Linear LIDAR versus Geiger-mode LIDAR:
Impact on data properties and data quality
A. Ullrich*, M. Pfennigbauer*
*RIEGL Laser Measurement Sytems GmbH, Riedenburgstrasse 38, 3580 Horn, Austria
ABSTRACT
LIDAR has become the inevitable technology to provide accurate 3D data fast and reliably even in adverse measurement
situations and harsh environments. It provides highly accurate point clouds with a significant number of additional
valuable attributes per point. LIDAR systems based on Geiger-mode avalanche photo diode arrays, also called single
photon avalanche photo diode arrays, earlier employed for military applications, now seek to enter the commercial
market of 3D data acquisition, advertising higher point acquisition speeds from longer ranges compared to conventional
techniques. Publications pointing out the advantages of these new systems refer to the other category of LIDAR as
„linear LIDAR“, as the prime receiver element for detecting the laser echo pulses - avalanche photo diodes - are used in
a linear mode of operation. We analyze the differences between the two LIDAR technologies and the fundamental
differences in the data they provide. The limitations imposed by physics on both approaches to LIDAR are also
addressed and advantages of linear LIDAR over the photon counting approach are discussed.
Keywords: LIDAR, Geiger mode, laser scanning, airborne sensing, point clouds
1. INTRODUCTION
LIDAR in its traditional form as time-of-flight measurement with short laser pulses and a photodetector operated in the
linear regime has become the inevitable technology to provide survey-grade 3D in a vast variety of applications.
Applications include in the field of static laser scanning, e.g., acquiring data indoors, and providing as-built surveying of
industrial sites or long-range laser scanning in open pit mines. In the field of so-called kinematic laser scanning from a
broad range of platforms (land based vehicles, ships and all kind of aircrafts, all of these manned and unmanned)
application include acquiring 3D data in corridors or over large extended areas of thousands of square kilometers. One of
the specific strengths of LIDAR technology, in contrast to e.g. photogrammetry, is it multi-target capability enabling the
penetration of vegetation to reveal objects below the canopy or to provide data from the ground for deriving a high
resolution digital terrain model.
These traditional LIDARs come in two flavors, with so-called discrete returns based on analog signal detection and with
so-called echo digitization with subsequent offline full waveform analysis or online waveform processing. The echodigitizing LIDAR systems do not only provide highly accurate point clouds, but also a significant number of additional
valuable attributes per point. These attributes include calibrated amplitudes and calibrated reflectance readings for every
echo, but also attributes derived from the shape of the echo waveforms itself.
LIDAR systems based on Geiger-mode avalanche photo diode arrays earlier employed for military applications, now
seek to enter the commercial market of 3D data acquisition in airborne applications from high altitudes, advertising
tremendously higher acquisition speeds from longer ranges compared to conventional techniques [1]. Publications
pointing out the advantages of these new systems refer to the other category of LIDAR as „linear LIDAR“, as the prime
receiver element for detecting the laser echo pulses - avalanche photo diodes - are used in a linear mode of operation.
Subsequently we analyze the differences between the two LIDAR technologies and the fundamental differences in the
data they provide, especially with respect to the capability of penetrating the canopy of dense vegetation and to the
achievable accuracy level.
The information on Geiger Mode LIDAR presented in this paper is based on calculations and simulations carried out
with the authors’ best efforts. The parameters used for the calculations and simulations were obtained from publicly
available sources. The authors have executed their best efforts to respect any and all possible copyrights.
Laser Radar Technology and Applications XXI, edited by Monte D. Turner,
Gary W. Kamerman, Proc. of SPIE Vol. 9832, 983204 · © 2016 SPIE
CCC code: 0277-786X/16/$18 · doi: 10.1117/12.2223586
Proc. of SPIE Vol. 9832 983204-1
2. LIINEAR LID
DAR DATA
A PROPER
RTIES
Linear LIDA
AR as utilizedd in RIEGL LIDAR
L
enginnes and system
ms [2] make use of echo digitization and
a either fulll
waveform annalysis or onliine waveform
m processing to
t derive geo
ometry data annd valuable aattributes for each detectedd
target from a digital data stream [3,4]. Figure 1 deppicts a simplified block diaagram of the laser rangefin
nder part of a
LIDAR systeem. The laserr emits short pulses of som
me nanosecon
nd pulse widtth at wavelenngths ranging from 532 nm
m
(usually in baathymetry) to 1.5 µm for teerrestrial laser scanning, botth static and kinematic,
k
andd at 1.064 µm
m used for longg
range airbornne systems. Thhe receiver is usually an avalanche photo
o diode (APD)) operated typpically in the linear
l
mode of
operation. In this regime the
t output currrent of the APD
A
is proporttional to the optical
o
input ppower. Most linear LIDAR
R
systems provviding survey-grade accuraccies acquire data
d sequentiallly, employingg a single laseer and a singlle receiver. Too
provide spatiial data withinn a certain fieeld of view, deflecting
d
or refracting opttical elementss are used to scan the laserr
beam and recceiver instantaaneous field off view (iFOV)).
The lower lefft part of Figuure 1 shows example
e
signaals in the vario
ous regimes. It
I is assumed that the emittted laser pulsee
has interactedd with 3 targeets within its footprint, givving rise to three distinct eccho pulses at the receiver input. In echoo
digitizing LID
DARs, an anaalog-to-digitaal converter (A
ADC) converrts the amplifi
fied output off the APD intto a stream of
samples. Connversion ratess typically vaary between 500
5 MSPS an
nd 2 GSPS. Careful
C
signall processing prior to ADC
C
conversion alllows “linear” LIDARs to cover
c
dynamicc ranges of typ
pically more thhan 60 dB in tthe optical reg
gime.
