Journal of Transportation Technologies, 2014, 4, 44-62
Published Online January 2014 (
Measur in g Light an d Ge o me t ry D a ta of Ro adw a y
Enviro nments with a Camera
Hongyi Cai*, Linjie Li
Department of Civil, Environmental & Architectural Engineering, The University of Kansas, Lawrence, USA
Email: *hyc
Received September 23, 2013; revised October 27; 2013; accepted November 21, 2013
Copyright © 2014 Hongyi Cai, Linjie Li. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. In accor-
dance of the Creative Commons Attribution License all Copyrights © 2014 are reserved for SCIRP and the owner of the intellectual
property Hongyi Cai, Linjie Li. All Copyright © 2014 are guarded by law and by SCIRP as a guardian.
Evaluation of the conspicuity of roadway environments for their environmental impact on driving performance
is vital for roadway safety. Existing meters and tools for roadway measurements cannot record light and geome-
try data simultaneously in a high resolution. This study introduced a new method that adopted recently devel-
oped high dynamic range (HDR) photogrammetry to measure the luminance and XYZ coordinates of millions of
points across a road scene with the same devicea camera, and a MatLab c od e for d ata tre atm e n t and vi s u aliz ation.
To validate this method, the roadway environments of a straight and flat section of Jayhawk Boulevard (482.8 m
long) at Lawrence, KS and a roundabout (1 5.3 m in diameter) at its end were measured under c lear and cloudy sky
in the daytime and at nighttime with dry and wet pavements. Eight HDR images of the roadway environments under
different viewing cond itions were g enerate d using the H DR photo grammetric t echniqu es and cali brated. Fro m each
HDR image, synchronous light and geometry data were extracted in Radiance and further analyzed to identify po-
tential roadway environmental hazards using the MatLab code ( .htm l ). The
HDR photogrammetric measurement with current equipment had a margin of errors for geometry measurement
that varied with the measuring distance, averagely 23.1% - 27.5% for the Jayhawk Boulevard and 9.3% - 16.2%
for the roundabout. The accuracy of luminance measurement was proven in the literature as averagely 1.5% -
10.1%. The camera-aided measurement is fast, non-contact, n on-destructive, and off the road, thus, it is deemed
more efficient and safer than conventional ways using meters and tools. The HDR photogrammetric techniques
with current equipment still need improvements on accuracy and speed of the data treatment.
Measurement; Geometry; Light; Roadway Environment; High Dynamic Range Photogrammetry
1. Introduction
The conspicuity of roadway environments is affected by
many factors, including the transitory lighting and
weather conditions, the geometry and layout of roadway
elements, the legibility of roadway signs and plaques,
and the visibility of traffic control signals, pavement
markings and markers, object markers, delineators,
channelizing devices, and pedestrian hybrid beacons, etc.
As a result, the conspicuity of roadway environments
varies over time under different viewing conditions, e.g.,
in the daytime or at nighttime; in sunny, rainy, foggy, or
snowy weathers; wi t h dr y or wet pavements, etc.
When the conspicuity of essential roadway environ-
ments (e.g., intersections) is harmed at nighttime or even
in the daytime under less-than -optimal viewing condi-
tions, roadway lighting is often needed to increase the
visibility of the pavement, traffic control devices, and
on-road/off-road hazards to facilitate drivers conducting
driving tasks safely. However, roadway lighting with low
quality may provide insufficient light, cast shadows on
the road pavement, or bring discomfort glare to motorists
and pedestrians, which all reduce the visibility of the
roadway environments.
Therefore, evaluation of the roadway environments is
vital for their environmental impact on driving perfor-
mance and roadway safety. Routine measurements of the
*Corresponding a uthor.
H. Y. CAI, L. J. LI
roadway environments could help to maintain a good
conspicuity of essential roadway elements, and also id en-
tify potential roadway hazards (e.g., illegible roadway
signs, insufficient light on pavement, glaring roadway
lighting, etc.) that may jeopardize the roadway’s visibili-
ty and the drivers’ visual comfort. Both ligh t and geome-
try data of the roadway environments need to be meas-
ured simultaneously.
