines represent two
different periods in time
Frame transfer sensors are organised such that the light sensitive regions are
also used to transfer charge and the
image is read-out as a single frame.
Method depends on an independent storage and readout zone or a
mechanical shutter to prevent light
reaching the sensor whilst the
image information is read out.
Sensor
Element
Row Bus
Column Bus
Horizontal Scan Register
Vertical S
can R
egiste
r
Output
Amplifier
Video Out
Digital Image Sensors (CCD and CMOS)
-
Bucket Array Analogy
Photons
Gauge
Conveyors
Conveyor
-
It is conventional for a photogrammetric image coordinate system to
have an origin at the centre of the image format, coinciding with the
optical axis of the lens in an ideal central perspective projection. Image
sensor arrays are highly regular structures, which given consistent
electronic signal timing, provide an excellent image coordinate system.
x photo co-ordinate axis
y photo co-ordinate axis true origin
Y pixel axis (0,0)
(0,0)
Digital image
Y pixel size
X pixel size false origin
X pixel axis
However, the
sensor scanning
process
conventionally
reads from the top
left corner of the
array, line by line,
towards the
bottom right
corner. A simple
2D transformation
is therefore
necessary to
obtain the familiar
photo coordinate
system.
Image Coordinate Systems
-
Sub-pixel Location of a Circular Target Image on a Dark
Background
Pixel
Value
Threshold
Pixel
Value
Pixel
Value
Pixel Number
Pixel Number
Centroid
Sub pixel location
T
0
0 0
255
255 255
Intensity
A/D conversion
-
One or more cameras probing a single point
Y
X
Z
(Xo, Yo, Zo
w,f, k)
P
yx
z
X
Y
Z
camera 1 camera 2
OnlinesystemPrinzip.ppt
probe
object
-
Two or more cameras measuring multiple points
Right image Left image
Stereoanordnungen.ppt
b
Y
X
Z
-
General case: A Multi-photo Network
mehrbild1.ppt
-
Fundamental angular uncertainty
The ellipses represent the uncertainty of positions determined by the intersection of direction observations from two camera positions.
?
-
Essentials camera calibration
-
Essentials bundle adjustment (network adjustment)
Scale
Network of mechanically unconnected cameras
If the network geometry is strong enough (# images and intersecting rays per feature point) it is possible to determine parameters describing the systematic distortions in the camera(s) used. This process is termed self-calibration and allows the use of a wide variety of imaging sensors.
-
A carving from Sumatra
Examples of similar outputs to the aerial photogrammetry case with automated area and feature based matching techniques
Simple stereo pair
-
-200.00 0.00 200.00 400.00 600.00 800.00 1000.00
-200.00
0.00
200.00
400.00
Automatically generated contoured surface
-
Orthophoto one of the original images draped over the 3D surface model
-
Automated surface measurement examples
-
F. Guerra (Italy), C. Baletti, D. Miniutti The Arena of Verona
Instituto Universitario di Architettura di Venezia
A traditional stone by stone output based on multi-image registration, followed by stereo plotting into a CAD package Automation difficult, but possible with less structured outputs
-
Set of stereo images from a petro-chemical plant survey
Following a network adjustment, data might be manually plotted, or
extracted automatically based on edge and feature extraction coupled with
expected component geometries e.g cylinder reconstruction from tangents
and centre lines
-
A portion of the as built model (isometric view) Example CAD output, modelled into PDS/PDMS
-
Panoramic imaging - Optag: infrastructure tracking
system
Combined panoramic photogrammetry and radio frequency tagging real time photogrammetric panoramic camera
far field radio tag system
integrated together to track individuals within an airport environment
-
Photogrammetry with Targets
Targets provide a unique feature that is purpose designed to produce a signature image on a sensor
Automatically measured based on image scanning at a specified threshold
Used for resection of camera, locations of targets or probe systems
Variety of coding techniques
Note - similar basis to surveying with targets, but machine readable numbering offers
many advantages
-
Automatic processing with targets
Measure images: identification of known targets and
measurement of other imaged targets
Location of cameras: given appropriate spatial
information the location and orientation of each camera is
determined at the time the image was taken.
Identification and location of targets: given camera
orientations the identities and locations of new targets
are established.
Compute parameters of interest: for example the attitude
of the object, change in shape or motion parameters.
Re-compute the solution for the next set of images:
Using, for example, target tracking to enable a very rapid
update of the parameters of interest.