1I
PTICAL
1
DIGITAL
ANAL(
2
2LASER
É 3> ó p,m
Sa
' II
30 MByte /sec
to
150 MByte /sec
o
I VZEE
ON
1.5 MPts/s
to
6 MPts/se
ESTI
y
SIGNAL D
sample t)lock / sample datagram for
Full Waiteform Analyses
internal ssample datagranns for
Online WJaveform Proces;sing
SIGNAL E TIMATION
1
temporalI position / range
signal strrength / amplitucie / reflectance
pulse wicith / pulse shapea deviation
backscattter coefficient o1f turbid media
Figure 1. Simplified blocck diagram of liinear LIDAR ass utilized in RIE
EGL LIDAR enngines and systeems depicting the
t optical
e
regim
me (blue), and a digital regime (black).
(
regime (reed), an analog electrical
The first taskk in processinng the data strream is signaal detection, i.e., discriminaating betweenn signal and noise.
n
In mosst
linear LIDAR
R systems dettection threshhold is set forr a very low false alarm rate. The detection threshold is typicallyy
correspondingg to about 2550 photons at a wavelengthh of 1.064 µm
m. After signaal detection oonly the samples containingg
target inform
mation are maiintained thus reducing the data rate to about
a
up to 150 Mbytes/seec. These dataa are providedd
either for offf-line waveforrm analysis [55] or for onlinne waveform processing [66]. The seconnd task is sign
nal estimationn.
Signal estimaation determinnes the tempooral position of
o the received
d echo signal giving the rannge to the target, the signaal
strength yieldding by calibrration the ampplitude of the reflected ech
ho signal in thhe optical regiime and also an
a estimate of
the reflectancce of the targeet (more precissely the laser radar cross-seection normaliized to the areea of the footp
print). In somee
instruments the
t analysis off the echo puulse shape provides an addiitional valuablle attribute foor, e.g., the classification of
the resulting point
p
cloud innto bare earth returns or veggetation return
ns.
Linear LIDAR provides 3D
D point cloudss with low rannge noise, i.e.,, high precisioon. The exampple given in Figure
F
2 showss
data from an airborne acquuisition 600 m above groundd with a RIEG
GL LMS-Q680i [7], as a pooint cloud colo
orized by echoo
amplitude.
Proc. of SPIE Vol. 9832 983204-2
Range .ogrom
---- -----
§
-0016-0014-0012 -001 -01008 -0 006 -0 004 -0 002
0
0 002
2
Distance INleasurement,Plen,
0004 0006 0008 001
0012 0
181
Accquisition param
meters:
600
0 m AGL
14 measurements//m²
nging precision:
ran
5 mm
m standard deeviation on 30 m
m² patch
Figure 2. Sample data off linear LIDAR from airborne acquisition
a
dem
monstration highh measurement precision. Leftt: Point
cloud coloorized accordinng to echo amplitude with test area
a highlighted
d in yellow. Uppper right: histoogram of residuaals to
fitting plaane.
The dataset demonstrates
d
thhe excellent ranging
r
precision of about 5 mm over a plane
p
patch of a parking lot.
The attribute from the anallysis of the eccho’s pulse shhape (pulse shaape deviation or pulse widtth) can be useed for cleaningg
up point clouuds. The laseer beam has a finite beam
m divergence and thus a fiinite footprintt size on the target. Whenn
measuring intto vegetation different partts of the laser footprint may
y interact withh different tarrgets giving riise to multiplee
echoes for a single
s
laser puulse. The rangge resolution iss limited by th
he temporal puulse width of the laser and the bandwidthh
of the receiveer. Multi-targeet resolution – the power too resolve targets illuminateed by the samee laser footpriint close-by inn
range – is in the range of 0.6 m for RIEGL instruments. In case targets are clooser than thiss range differeence, the echoo
pulses tend too merge and a single returrn will be deteected with a ranging
r
resultt somewhere iin between th
he true rangess.
These points show up in the
t final poinnt cloud of anny linear LIDA
AR. In contraast to “discrette return LIDA
AR” [4], echoo
digitization annd waveform analysis provvide the meanss to eliminate these returns based on merrged echo pulsses by filteringg
as depicted inn Figure 3.
d'
t
k
\#:
Ij
n:.
ty,,
_
..?
.T''yQ;
at;
ti
, ...
te-e,
,-. f:
141Y1
,;a.
*Qt.
_
,
.T
/
-
-
IIIMMI
A
Figure 3. Detail of a scann from a static laser
l
scan with a terrestrial laseer scan. Left: pooints highlighteed in yellow steem from
merging echo
e
pulses withh close-by targeets within the laaser footprints. Right: cleaned--up point cloudd with echoes frrom
merged puulses removed by
b simple filterring operation.
Proc. of SPIE Vol. 9832 983204-3
3. BASIC GEIGER-MODE LIDAR PROPERTIES
Geiger-Mode LIDAR (GmLIDAR) advertised for commercial surveying applications utilize not a single photodetector
as the current linear-mode LIDAR systems for airborne surveying but an array of avalanche photo diodes, typically with
32 x 128 pixels [8]. Each of the APDs is biased above the breakdown voltage so that a single photon may trigger the
APD with a certain detection probability.
In the subsequent discussion of GmLIDAR we neglect the dark count rate of any actual APD array and also photons due
to solar background radiation, i.e., we assume to have an ideal APD array operated at nighttime.
Figure 4 depicts a very simple block diagram of a GmLIDAR indicating the signals and data rates to be expected in the
various regimes ranging from optical (left in red), analog (in the middle in blue) and the digital regime (right in black).