What light and geometry data need to be measured in
the roadway environments? Lighting metrics measured
on the road often include surface illuminance and lu-
minance, their spatial variations, luminance contrast be-
tween a target and its background, etc. The commonly
measured geometries of the roadway environments in-
clude size and shape of the pavement, sidewalks, road-
way markings and markers, and surrounding objects and
their distances, e tc . In addition, the acquired light and
geometry data, if concurrent, are used for derivation of
some advanced metrics for timely roadway environment
assessment. Such metrics with which the light and geo-
metry data are jointly involved include visibility of ob-
jects on the road, legibility of roadway signs, discomfort
glare of roadway lights, roadway surface reflectance,
uniformity of pavement light distribution, and luminance
gradient (the frequency of the spatial lu minance variation)
[1] across the roadway environment, etc.
How to measure light and geometry data in the road-
way environments? Conventionally, meters and tools
have been used. Minolta LS-100 luminance meter, Mi-
nolta T-10 lux meter, and Minolta CL-200A chroma me-
ter are popular handheld light meters. Handheld retroref-
lectometers, such as Delta LTL-X, Mirolux Ultra 30 , can
provide traceable, accurate, repeatable and reproducible
retroreflectivity measurements. For measuring distances
and angles, rulers, tapes, wheel rolls, laser distance me-
ters, laser rangefinders, protractors, and magnetic angle
locators have been used. The measurement using meters
and tools is a reliable but tedious point-by-point process
with very low measurement resolution (number of meas-
ured points per surface area). The dynamic daylight,
transient weather, moving traffic, and other changing
viewing conditions on road all affect the accuracy of
overall roadway measurement using meters and tools that
take hours or even longer [2,3]. In addition, it is danger-
ous for workers to conduct the measurement on road with
active traffic flow.
To solve these problems, remote sensing technologies
are preferred for fast, non-contact, non-destructive and
off-road measurement. This paper introduced a new
camera-aided method for measuring millions of syn-
chronous light and geometry data of the roadway envi-
ronments. To validate this method, the roadway envi-
ronments on Jayhawk Boulevard in Lawrence, KS and a
roundabout at its end were measured under clear and
cloudy sky in the daytime and also at nighttime with dry
and wet pavements.
2. Literature Review
In the past 20 years, several non-contact measurement
techniques were developed for measuring the roadway
geometry or light. They are reviewed below.
2.1. Roadway Geometry Measurement
Drakopoulos and Ornek developed a non-contact method
for measuring roadway geometry [4]. A vehicle was used
to carry a distance-measuring device, a vertical gyros-
cope, and a gyrocompass. Data collected include odome-
ter reading, gradient, transverse slope, and compass
reading. From these data, an algorithm was developed for
calculating the roadway geometry, including road length,
deflection angle, and degree of curves. The conventional
inertial data logging and advanced GPS (global position-
ing system) data logging were used in their study, both
had low accuracy. Inertial hardware has acceleration ef-
fect on readings, which needs frequent calibration. Al-
though the GPS technology has no acceleration effect,
atmospheric disturbances, receiver errors and other fac-
tors reduce the accuracy of GPS devices. Due to these
problems, the measurement errors of this algorithm were
relatively large, up to 63.8%.
Tsai et al. developed another algorithm for roadway
geometry computation based on video logs [5-8]. Pave-
ment images were collected using a camera fixed on a
moving vehicle. This algorithm used vanishing points
and vanishing lines to calculate the projection metric
from the 3D world coordinate system (WCS) to the 2D
camera coordinate system. Vanishing points and lines
were calculated using orthogonal edges of rectangular
pavement markings in the images. Next, roadway geo-
metries such as line width, roadway length, lane width
and marking size were extracted. The measurement error
was within 0.01 - 0.3 m, depending on the measuring
distance. An advantage of using vanishing points and
vanishing lines is that this algorithm does not need any
intrinsic parameters of the camera, such as the focal
length and the focal point [9-11]. On the other hand, this
method has a limitation that the camera must capture
rectangular pavement markings or other objects with
orthogona l edges in th e picture f or 2D/3D recons truc tion.
If a road section is not linear, e.g., a horizontal curve or a
roundabout, and doesn’t contain any rectangular pave-
ment markings, this algorithm is not applicable. Fur-
thermore, using this method, the pavement markings
were segmented based on colors and length [7]. As a
result, only pavement markings with specific colors and
certain lengths were extracted. Other objects on the road
or on roadside were left untreated.
H. Y. CAI, L. J. LI
GPS was also used for measuring roadway geometry.