Sett
ing
up
Repe
ate
d
Inf
ormation
ava
ilable a
t ca
mera
syst
em
Significant research on automation: red light green light systems
-
R&D with NASA Langley: Stretched lens array
-
Monitoring 3D change during a structures test R&D with UCL Mech. Eng Oil Rig components
-
R&D with NASA Langley: Parachute flight performance
-
Dimensional and Accuracy Control Automation for shipbuilding
Photogrammetric edge measurements in multiple images to 3D reconstruction
-
Medical: surface measurement in support of optical tomography
An infant born after 24 weeks gestation
(~6 months)
Phot
ogra
mmetr
ic
surf
ace
Validation
CT S
can
Medical Physics, Computing Science and Geomatic
Engineering
-
Terrestrial Laser Scanning
Principles of laser scanning Time-of-Flight, Phase & Triangulation systems
FoV, scan pattern, specifications
Data acquisition Error sources, surface effects, range
Data processing Registration, points or triangles? surface-
growing, thinning, building CAD models
Applications
-
Overview of laser scanning systems
3D laser scanners record three-dimensional coordinates of numerous points on an object surface in a relatively short period of time.
A laser beam is projected onto the surface of the object to be measured and the horizontal angle, vertical angle and range are recorded to deliver 3D information.
Accuracies are between several tens of mm and a few cm, depending on object surface properties, instrument design and the range to the object from the scanner.
Applications City Modelling & Urban Planning Architecture & Facade Measurement Tunnel Surveying Archaeology & Cultural Heritage Documentation Topography & Mining Process Automation and Robotics Scene Acquisition for Virtual Reality Reverse Engineering
-
What makes it possible? - the semiconductor laser
Laser scanners use small semiconductor (diode) lasers to convert a pulse of electrical energy into a pulse of optical energy with high efficiency and high reliability.
A laser diode is a small cube of semiconductor material with two flat and parallel faces which form the mirrors of the laser cavity.
Light generation takes place in the very narrow active region ~ 1 m thick
The divergent laser radiation emitted by the semiconductor is collected by a collimating lens to form a narrow beam.
Laser wavelengths used in scanning are in the infrared (invisible) and green (visible) part of the spectrum (1 mm to 700 nm).
Regulations require manufacturers to certify each laser product as Class I (least hazardous), II, III, or IV, depending on the characteristics of the laser radiation emitted (http://www.fda.gov).
-
Laser Scanning system examples
Range measurement
Pulsed or phase measurement
Full waveform
Triangulation systems
FoV, scan pattern
Typical specifications
-
Total stations > direction and range
-
Time of flight laser scanning
Laser scanning systems emit a laser signal (1) to record the position of a point in object space.
Scanning of the laser beam (2) is achieved using one to two reflective surfaces (3) which are linked to accurate angle motors and angle encoders to allow changes of the deflection angle in small increments.
In addition, the entire instrument may be rotated to achieve a complete 3-dimensional point coverage (4). Alternatively a second mirror may be used
Angle encoders deliver the direction of the beam
The method used to measure range depends on the accuracy and distance capability required of the device. Example: Riegl (www.riegl.co.at)
-
Pulsed time of flight measurement
Time of flight sensors derive range from the time it takes light to travel from the
sensor to the target and return.
A laser diode sends a pulsed laser beam to the object. The pulse is diffusely
reflected by the surface and part of the
light returns to the receiver.
The time that light needs to travel from the laser diode to the object surface and back
is measured and the distance to the object
calculated using an assumed speed of
light.
Pulse-type time of flight systems are typically used over ranges of several
metres to several hundred metres.
The accuracy of these sensors is typically limited by the accuracy with which the time
interval can be measured, and the rise
time of the laser pulse.
-
Operation of a pulsed laser distance meter An electrical pulse generator periodically
drives a semiconductor laser diode
sending out light pulses, which are
collimated by the transmitter lens.
Via the receiver lens, part of the echo signal reflected by the target hits a
photodiode which generates an electrical
receiver signal.
The time interval between the transmitted and received pulses is counted by means
of a quartz-stabilised clock frequency.
The calculated range value is fed
into the internal microcomputer
which processes the measured data
and prepares is for range (and
speed) display as well as for data
output.
It is possible to the select different
data processing algorithms,
according to the prevailing
conditions and requirements
-
HDS2500 (Leica) ~1998
40 x 40 degree field of view
1000 points per second
Produces a 3D point cloud
Single point accuracy of 6mm
Uses two rotating mirrors with their axis of rotation set at 90 degrees to each other
Includes a digital camera designed to work as a viewfinder for the system
Data in point cloud coded according to the strength of the laser return signal
http://www.ascscientific.com/cyrax.html
-
Leica ScanStation C10 (2009)
Range up to 300 m 90% (134m 18%)
Laser Class 3R (532nm)
Field of View 270 x 360
Measurement capability 6 mm (single shot)
2 mm (surface average)
Spot diameter 4.5mm to 7mm
Angular accuracy 12 arc
Measurement rate 50 000 pts/sec
Additional Capabilities Integrated camera 2K x 2K pixels
Laser plummet
Dual axis compensators
Target capability to 2mm std dev.
http://www.leica-geosystems.com/en/HDS-Laser-
Scanners-SW_Leica-ScanStation-C10_79411.htm
-
Phase / modulated beam range measurement
The "modulated-beam" sensor also uses the time light takes to
travel to the target and back, but
the time for a single round-trip is
not measured directly.