A Geiger-mode array is usually activated sometime after the laser pulse has been emitted. The time period for which the
array is active, i.e., when the APDs are biased above breakdown voltage, is denoted as a range gate. The delay of the
start of the range gate with respect to the emitted laser pulse is set according to an a priori knowledge based on the flight
path, the scan pattern of the LIDAR system, and the terrain data has to be acquired on.
OPTICAL
ANALOG
DIGITAL
LASER
II
400 MB/sec
APD ARRAY
4096 pixels
TDCs
I
400 MB/sec
400 MB/sec
ESTIMATION
I
SIGNAL DETECTION
800 kW
inside Geiger mode APD array
changed by varying laser power
changed by varying photon detection
efficiency (PDE)
0W
1mw=
1 pW
11
1Vr
SIGNAL ESTIMATION
range gate
0Vf¡
time delay
temporal position / range
signal strength / amplitude / reflectance
pulce width / pulce chape deviation
bac I ttercocfficicnt of turbid media
Figure 4. Simplified block diagram of a Geiger-Mode LIDAR, again depicting the optical regime (red), an analog electrical
regime (blue), and a digital regime (black). Signals are examples for a single pixel of the APD array.
In contrast to linear LIDAR, GmLIDAR in its current state of development triggers each pixel of the receiver array
maximally once per laser pulse. In case the return of the first target contains some photons, the first target will most
probably trigger the pixel, and the subsequent returns from the second and third target will be lost. Instead of an ADC,
the GmLIDAR utilized an array of TDCs (time-to-digital converters) providing the time delay between the start of the
range gate and receiving the first photons. For typical values of timing resolutions (0.5 ns), range gate lengths (4 µs), and
pulse repetition rates (PRR) (50 kHz) the amount of data is about 400 MByte/sec. In the online part of the digital regime
there is, in contrast to linear LIDAR, no signal detection as this already happens already in the APD array. Signal
estimation in GmLIDAR is rudimentary as it just provides the temporal position of the trigger event and thus range, but
no information of the signal strength, i.e., the number of photons that have actually triggered the pixel, or on the pulse
width of the received echo pulse.
4. SPATIAL SAMPLING OF TARGET OBJECTS
Current state-of-the-art linear LIDAR systems used in airborne 3D surveying acquire data sequentially at high laser
PRRs of several hundred kilohertz. Depending on height above ground (above ground level, AGL) and platform speed
the measurement density typically varies between a few 100 measurements per square meter – usually addressed as
Proc. of SPIE Vol. 9832 983204-4
points per sqquare meter (pts/m2)
(
– annd just a few pts/m2. How
wever, also att very high leevels of pts/m
m2 the spatiaal
resolution is fundamentally limited by the laser foottprint size on the target [9]]. The footpriint size is sim
mply the beam
m
divergence times the rangee to target. Beeam divergennce is typically
y about 250 μrad,
μ
as statedd e.g. for the RIEGL LMSQ1560 [10]. From
F
an AGL
L of 1,000 m thhe footprint diameter is abo
out 0.25 m in nadir
n
directionn.
GmLIDAR syystems emplooying APD arrrays gather thhousands of po
otential samples of the surfface with every laser pulsee.
The laser pulse illuminatess the completee footprint of the
t APD array
y on the grounnd and the spaatial resolution
n is limited byy
the footprint of a single pixel
p
of the array
a
on the ground,
g
similaar to digital photogramme
p
try where thee resolution iss
limited by thee so-called groound samplingg distance (GS
SD) of a pixell of the cameraa array.
The GmLIDA
AR system currrently adverttised for large--scale commeercial surveys are designed tto acquire datta from 27,0000
feet at a finall post-processsed point denssity of 8 pts/m
m2 [1]. With a stated instanntaneous field--of-view (iFO
OV) of a singlee
pixel of 35 μrad,
μ
and for a stated scan angle
a
cone off 30 deg, the GSD
G
is about 0.3 m in goood agreement with 8 pts/m22
and similar too the footprintt size of a lineear LIDAR operated from 1,000 m (3,2800 ft) AGL.
Although the GmLIDAR discussed
d
heree uses an area sensor as deteector, the iFOV of the arrayy has to be scaanned over thee
FOV of the system
s
to acquuire data in a broad swath. The commerrcial GmLIDA
AR is advertissed to scan thee area withouut
any scan shadows due to a 360 deg loook on all targgets [11]. Figu
ure 5 shows the
t scan patteern of a dual channel
c
linearr
LIDAR systeem, the RIEGL
L LMS-Q15600 on the left. The
T FOV is 60
6 deg. Each channel
c
covers the full FOV
V by a straighht
scan line and the two channels are tiltedd by 28 deg wiith respect to each other. Inn the center off the swath thee two channelss
are nadir looking, whereass at the edgess of the swathh the angle off incidence iss about 30 degg with a forw
ward/backwardd
component off about 8 deg.. The scan patttern of the GmLIDAR is shown
s
on the right. The FO
OV is 30 deg. In nadir, theree
is a forward/bbackward lookk of +/- 15 deg. At the edgees of the swath
h there is a strrict 15 deg sidde look withou
ut any angularr
component allong the flightt path.
:enter
C
of
swath
edge
ce!nter of
averlap ---p 50%
C
60 deg
30 deg
Figure 5. Illustration of angle
a
of incidennce of measurem
ment beam on targets
t
within a single swath. L
Left: Scan patteern of a
AR, RIEGL LM
MS-Q1560. Righht: Scan pattern
n of GmLIDAR
R with palmer sccanner.
dual channnel linear LIDA
In case a target area is coovered with 50%
5
side overlap of the sw
waths, a targeet in the centter of the oveerlap region iss
covered exacctly 4 times with
w 4 differennt angles of incidence.
i
Th
hus there is noo principle diifference in prroducing scann
shadows or data
d gaps if theere are e.g. buuildings on thee ground between the two syystems discussed.