Nehate and Rys used GPS data to calculate stopping-
sight dis t a nc e of hi ghways [12]. A profili n g va n c ol lected
GPS raw data such as latitude, longitude, and altitude of
road surface and sight obstructions, which were trans-
formed into B-spine curves. Drivers’ sight line inter-
sected with these two curves. By examining the intersec-
tions, the sight distance was derived. The measurement
error was ±0.2 m. This algorithm was not capable of
showing the actual image of sight obstructions and de-
tailed information of the road surface. Castro et al. used a
Leica System 500 GPS to track the geometry of highway
system [13]. The GPS were mounted on a vehicle to
measure the coordinates when the vehicle was moving
along the highway, with maximum err or of 1 m. Koc also
used GPS for rail-track surveying [14]. He used a trailer
equipped with GPS antennas. The GPS measured the
coordinates of the trailer at a frequency of 20 Hz with
error of 0.01 - 0.03 m. It was assumed that the rail-track
was on a horizontal plane, thus, lacked information for
3D modeling. This method was also not capable of
showing detailed visual information of the scene.
Tsai and Li used a laser system to detect the cracks on
the pavement [15]. Two laser units were mounted on a
vehicle to cover a full lane width. Laser beams were pro-
jected on the pavement. CMOS cameras were used to
detect the reflected light of the laser. Likewise, Amarasiri
et al. used a CCD camera for measuring pavement ma-
cro-texture and wear of pavement [16,17]. The CCD
camera sensed the optical reflection properties of the
pavement and recorded them on the image.
2.2. Roadway Light Measurement
Arditi et al. developed a system called LUMINA [18].
This system deployed an analog video camera and a
photometer to record the image and luminance value of
the safety vests synchronously. A device called AI-
GOTCHA was used to convert the frames of the analog
videotape to BMP format. The luminance values of the
vests were calculated using RGB values extracted from
the image. The luminance values were also measured
with a photometer for physical calibration of the calcu-
lated luminance. This study had two limitati ons. First, the
analog/digital conv ersion introduced new errors. Second,
the camera had a very low resolution of only 400 lines,
which was inadequate and outdated compared to digital
To improve, other researchers used digital cameras for
recording the roadway luminance. Zatari et al. used two
CCD cameras mounted on top of a vehicle for measure-
ment of glare, luminance and illuminance of road light-
ing [19]. Due to the limited field of view of the cameras,
one camera was aiming straight forward to cover the road,
the other was aiming upward to photograph the lumi-
naires on poles. The measurement error of absolute illu-
minance was within 15 lx, while the average error of
luminance measurement was 0.3185 cd/m2. Likewise,
Aktan et al. used a CCD photometer that was mounted
outside a vehicle and automatically took photos of the
road scene while the vehicle was moving [20]. The reso-
lution of the CCD photometer was 1 megapixel and the
measurement error of road luminance was within 10%.
Ekrias et al. developed a technique to measure the lu-
minance on road with a ProMetric 1400 photometer [2,3].
The CCD camera of the photometer captured the scene
and luminance values a across the whole scene in a few
seconds. Due to the reduced measurement time, the error
of this CCD photometer system was reduced to ±3%.
However, the pixel resolution of this CCD ProMetric
1400 photometer is only 500 × 500, which was not
enough to measure large scenes. Armas and Laugis used
an LMK camera developed by the TechnoTeam Compa-
ny for measuring roadway luminance [21,22]. The LMK
camera could record both the image and luminance val-
ues of the scene with a pixel resolution of 5184 × 3456
for image acquisition and 2592 × 1728 for luminance
mapping [23]. Ylinen et al. also used the LMK Mobile
Advanced luminance measuring camera for evaluation of
LED street lighting [24]. In spite of their potentials for
wide uses in roadway environments, all these measure-
ment techniques proposed by Ekrias et al. [2 ,3], Aktan et
al. [20], Armas and Laugis [21,22], and Ylinen et al. [24]
require d hi gh-cost equipment.
On the other hand, high dynamic range (HDR) photo-
graphy, as one of the seven types of known HDR imag-
ing techniques [25], could be used for luminance map-
ping roadway environments with lower entry barriers.