The strength of the laser is rapidly varied to produce a
signal that changes over time.
The time delay is indirectly
measured by comparing the
signal from the laser with the
delayed signal returning from
the target.
Given several frequency modulations it is possible to
compute the number of full
wavelengths to the target (cycle
ambiguity) and to add these to
the offset T1
Waveform offset
(Thiel & Wehr, 2004)
-
Phase shift distance measurement
B = brightness A = amplitude f = phase d = range lmod = modulation wavelength
Kahlmann, Remondino, Ingensand (2006)
-
Z&F 5006i
Range up to 79 m 90%
Laser Class 3R (visible)
Field of View 310 x 360
Measurement capability 0.7mm rms at 10m (20%)
0.4mm rms at 10m (100%)
3.5mm rms at 50m (20%)
1.8mm rms at 50m (100%)
Spot diameter 3mm at 1m
Beam divergence 0.22mrad
Measurement rate 508 000 pts/sec
Additional Capabilities Tilt compensation
http://www.zf-laser.com/e_imager5006.html
-
Metris MV330 / 350 from MetricVision (USA), now Nikon
Field of View
90 x 360 degrees
Multiple lasers (Class 1)
2 visible lasers to point and focus
1 infra laser for time-of-flight distance measurement
Two range options - 30m, 50m
Several measurement modes
from 4000 pts/sec with 0.3mm typical accuracy
to 2 pts/sec with 102 mm at 10m
Accuracies are achieved through the use of beam modulation and extensive signal processing
www.nikonmetrology.com/large_volume_metrology/laser_radar
-
Full waveform scanning (after Riegl 2008)
-
Riegl VZ400 (2009)
Range up to 500m 80%, (160m 10%)
Laser Class 1 (near IR)
Field of View 100 x 360
Measurement capability
5 mm (accuracy)
3 mm (precision)
Beam divergence 0.3mRad
Angular resolution 1.8 acr
Measurement rate
125 000 to 42 000 pts/sec
Additional Capabilities
Fitting for Nikon digital camera
Laser plummet
Dual axis compensators
Target capability to 2mm std dev.
GPS receiver
-
Triangulation
A light spot or stripe is projected onto an object surface and the position of the spot on the object is recorded by a CCD cameras.
The angle of the light beam leaving the scanner (a) is internally recorded.
The fixed separation (D) between laser source and camera is known from calibration.
The direction of the reflected
laser spot (b) is computed by
measuring the location of its
image on a sensor array (P1)
The distance from the object to
the instrument is geometrically
determined using a, b and D.
The diagram shows that a
second spot at a differing range
would yield P2 and a different
value for b.
-
Example -Minolta VIVID 910
Object distance (range) 0.6m to 2.5m
The object is scanned by a plane of
laser light which is swept across the
field of view by a mirror, rotated by a
precise galvanometer
Reflected light from each scan line is
observed by a single frame, captured
by the CCD camera to provide over
300,000 vertices per scan
Scanning field of view depends on
interchangeable lenses used
Data captured in 2.5 seconds
A (24-bit) colour image is captured at
the same time by the same CCD to
provide RGB information for each 3D
data point
http://www.konicaminolta.eu/index.php?id=2079
-
Object coverage for Minolta Vivid 910
Lens Near field (mm)
(@ 0.6 m)
Far field (mm)
(@2.5m)
max depth resolution
Tele: 25mm 111 x 84 x 40 460 x 350 x 130 0.039 mm
Mid : 14mm 196 x 153 x 70 830 x 622 x 220 0.068 mm
Wide : 8mm 355 x 266 x 92 1200 x 903 x 400 0.090 mm
A triangulation scanning system reaches 3D point standard deviations of less than one millimetre at very close range (less than 2 meters).
The accuracy depends on the length of the scanner base, the optics used and the object distance.
With a fixed base length, the standard deviation of the distance measurement will increase in proportion to the square of the distance.
-
Error sources
Broadly similar to total station with REDM
Non-optical
Vibration
Air turbulence
Mechanical error
Human error
Optical
Speckle, signal buried in noise
Spot size
Range shift and noise: laser light surface penetration
Range artefacts: edge and reflectance jumps
Strength of laser return signal
Level of background illumination
Reflectivity of surface / colour
Angle of incidence
Calibration of instrument
-
Reflectivity of Various Surfaces / Materials
The amount of light that is returned from a target's surface is characterised by the reflection coefficient r.
For a diffusely reflecting target, the maximum value of r is 100 %.
For mirror-like or retro reflecting targets, the (theoretical) value of reflectivity can exceed 100 %.