Proc. of SPIE Vol. 9832 983204-5
5. WAVEFORM INFORMATION AND MULTI-LOOK PROCESSING
The key to the best possible multi-target resolution and thus penetration of vegetation in airborne laser scanning with
linear LIDAR is echo digitization and subsequent full waveform analysis. Figure 6 shows an example waveform when
measuring into dense vegetation. The first pulse/peak may represent the top of the canopy, the last peak bare earth and
the intermediate peaks the different layers in the vegetation.
25
Laser pulse
20 - Gaussian model
15
10
5
0
534
535
536
537
538
539
540
541
542
Figure 6. Example waveform from a RIEGL LMS-Q680i when measuring into dens vegetation. Dots represent the samples
from echo digitization, the solid curve the reconstruction by Gaussian decomposition.
It seems difficult to penetrate vegetation with a detector which triggers only once per laser pulse at the first few photons
arriving at the detector. An approach to gain penetration capability is to adjust the detection in the APD array in a way
that for every single illumination the detection probability for a single target return is so low, that there remains a nonzero detection probability for all subsequent targets [12]. In order to achieve a high detection probability for every target
the scene has to be illuminated numerous times, i.e., to have numerous looks (multi-look) onto the same spot on the
ground. This can be easily achieved for a stationary GmLIDAR system looking in the same direction in space all the
time. In doing so, the temporal integration over all events gives a histogram of detections over range, which resembles a
waveform from a linear LIDAR system from a single acquisition. However, for a kinematic acquisition from a fast
moving airborne platform having numerous illuminations of the same spot imposes a severe challenge. For the
commercial GmLIDAR system it is claimed that every spot on the ground is illuminated hundreds of times [1].
Subsequently, we derive the number of looks of a commercial LIDAR system for the advertised acquisition parameters
as summarized in Table 1.
Table 1. Parameters published for a commercial GmLIDAR to acquire 8 pts/m² [1].
above ground level
platform speed
FOV of scanner
AGL
v
α
27,000 ft
290 kts
30 deg
FOV of single pixel
array dimensions
laser repetition rate
iFOV
PRR
H
D2
d1=
35 µrad
32 x 128
50 kHz
D1=
rps
L = 27r AGL tan
PRR
cos(a / 2)
D2-
D1
vf = L rps
d2- of
AGL
N`°°k,
AGL
Nrooks =
(a)
V
iFOV
AGL
cos(a / 2)
128
cos(a / 2)
iFOV.32
Dl D2
=
zFOV z
dl d2
4096PRR
7r sin(a).cos(a/2)
(b)
Figure 7. Deriving the number of looks for an airborne GmLIDAR with a Palmer scan. For symbols refer to Table 1 and
text.
Proc. of SPIE Vol. 9832 983204-6
Figure 7(a) depicts the scan path on the ground of the GmLIDAR system employing a Palmer scanner (blue line). For a
platform speed of v and a given number of rotations of the scanner per second (rps) the platform moves by d1 within a
single rotation of the scanner. The length of the scan line, L, on the ground is easily calculated from the FOV of the
scanner and the AGL. The speed of the footprint on the ground (vf) with the pulse repetition rate of the laser, PRR, gives
the displacement of the footprint between two looks, d2.
The size of the footprint (projection of the APD array) on the ground is also calculated easily as shown in Figure 7(b).
D1 and D2 denote the size across the scan line and along the scan line in the center of the swath. The number of looks of
a spot on the ground in the center of the swath within a single line is given by D2/d2, the number looks of consecutive
lines is D1/d1. The formula derived for the number of looks, Nlooks, reveals that Nlooks is independent of the rotational
speed of the scanner, rps, as long there is overlap of footprints along the scan line and from scan line to scan line. The
number of looks can be changed by the operator only by changing the acquisition parameters AGL and v, as all the other
parameters are fixed by system design. For the advertised acquisition parameters of Table 1 the number of looks is just
about 10 but not hundreds of looks as claimed. The number of looks as a function of AGL is shown in Figure 8 on the
left. On the right the number of looks is given as a function of the scan angle. In the center of the overlap of two
neighboring scan swaths at 50% side overlap, the number of looks is 12 compared to 10 in the center of the swath.
1.03
45
speed over ground 290 knt
40
- iFOV = 35mrad
- iFOV = 70prad
- iFOV = 150Arad
35
F. 30
E. 102
Y '2
o 25
o
2
Ó
o
á, 20
a
101
c
15
N = 12
10
50% overlap
5
1000
0
5000
15000 20000
above ground level (ft]
10000
25000
30000
o
50
100
200
250
scan angle [deg]
150
300
350
400
Figure 8. Left: number of looks as a function of AGL and the system design parameter iFOV for a platform speed of
290 kts. Right: number of looks versus the scan angle (0 deg and 180 deg in the center of the swath, 90 deg and 270 deg the
edges of the swath).
6. DETECTION PROBABILITY AND PENETRATION OF VEGETATION
The detection probability for both systems, GmLIDAR and linear LIDAR strongly depends on the number of photons
received within a single echo pulse. Subsequently, we estimate the number of photons received from a white diffusely
reflecting target for the acquisition example advertised to achieve 8 pts/m2, i.e, AGL 27,000 ft and 290 kts platform
speed for the GmLIDAR and AGL 3,280 ft and 117 kts for the linear LIDAR system [10] taking into account the system
parameters like laser power, laser pulse repetition rate, receiver aperture and assuming a visibility of 23 km by making
use of the LIDAR equation. For the estimation for the GmLIDAR published system parameters have been used [1, 13].
Table 2 summarizes the parameters used.