The HDR photography uses affordable consumer grade
digital cameras fitted with wide-angle or standard lenses
to acquire luminance data across a static scene within 1 -
2 minutes [26]. The scene is photographed via sequential
exposures to cover a high dynamic range of light, given
limited recording capability of current image sensor
technology. The low dynamic range (LDR) photographs
are then picked, those fallen in an appropriate range of
RGB values are fused into a raw HDR image, which is
calibrated for luminance mapping the scene at pixel level
[26]. Free data-fusion software programs include Pho-
tosphere, hdrgen, Radiance and Photolux. The HDR
photography has proven average errors of 1.5% - 10.1%
for luminance mapping gray, black, color, and light-
emitting surfaces [26,27]. Due to time lapse between
exposures, the HDR photography is ideal for static
scenes under constant light. This limitation w ill n o longer
exist with the emergence of superb image sensors
(>1,000,000:1) that will capture a high dynamic scene
with only a single shot [2 5].
H. Y. CAI, L. J. LI
2.3. Roadway Geometry and Light Measurement
Other researchers developed techniques to measure both
geometry and light of the roadway environments. Zhou et
al. mounted a light meter and a longitudinal distance
measurement instrument on a vehicle [28]. A computer
was used to control these devices for measuring the loca-
tion and illuminance of the road elements while the ve-
hicle was moving. This technique did not use the same
device to measure the concurrent light and geometry data.
As a result, the light and geometry data obtained sepa-
rately were not aligned at the same measurement points.
It also had low measurement resolution, a problem affi-
liated with the conventional meters.
Gutierrez et al. developed a camera-aided method to
measure the luminance of a traffic sign lighted by the
vehicle headlights solely from the driver’s view, as well
as its distance [29]. A van was equipped with two mo-
nochrome CCD cameras mounted close to the driver, and
LEDs in the front to simulate the headlight. When the
van was moving, a computer was used to switch the
LEDs on and off. One camera took an image with the
LEDs turned on, later the other camera took an image
with the LEDs turned off. From the two consecutive im-
ages, luminance values of the traffic sign were extracted.
The difference between the extracted luminance values
of the two images was assumed the luminance of the
traffic sign lighted by the LEDs only. This technique
could exclude the influence of ambient light from road
lamps and the headlamps of other vehicles. The distance
from the van to the sign was calculated from the norma-
lized size of the sign and the traveled distance of the van
over the time delay when the two consecutive images
were photographed. Nonetheless, Gutierrez et al.’s tech-
nique had two insufficiencies. First, their technique did
not capture a high dynamic range of light in the roadway
environment, since both cameras, which lacked a superb
image se nsor, did not ph ot ograph the sc ene via sequential
exposures. Second, an accurate measurement of lumin-
ance is dependent on the measurement direction. The
luminance differential extracted from the two consecu-
tively photographed images was not truly the sign lu-
minance when lighted only by the LEDs, because the
viewing direction from the cameras to the sign was
changed over the time delay.
3. Research Gap and Problems
The non-contact measurement techniques developed by
those researchers aforementioned are of great improve-
ment upon meters and tools in light of the measurement
speed, resolution, overall accuracy under varying road-
way conditions (due to the shortened measurement time),
and the capability of measuring large non-uniform sur-
faces. However, none of those techniques is able to
measure geometry and light level of the entire roadway
environment simultaneously with the same device. The
separated measurements increase the field measurement
labor, slow the data collection, and cause difficulties in
follow-up data alignment. Capturing synchronous light
and geometry data across a roadway environment with
the same device can speed up the measurement in that
there is no need to measure twice, and no pixels on the
photographic images are wasted in data treatment later
(e.g., in the study by Wu and Tsai [7]. The concurrent
light and geometry data could serve more needs than
separately measured data for roadway environment as-
sessment, e.g., visibility of objects on the road, legibility
of roadway signs, glare, uniformity, roadway surface
reflectance, and luminance gradient, etc.
On the other hand, on-road measurement, using either
handheld meters and tools or non-contact measurement
equipment mounted on vehicles running on road, is
harmful for roadway safety and overall efficiency. Dur-
ing the measurement, the workers and instrumented ve-
hicles constantly on the road could hinder the roadway
traffic, and increase the risk of rear-end collisions. Off-
road non-contract measurement does not interrupt road-
way traffic, thus, is safer. The HDR photography could
be used for luminance mapping an entire roadway envi-
ronment off the road by approximating the sightline and
field of view of a typical observer. It has the attraction
due to its low entry barrier, extraordinarily high resolu-
tion, rapid measurement speed, and acceptable accuracy.