The reflection coefficient also depends on the wavelength.
Diffuse reflection:
The signal is reflected omni-directionally according to Lambert's cosine law
Specular reflection:
The angle of the reflected beam with respect to the targets surface is equal to the angle of incidence. Incident beam and reflected beam lie in
the same plane.
Retroreflection:
The retroreflected beam is returned in the same direction from which the incident beam came. This property is maintained over a wide range
of directions of the incident beam
-
Maximum Range versus Target Reflectivity
The maximum range achievable with a laser
scanner depends strongly
on the reflectivity of the
target.
Range performance (as specified by RIEGL) is
given for a diffusely
reflecting (lambertian)
target with a reflectivity of
80 percent.
For a target of different reflectivity, the maximum
range can be found with
the range correction factor
as given in the diagram.
-
Maximum Range as a Function of Visibility
For long range systems the maximum range achievable
with a laser rangefinder
depends strongly on the
meteorological visibility.
Range performance can be given with respect to a
meteorological visibility of 20
km (clear air).
At lower visibility, the maximum range is reduced due to the
atmospheric attenuation
according to a range reduction
factor
-
Laser scanning acquisition and processing
Acquire scan data from multiple view points
Multiple clouds of 3D points, each point with its own sources of error scanning method, scan spot size, object surface qualities, colour information
Clean and Register data together
initial cleaning to remove any data that may obstruct the registration process
Register with Common points or mathematical fitting based on surface similarities
A single point cloud with overlapping areas and data of varying degrees of quality
Convert data into a model suited to final purpose
Points, triangles, mesh or NURBS model
Web delivery
Sharing information between institutions
Archive
-
Data processing steps Cleaning
The Cleaning Process involves removing unnecessary, unwanted, or bad data from the component image. Although cleaning of the data can occur throughout
the process the initial cleaning should remove any data that may obstruct the
alignment process.
Alignment
The alignment process transforms one image into position relative to another image. The scanning process results in several view oriented image of physical 3
dimensional object. The multiple view orientated scans (component images) can
be aligned until a completed 3d object is created (composite image). Overlapping
data between the component images is used to align them together.
Editing
The Editing process includes many methods of manipulating spatial (xyx), colour (rgb), and normal (ijk) data. Measurement data is edited to improve the quality,
filter data, enhance the colour, or segment the data into structures.
Hole Filling
The Hole Filling process creates new data within a hole. A hole is essentially a region where no measurement data exist. Holes are filled by blending new data with the surrounding data.
-
A single scan (Leica HDS 2500)
-
A second scan position
-
Registration Coordinate Systems
Individual scan data are based on a coordinate system defined according to the orientation of the scanner
Parameters of a 3D similarity transformation (3 translations and 3 rotations) are required to register data from two or more independent
scans
A minimum of 3 common points between scans are required
-
Registration Sources of common data
Control targets, identifiable in the scan
Must locate physical targets in the scan volume, positioned so as to provide sufficient common points
Optionally use a high definition scan to find target centre
Possible to link to external coordinate system through a target survey
Key advantage is that control targets provide clearly identifiable common points
Common natural features
Rely on natural features of interest
Natural features typically identified after scanning
Transformation parameters computed by minimising computed discrepancies between surfaces from different
scans
using iterative closest point
least squares surface shape matching
Dependant on appropriate features being available
-
The combined (registered) view
-
Engineering Applications
-
Natural Features
-
Natural Features Grimes Graves
-
Film Special Effects
-
Architectural Applications
-
Rialto Church, San Bernardino, California
Model created with a Leica HDS scanner, then modelled in CloudWorx and Autocad
-
Rialto Church, San Bernardino, California
-
Adding Colour..
Example - Riegl Z420i (2005) Measurement Accuracy 12mm (topo mode), 5mm (survey mode)
Red, Green and Blue Lasers
Optical combination
Schematic for a colour triangulation system
-
Combination of imaging and laser scanning
Carpiniana - the Italian delegate to World Summit Award in the e-Science category
-
Arius3D Foundation System
RGB colour from three lasers,
80mm spot diameter
100mm sampling interval
maximum dimensional error 25mm
Scanning cross section ~ 0.6 x 0.8 m
Arius im
ages c
ourt
esy o
f R
OM
http://www.arius3d.com/
-
Egyptian childs skull Arius 3D scanner
-
Laser scanning - Summary
Directly acquire a set of 3D surface measurements from a single instrument position provided that the surface concerned will reflect a laser beam
Distance measurement principle based on either on time of flight, phase, or triangulation
Data acquired in a regular fashion
Multiple scans required to overcome object occlusions
Registration between multiple scans required, either by use of physical targets, or through matching common surface features
Scanning systems tend to be built for specific purposes, e.g Cyrax 2500 or Minolta Vivid
Generate massive quantities of data which require significant post processing to produce a surface model