Table 2. System parameters used to estimate the number of received photons.
above ground level
platform speed
laser repetition rate
average laser power
laser wavelength
receiver aperture
receiving elements
Linear LIDAR
3,280 ft
117 kts
2 x 400 kHz
2 x 10 W
1064 nm
2 x 42 mm
2x1
Proc. of SPIE Vol. 9832 983204-7
Geiger-mode LIDAR
27,000 ft
290 kts
50 kHz
20 W
1064 nm
250 mm
4096
The average number of phhotons for a single
s
pixel of
o the GmLID
DAR system is
i about 10, ffor the linear LIDAR abouut
44,000.
The detectionn probability of
o a linear LID
DAR receiverr an be modelled as a threshhold receiver ttaking into acccount thermaal
noise at the receiver inpuut, excess noisse of the APD
D, shot noisee, and specklee noise. The detection threeshold for thee
RIEGL LMS-Q1560 is aboout 250 photoons. Figure 9 shows
s
the deteection probabbility as a funcction of the av
verage numberr
of photons received in an echo
e
pulse.
iL
:
o
1-
o
:
N
|
1-
detection probabilit
/
i5
0
150l
200
250
300
350
averEige number of photons
Figure 9. Detection probability of the linnear LIDAR syystem RIEGL LMS
L
Q1560 as a function of thhe average numb
ber of
photons reeceived in an eccho pulse.
The detectionn probability, PD, for a Geiger-mode recceiver with a given photon detection effi
ficiency for a single photonn,
PDE, and for a given deterrministic numbber of photonss in a pulse, N,
N is
PD (N) = 1 - (1 - PDE))N
(1)
as the probabbility that any of N photons triggers is com
mplementary to
t that that noone of the N phhotons triggerrs.
A Poisson prrocess describes the actual number of phhotons receiveed within a puulse. The probbability mass function, pmff,
for detecting N photons forr a given averaage number of photons, Navvg, is
!
(2)
f a given PD
DE and a given average nu
umber of phottons per pulsee is given by the
t product of
The detectionn probability for
the probabilitty mass functtion and the detection probability for a deterministicc number of photons, integrated overalll
number of phhotons, as illusstrated in Figuure 9.
Proc. of SPIE Vol. 9832 983204-8
n«KOao o,aweiirN
2
06
X
02
0
02
o
10
a
os
0.
02
o
.1 0.05
avg. num
V photons:
12
P
0e
14
P
a0 :s
o
*balm protaty M N photon o Klo On
10
0.30
x
N.
6
00
00
o
0.00
6
umber of photons I echo pulse
number o photons m en
o
6
8
of photons In echo pulse
Figure 10. Probability mass
m function forr receiving a ceertain number of photons for a given average nnumber of phottons in a
pulse (leftt), detection proobability for N photons
p
in a puulse for PDE = 0.3
0 (center), ressulting pmf for an average num
mber of
photons of
o 2.2.
∑
!
1
1
(3)
Figure 10 shows the detecction probabillity for a Geiiger-mode recceiver with PD
DE = 0.3 verrsus the averaage number of
photons in ann echo pulse. For
F 90% detecction probabiliity an averagee number of phhotons of about 8 are requirred.
10°
:
E
al
o
L
Q 10-1
c
,
o
u
ti
',
1
10 -2
10
10°
101
e number of ph,otons in echo piulse
Figure 11. Detection probability for a Geiger-mode
G
recceiver with PDE
E = 0.3 versus the
t average num
mber of photonss in an
echo pulsee.
In order to demonstrate
d
a LIDAR’s caapability to penetrate
p
vegeetation in an airborne dataa acquisition we assume a
scenario as sketched in Fiigure 11. Thee iFOV of a pixel
p
of an AP
PD array “seees” not a singgle target but two layers of
vegetation, which
w
might giive rise to targget returns 1 and
a 2, and also bare earth, the
t potential tthird target. Not
N untypicallyy
for dense forrests in the leaave-on condittion in North America, we assume a traansparency of the canopy with
w respect too
ground of 5%
%. Ground refflectance is assumed to bee 40% at the laser
l
wavelenngth. Fill factoors for the tw
wo layers withh
according refflectance values are also staated in Figuree 11. The averrage number of
o photons to expect from the
t acquisitionn
parameters annd system parameters as staated in the tables above are 3.2, 1.5, and 0.2.
0
Proc. of SPIE Vol. 9832 983204-9
Accordingly, for the lineaar LIDAR RIE
EGL LMS-Q11560 for the acquisition
a
paarameters statted in the tables above, thee
number of phhotons from thhe three targeets amount to 14,000, 6,500, and 900 reespectively. A
As the detectio
on threshold iss
about 250 phootons, all threee targets are detected
d
with almost 100% certainty.
h iFOV
section throu
of a single pix
e 50%
rage
,5 %, reflectanI
ce 50%
photons on av
verage
30 %, reflectan
Figure 12. : Scene used to
t demonstrate the
t capabilities of GmLIDAR and linear LID
DAR to penetratte vegetation. Two layers
t
returns. The
T insert on thhe left gives the fill factors, thee
of vegetattion and bare eaarth are assumed to provide 3 target
reflectancce values and avverage number of photons for the
t GmLIDAR example descrribed above.
Making use of the equattions above and
a
of the faact that the Geiger-mode
G
array can onnly trigger once
o
per laserr
illumination, the probabilitty of detectioon for the threee targets can be calculatedd as a functionn of the averaage number of
photons to bee expected forr a white diffu
fusely reflectinng target at th
he same rangee. The detectioon probability
y for detectingg
the second tarrget is the proobability for detecting the taarget in the ab
bsence of targeet 1 times the probability th
hat target 1 hass
not triggered the detector. Similarly for detecting targget 3, neither target
t
1 nor tarrget 2 have too trigger the deetector. Figuree
gets 2 and 3.