Yet the HD R photography does no t record any geometry
of the roadway scene. The geometric features, which are
needed for roadway condition evaluation, have to be
measured separately.
Recently, based on the HDR photography, Cai devel-
oped the HDR photogrammetry for measurement of
synchronous light and geometry data of an entire scene
with the same device (the camera) [30]. When used for
close- and middle-range scenes (i.e., 8.14 m), the accu-
racy of the HDR photogrammetry was validated, with
average errors of 1.8% - 6.2% for luminance mapping
and 12.9 - 24.3 mm for measuring geometry. The HDR
photogrammetry has great potentials to solve aforemen-
tioned problems to facilitate the measurement and evalu-
ation of roadway environments. However, the accuracy
of the HDR photogrammetry for measurement of road-
way environments, which are often in far ranges > 8.14
m, has not yet been tested. This study was thus aimed to
validate the feasibility of the HDR photogrammetry for
synchronous roadway light and geometry measurement
under different viewing conditions, and also explore its
potentials for facilitation of expedited roadway environ-
ment evaluation.
H. Y. CAI, L. J. LI
4. The HDR Photogrammetric M ea su re men t
of Roadway Environments
The HDR photogrammetry was developed for acquisition
of luminance of millions of target points across a scene
and their right-handed Cartesian coordinates XYZ in the
field [30]. It deploys a consumer grade digital camera
fitted with a standard, w id e-angle, or fisheye lens
mounted on a portable photogrammetric measurement
platform. The camera is mounted in the test scene with
yaw angle κ, pitch angle η, and roll angle φ related to the
world coordinates XYZ, following the right hand rule. A
notebook computer controls the camera to take multiple
LDR photographs of the test scene, which are later fused
into an HDR image. The HDR image plane is located on
the image sensor of the camera, with pixel coordinates xz.
Any single target plane, on which the target points are
randomly distribu ted, also has ya w ang le θ, pitch angle τ,
and roll angle ρ in light of the world coordinates XYZ. A
target P (X, Y, Z) and a reference point Pi (Xi, Yi, Zi) are
both located on the target plane. The position (Xi, Yi, Zi)
of the reference point Pi is measured in the field. Mini-
mum three, ideally four, reference points are needed for
identifying a target plane. If the yaw, pitch, and roll an-
gles of the target plane, e.g., a flat road pavement, are
known, then only one reference point is needed. Aided
by the reference point Pi, the location (X, Y, Z) of the
target P in the field can be converted from its pixel loc a-
tion (x, z) on the HDR image, by using some photo-
grammetric transforming equations [30]. Meanwhile, the
luminance value of the target P can be obtained from the
HDR image. Then it is possible to extract the luminance
values and XYZ coordinates of the entire scene from the
HDR image at pixel level.
Below is an introduction to the deployment of the
HDR photogrammetry for roadway measurement, in-
cluding the field measurement procedure and data treat-
ment. Figure 1 is the flow chart to depict the field mea-
surement procedure and follow up data treatment.
4.1. Field Measurement Procedure
Below is a protocol divided into five steps of the HDR
photogrammetric measurement in the field.
Step 1: field preparation. The camera location in the
road scene may be any typical viewpoint of the observer
(e.g., a driver, a pedestrian, etc.). An off-road mounting
location is preferred for the benefit of roadway safety and
traffic flow, if the sightline of the observer could be ap-
proximated off the road. With the right lens, the camera’s
field of view shall cov er all interested target poin ts in the
roadway environments. In addition, four reference points
need to be identified for each target plane (e.g., pavement,
sidewalk, roadway sign, etc.). Common reference points
include visible roadway marks, edges of the curb, or field
Figure 1. The flow chart of the field measurement and fol-
low up data treatment.
objects such as light poles, plants, etc. Some reference
objects, such as wood white boards, could be placed on
interested points for the benefit of meter measurement.
An X-Rite 18% gray checker is also mounted on a tripod
and placed close to the camera shooting line for photo-
metric calibration.
Step 2: platform setup and leveling. The camera is then
mounted on a 3D angular measurement tripod head [30],
which is mounted on a heavy-duty flat head tripod and
leveled. The tripod head has an adjustable base with
three screws for micro-adjustment to ensure the platform
is truly horizontal at its initial position. The tripod head
has three angular dials on the base, the front and the side,
respectively, to record the aiming yaw, pitch and roll
angles of the camera with a precision of 0.1 ˚. A Dell La-
titude notebook computer is used for remote control of
the camera and data recording.