13 displays the detectionn probabilitiess. There are flat maxima for the detecction probabiilities for targ
Increasing thhe average nuumber of phootons, e.g., by
b changing system
s
param
meters or assuuming a high
h atmosphericc
visibility, thee detection proobability dropps due to the fact, that it beecomes more likely that allready target 1 triggers thuss
obscuring thee subsequent targets.
Proc. of SPIE Vol. 9832 983204-10
-
'
1st
2nd
Rrd
10 -1
I
i
103
o
10
5
20
15
age number of F)hotons per pixe I at p = 100% arid
f = 100%
Figure 13. Probability off detection for thhe three targetss (first layer of canopy
c
(blue), second
s
layer off canopy (green
n), and
xpected for a whhite diffusely reeflecting target at the
ground (reed)) as a functioon of the averagge number of phhotons to be ex
same rangge.
In order to deemonstrate thee penetration of
o vegetation of
o GmLIDAR
R and linear LIIDAR visuallyy, we simulatee the detectionn
of an object hidden
h
beneatth a syntheticc dense canopy with homog
geneous transsparency of 5%
%. The objectt consists of 6
letters with a letter height of
o about 2.5 m.
m The letters are modeled as 3D objectss with a depthh of more than
n 1 m arrangedd
over flat horiizontal area. Figure
F
14 (left
ft) shows the object
o
color-coded with resspect to heighht above groun
nd. The centerr
image represeents the simullation result foor the GmLID
DAR. White arreas indicate that
t no data haave been received; blue andd
red rectangles represent prrojections of the
t iFOVs of pixels
p
of the APD
A
array inn case they havve triggered. Blue
B
indicatess
results obtainned from the ground level, red from thee elevated lev
vel according to the top off the object. The
T simulationn
result is compposed of 10 loooks to the taarget area withh a detection probability
p
off 1.5% for thee ground targeet according too
Figure 13. Due to the irreegular samplinng, the low nuumber of look
ks, and the loow detection pprobability fo
or target 3, thee
object cannott be recognizeed or deciphereed.
Figure 14 (right) shows thhe simulation result for thee linear LIDA
AR system. Due to the reguular sampling
g and the highh
detection proobability for the
t linear LID
DAR the objeect can easily
y be recognized as 6 letterrs and the meessage can bee
deciphered.
a
f
..``. f
;
. I[ñlrrif. ..
a
4,4
.
.
t'2
o-o
o
'1'i
}
.
.
°! I.
.
u
o
¡
u
"
0
,.0,_
'' r::
:
f
'
t
.'
o
-. 1-
i
° ,4
i. i¡
S
-
2 6U6. ' ... - Oao
N
_
Figure 14. Left: object inn perspective viiew hidden beneeath a canopy for
f simulation. Center:
C
simulatted results of Gm
mLIDAR.
mulated results for
f linear LIDA
AR.
Right: sim
Proc. of SPIE Vol. 9832 983204-11
NEOUS SU
URVEYING
G OF LOW
W-CROSS-S
SECTION O
OBJECTS AND
7. SIIMULTAN
TERRAIN BENEATH
B
H THE CAN
NOPY
In commerciaal airborne suurveying the final user of the data is not only intereested in the ddigital elevatio
on model, buut
frequently to also survey thhe complete innventory of man-made
m
strucctures like pow
wer lines, com
mmunication lines,
l
or poless.
For example, wires have a significantly lower laser raadar cross-section comparedd to flat diffussely reflecting
g targets largerr
than the laserr footprint or the
t iFOV of a single pixel of
o the GmLIDAR on the groound.
diameter
vire diame
l4
I
1/e2
AOMDONAOOD
000000000
i--,--,--,--,--,--,--^,--
010000000000
photon count
dicnint-tar
icnImrnmc
l.,r11.......11111 IIV
4
2
o
8
_detection
6
- threshold
4
2
0
-0.3 -0.2 -f
'
1.3 -0.2 -0.]
displa
disp
Figure 15. Sketch of a wire target (1 cm
m diameter, 40%
% reflectance) in
nteracting with the laser footprrint of a linear LIDAR
L
(left) and the image of thhe iFOV of a Gm
mLIDAR (rightt) for the acquissition example as advertised foor 8pts/m2. The diagrams
on the botttom give the nuumber of photoons versus the displacement
d
off the wire withinn the footprint.
Figure 15 shoows a sketch of a wire targget interactingg with the laseer footprint off a linear LID
DAR from AG
GL 1,000 m onn
the left, and with
w the imagge of a single pixel of the Geiger-mode
G
APD
A
array froom 27,000ft oon the right. The
T number of
photons receiived by the linnear LIDAR varies
v
with thee displacemen
nt as shown inn the diagram on the left. The
T example iss
given for a wire
w with a diameter
d
of 1 cm and a diiffuse reflectaance of 40%. The wire wiill be detected
d with a highh
probability foor a displacem
ment of ±0.12 m. The averaage number off photons from
m a 1 cm wiree with 40% refflectance for a
single pixel of
o the GmLID
DAR is about 0.1 as shown in Figure 15 on the right. The correspoonding detection probabilityy
would be aboout 3% (compaare Figure 11)).
In order to detect a wire target
t
with a higher
h
probabbility, the imaage size of a pixel
p
on the oobject has to be
b reduced byy
p
deteccted by a sing
gle pixel of a
lowering the height abovee ground. Figgure 16 shows the averagee number of photons
GmLIDAR on
o wire targetts with 1 cm (green)
(
and 2 cm (blue) width
w
and on an
a extended w
white diffuse reflector (red)
versus the heiight above groound for a Gm
mLIDAR systeem with system
m parameters summarized in Table 2.
Proc. of SPIE Vol. 9832 983204-12
2 cm
1 cm
--,
o
- p=1.O fill =10 0%
detection probatliqty:r
e10%:
1
oo
average number of photons [1]
- p=0.4diam=
- p=0.4diam=
10
15
abovei ground
25
20
level [1 000
30
ft
Figure 16. Average numbber of photons detected by a siingle pixel of a GmLIDAR onn wire targets w
with 1 cm (green
n) and 2
e
white diffuse reflectoor (red) versus the
t height above ground.
cm (blue) width and on extended
1 st
I
I
- 2:nd
- 3Ird
I
OI
y
fD
<
DJ
F+
probablity of detection for each target
F+
F+
F+
F+
Y F+ Y F+ N
6
6
J
6
n
Á
W
O
7H
F+
O O O O O O O O O O OY
In order to detect
d
wire taargets with a high probabiility, e.g. 80%
%, about 6 phhotons on aveerage are req
quired. This iss
achieved withh the GmLIDA
AR by reduciing the height above ground
d to about 11,,500 ft and 133,800 ft for th
he 1-cm and 2cm wire diam
meter targets, respectively. However, att these heights, the numberr of photons received from
m an extendedd
diffusely refllecting white target amounnt to about 2223 and 132, respectively. These averagge numbers of
o photons aree
indicated as doted
d
vertical lines in Figure 17. The deetection probab
bilities for thee two layers oof vegetation and
a bare earthh
from the exam
mple defined in Figure 12 are
a now displlayed similar to
t Figure 13, but for an exttended range in the averagee
number of phhotons from thhe white extennded target. The detection probability
p
forr the ground iis well below 10-8 in case a
2-cm wire is detected withh better than 80%,
8
and welll below 10-12
2 for the 1-cm
m wire. This im
mplies that it is
i not possiblee
to survey low
w-cross-section targets like wires and thee ground beneeath vegetatioon with a Gm
mLIDAR systeem, as a closerr
inspection off published datta of GmLIDA
AR reveals [1]].
50
100
150
200
lumber of photcins per pixel at p = 100% and
250
f = 100%
Figure 17. Detection probabilities for thhe two layers off vegetation and
d bare earth from
m the example defined in Figu
ure 12. The
w below 10-8 in case a 2-cm wire is detectedd with better thhan 80%, and well
w below
detection probability for the ground is well
10-12 for thhe 1-cm wire.
Proc. of SPIE Vol. 9832 983204-13
For the linear LIDAR RIEGL LMS-Q1560 simultaneous acquisition of data on wire targets and on the ground beneath
vegetation imposes no challenge, as shown by the example data in Figure 18, showing a perspective view of the point
cloud with color-coding according to height on the left, and a cross section through the data on the right (wires of the
power line show up as dots in the right hand side of the image).
Figure 18. Perspective view of data acquired with RIEGL LMS-Q1560 on two power lines and group of trees (left) and a
cross-section through the same data set (right) demonstrating the capability to acquire data on low-cross-section targets like
wires and ground beneath vegetation simultaneously.
8. MEASUREMENT ACCURACY AND MEASUREMENT NOISE
Measurement accuracy and measurement noise of a LIDAR system are determined by a large number of phenomena,
like noise within the receiver, background noise, shot noise of the optical signal itself, trigger walk due to the finite
bandwidth of the laser pulse and the receiver, beam walk due to atmospheric turbulence and inhomogeneity to name just
a few. Specific error sources in LIDAR are ranging noise and systematic ranging error. In echo-digitizing linear LIDAR
systems, ranging is done by estimating the temporal position of a received echo signal, which in turn is based on digital
signal processing schemes. This proves to give a very low change of the estimated range versus the signal strength of the
echo signal over a very wide dynamic range and also a low ranging noise. Typically for these airborne linear LIDAR
systems the trigger walk and the range measurement noise is about 20 mm. In non-echo-digitizing linear LIDAR
systems, so-called discrete return systems, the inherent trigger walk imposed by the detection scheme is compensated for
by estimating the amplitude of the echo signal and to correct for by adding a correction value from a look-up table.
In GmLIDAR systems an additional effect contributes to a significant ranging error on flat tilted surfaces: as the number
of photons per return echo is quite low, the temporal position of the photon actually triggering the pixel of APD array is
unknown. Thus a large range noise is to be observed. Furthermore, as the arrival times of the photons follow a Poisson
process and the detector triggers just once per laser pulse, the early photons have higher detection probabilities than the
ones arriving later, giving rise to a systematic range error. In contrast to discrete-return linear LIDARs, GmLIDAR does
not provide any information on signal strength, thus systematic trigger walk cannot be compensated for at all.
Proc. of SPIE Vol. 9832 983204-14
%00
_x3
d
-10
-p5
,e°o im
Figure 19. Impact of rangging to a vertical flat target wiith a linear LIDAR from AGL 1,000 m (left) and with a GmL
LIDAR
r
d v,
from 27,0000 ft (right), booth with an off--nadir scan anglle of 15 deg. Beeam diameter annd pixel image size d, height range
and variattion of range Δrrv.
Figure 19 deemonstrates thhe impact of measuring too a flat verticcal surface onn ranging noiise and rangiing error. Thee
example takees the same accquisition paraameters for thhe linear LIDA
AR (left) and the GmLIDA
AR (right) as summarized
s
inn
Tables 1 and 2. For an off--axis scan anggle of 15 deg, the variation in height andd range withinn the footprint is 1.16 m andd
1.12 m, respeectively. The insert for the linear LIDAR
R shows the waveform
w
for a flat target aat normal inciidence in bluee
and the waveeform for the slanting targeet of the exam
mple in red. Th
he width of thhe waveform increases, butt the center of
the waveform
m remains neearly at the same
s
temporaal position indicating that the systemattic error to be
b expected iss
negligible. For the GmLIIDAR Figure 19 shows thhree simulatio
on results for the temporral distribution
n of photonss,
assuming 2.5 photons to bee received on average. The red stems ind
dicate the phottons triggeringg the detector pixel.