Step 3: determining the XYZ coordinates. In roadway
scenes, the World Coordinate System (WCS) is used,
which has fixed coordinates—X (east), Y (north), Z (up).
The origin point O (0, 0, 0) of the photogrammetric
coordinates is overlapped at the optical center of the
camera. To determine the initial yaw pitch and roll an-
gles (κXYZ , ηXYZ, φXY Z) of the XYZ coordinates, a compass
is put on the platform to adjust the 3D angular measure-
ment tripod head until the camera is aiming north. Yaw
angle κXYZ and pitch angle ηXYZ are recorded from the
angular dials. Roll angle φXYZ is assumed 0˚ for roadway
Step 4: HDR photographing. The camera is manually
focused on an aiming point in the field, which is careful-
ly selected to balance the field of view and make sure all
targets across the entire roadway environment are in fo-
cus. A chroma meter Minolta CL-200A is used to meas-
ure the correlated color temperature (CCT) and vertical
illuminance at the camera lens. The CCT is used for
white balancing of photographing. The vertical illumin-
H. Y. CAI, L. J. LI
ance at the lens is used to control the light measurement
in the field and as an input of potential glare calculation
(not covered in this study). The Dell notebook computer
is used to remotely control the camera to take a total of
18 LDR photographs via sequential exposures (1/4000 s
to 30 s @ aperture of f/5.0). Some of them fallen in a
good exposure range are fused into a raw HDR image.
During the photographing, the luminance meter Minolta
LS-100 mounted beside the camera is used to measure
the luminance of the X-Rite 18% gray checker for pho-
tometric calibrations of the raw HDR image.
Step 5: field measurements and calibration. From the
three dials of the tripod head, the camera’s initial aiming
direction (κ0, η0, φ0) is recorded. The actual aiming an-
gles (κ, η, φ) are calculated as κ = κ0 κXYZ, η = η0
ηXYZ, φ = φ0 φXYZ. Next, the camera is dismounted from
the tripod head and replaced with a laser distance meter
Leica DISTO™ D5 for measuring the distance di of the
reference point Pi. If faraway points exceed the mea-
surement range (650 ft) of the laser distance meter, a
Bushnell EliteTM 1600 laser rangefin de r is used to repla c e
the laser distance meter. The ai ming angles of Pi (ϕi,a, αi,a)
are then recorded from the angular dials on the side and
base of the tripod head. Note that due to d ifferent optical
structures, when replacing the camera with the laser dis-
tance meter or the rangefinder, the aiming direction may
change. To calibrate this error, the laser distance meter or
the rangefinder is re-aimed to the same aiming point of
the camera. The new aiming direction (κm0, ηm0, φm0) is
recorded from the three angular dials. The offsets of the
aiming angles are calculated using Equation (1). The
off-axis angles are then calculated using Equation (2).
The four reference points Pi (Xi, Yi, Zi) (i = 1, 2, 3, 4) are
later calculated using Equation (3).
κκ κ
ηη η
∆= −
∆= −
ii aXYZ
ii aXYZ
φφ κκ
αα ηη
= −−∆
= −−∆
(2 )
( )
( )
cos sin
cos cos
ii i
i iii
= −
4.2. Data Treatment
The field-measured data are then treated using personal
computers at six steps below, which are numbered after
the previous five steps.
Step 6: generation of raw HDR images. Raw HDR
images are generated from an appropriate exposure range
of those LDR photographs using Photosphere or Ra-
diance [26]. Before that, if necessary, the LDR photo-
graphs need treatments to filter any unwanted voltage
bias, dark current and conductive noise, and fixed pattern
noise [26,31].
Step 7: calibrations of the raw HDR images. The raw
HDR images are then calibrated in Radiance, including,
in order, 1) vignetting effect correction, which corrects
the light drop-off in the periphery of the field of view,
and 2) photometric calibration, which corrects the lu-
minance differential between the camera-aided mea-
surement and the luminance meter measurement of the
X-Rite 18% gray check er [26].
Step 8: data retrieval from the calibrated HDR images.
Of every single pixel, the luminance and pixel location
Ppix (xpix, zpix) are extracted from the calibrated HDR im-
ages in Radiance, using the command “pvalue” (pvalue
-o -h -b $HDR > L_$HDR.txt). The output text file en-
closes per-pixel luminance values across the scene in a
data format of {xpix, zpix, luminance}. In addition, from
the pixel location, the distorted geometric coordinate Pd
(xd, zd) of the target point P on the image plane xz is then
derived [30].