Figure 20 deepicts ranging noise (left) and
a systematiic ranging errror (right) for the RIEGL L
LMS-Q1560 linear LIDAR
R
(green) and a GmLIDAR (blue) on a vertical flat tarrget at an ang
gle of incidencce of 75 deg derived by simulation. Thee
signal strengtths of the echho signal havve been norm
malized to a detection
d
probbability of 80%. The rangee noise of thee
GmLIDAR system
s
is givven for a singgle look and may be red
duced by spattial averagingg in multi-loo
ok acquisitionn
scenarios.
Proc. of SPIE Vol. 9832 983204-15
o
0.30
1A
0.35
c
á 0.15
ó
A
!J
o
Á
41
0.20
o
ÿE
systematic error [m]
0.25
-s
;+
0
5
10
o
-10
ó
io
V
0.00
..........
o
0.05
o
0.10
-5
return Ignel strength Idl
0
5
1
retunn signal strength
Figure 20. Range noise (left)
(
and system
matic range erroor (right) for thee RIEGL LMS--Q1560 linear L
LIDAR (green) and a
AR (blue) on a vertical
v
flat target at an angle of
o incidence of 75
7 deg derived by simulation. As the systemaatic range
GmLIDA
error for the
t linear LIDA
AR is negligible, the specified accuracy
a
of 2 cm
m is given for the
t RIEGL LM
MS-Q1560.
9. SUMMAR
RY
Properties annd performancce of linear LIDAR
L
and Geiger Mode LIDAR
L
have been
b
analyzedd and comparred in view of
acquisition sppeed, oblique viewing angles, vegetation penetration, surveying of
o man-made structures likee power liness,
and range nooise and systeematic error. Vegetation
V
peenetration for generation of
o digital elevation models is limited forr
GmLIDAR systems,
s
especially when flying
f
low inn order to inccrease the chhance to captuure wires. In this case thee
probability off pick up echooes from the ground
g
undernneath the cano
opy becomes extremely
e
low
w. From high altitudes
a
whenn
exploiting thee merits of GM
MLIDAR systtems, measureements to pow
wer lines will show
s
a detectiion probabilitty of 3% whilee
the results onn targets subm
merged by veggetation will be
b improved but
b still compaaratively sparse. Due to thee nature of thee
detection process the rangee noise and syystematic error on vertical flat
f targets aree considerablyy worse for Gm
mLIDAR thann
for linear LID
DAR.
R
REFERENCE
ES
[1] Rhoads, R., “Geiiger-mode LiDAR mappinng: High density, high voolume airbornne 3D imagin
ng”, Capturingg
Reallity 2015, Ausstria
[2] RIEG
GL 2016a, ww
ww.riegl.com
[3] Ullriich, A., Reichhert, R., Riegl, J.,: “High resolution
r
laseer scanner wiith waveform digitization for
f subsequennt
full waveform
w
anaalysis”, Proceeedings of SPIE
E Vol. 5791, pp.82-88,
p
March 28th – Aprril 1st 2005, Orlando
O
[4] Ullriich A., Pfennnigbauer, M.: “Echo Digitiization and Waveform
W
Anaalysis in Airbborne and Terrrestrial Laserr
Scannning” Proceeedings of the 53
5 rd Photogram
mmetric Week
k, September 2011,
2
Stuttgarrt
[5] Waggner, W., Ullrrich, A., Ducic, V., Melzer, T., Studnick
ka, N.: “Gausssian Decompoosition and Caalibration of a
Novvel Small-Foottprint Full-Waaveform Digittising Airborn
ne Laser Scannner”, ISPRS JJournal of Ph
hotogrammetryy
& Remote
R
Sensing, 60 (2006), pp. 100 – 1122, 2006
[6] Pfennnigbauer M.,, Ullrich, A.: “Three-dimeensional laserr scanners with echo digittization”, Voll. 6950: Laserr
Radaar Technologyy and Applicaations XIII, 166 April 2008
[7] RIEG
GL
2016aa,
http://w
www.riegl.com
m/uploads/tx_p
pxpriegldownlloads/10_DataaSheet_LMS-Q680i_28-0920122.pdf , retrieveed March 2016
[8] Cliftton, W.E, ett al.: “Mediuum Altitude Airborne Geiger-mode Mapping
M
Liddar System”, Laser Radarr
Techhnology and Applications
A
X Proc. of SPIE
XX,
S
Vol. 946
65, SPIE, 20155
Proc. of SPIE Vol. 9832 983204-16
[9] Ullrich, A. Sampling the World in 3D by Airborne LIDAR – Assessing the Information Content of LIDAR
Point Clouds, Photogrammetric Week, Stuttgart, 2013.
[10] [RIEGL 2016a, http://www.riegl.com/uploads/tx_pxpriegldownloads/DataSheet_LMS-Q1560_2015-03-19.pdf ,
retrieved March 2016]
[11] Romano, M.: “A New Industry Standard: Commercial Geiger-mode LIDAR”, LIDAR News webinar, March
24th, 2015
[12] Fried, D.G.: “Fast, Cost-Efficient Airborne 3D Imaging With Geiger-mode Detector Arrays”, MIT RLE &
3DEO, Inc. ILMF 2015, February 23, 2015, Denver, CO
[13] Clifton, W.E, et al.: “Medium Altitude Airborne Geiger-mode Mapping Lidar System”, Laser Radar
Technology and Applications XX, Proc. of SPIE Vol. 9465, SPIE, 2015
Proc. of SPIE Vol. 9832 983204-17