Step 9: correction of the lens distortion. A wide-angle
lens often has radial distortions that need corrections for
geometry measurement. Of every single pixel, the dis-
torted geometric coordinates Pd (xd, zd) are then corrected
as undistorted Pu (xu, zu). More details of the correction
models are available in the literature [30].
Step 10: photogrammetric transformation & calibra-
tion. The XYZ coordinates of the target point P in the
field are then transformed from its undistorted geometric
coordinates Pu (xu, zu) on the image plane, using some
transformation equations developed by Cai [30]. The
local coordinates X′Y′Z′ of the target point P are first
calculated from Pu (xu, zu). Then the local X′Y′Z′ coordi-
nates are converted back to the WCS coordinates P (X, Y,
Z) with the aid of the field measured XYZ coordinates of
the reference points. The last step is the photogrammetric
calibration using the four reference points. The average
geometric deviation between their measured and the cal-
culated XYZ coordinates is used to furth er calibrate th e P
(X, Y, Z) [30].
Step 11: collection of synchronous data for further
treatment. Of every single pixel, its luminance L and
calibrated 3D geometric coordinates P (Xcali , Ycali, Zcali)
are then collected for further data treatments, e.g., calcu-
lations of visibility of roadway objects, roadway signs
legibility, discomfort glare of roadway lights, roadway
surface reflectance, pavement uniformity. The output text
file that encloses per-pixel luminance values of each ca-
librated HDR image is used to plot in MatLab a 2D lumin-
ance map {xpix, zpix, luminance}, a 2D luminance gradient
magnitude map {xpix, zpix, luminance gradient m agnitude}, a
2D luminance gradient direction map {xpix, zpix, luminance
gradient direction}, and a comb in ed 3D luminance gradient
map {xpix, zpix, luminance gradient magnitude, luminance
H. Y. CAI, L. J. LI
gradient direction} [1]. The underlined coordinate of each
map is in pseudo color. The MatLab code for such data
treatment and ima ge plotting is availa ble online
5. Validation of the Method in the Field
5.1. Field Measurements
The HDR photogrammetric measurement is fast, non-
contact, non-destructive, and off the road. To validate
this method, the roadway environments of a straight and
flat section of Jayhawk Boulevard (482.8 m long) at
Lawrence, KS and a roundabout (15.3 m in diameter) at
its end were measured. The Jayhawk Boulevard went
through the campus of the University of Kansas. On both
sides of this road, there were buildings, trees, light poles,
traffic signs, statues, newspaper booth, etc. The rounda-
bout had three asymmetrical non-straight approaches.
In this study, the Canon camera EOS rebel T2i with an
18-mega pixel CMOS sensor fitted with a S igma EX DC
HSM 10 mm diagonal fisheye lens was deployed. The
HDR photogrammetric measurement platform is shown
in Figure 2, on which the camera was mounted. The op-
timized lens aperture size of f/5.0 was used based on
previous studies [26,31]. Several pieces of wood white
boards were placed on interested locations serving as the
target and reference points. In both test scenarios, an
off-road measuring location was chosen.
For measuring Jayhawk Boulevard, four HDR images
were generated under clear or cloudy sky in the daytime,
and also at night with dry or wet pavement, as shown in
Figu re 3, which were photographed at the same off-road
location aiming at the same direction.
In addition, a to tal of 24 points on Jayhawk Bo ulevard
were selected for measurement of their XYZ coordinates,
as shown in F igure 4 . Points 1 - 4 were ref erence points.
Others were target po ints. Points 5 - 11 a nd 15 - 20 were
the centers of crosswalk strips. Point 14 was the focusing
target of the camera, which was a blue newspaper booth.
Points 12 and 13 were the bottoms of trees. Points 21 -
23 were the bottoms of light poles. Point 24 was a traffic
booth at the end of the road. Independent of various
viewing conditions, the field measurement of the 24
points was conducted in the daytime under clear sky due
to better legibility of the far targets. White wood boards
were placed at those points to help the laser distance me-
ter and rangefinder acquire the targets.
The roadway environment at the roundabout was also
measured off the road. Figure 5 shows the four HDR
images taken under clear or cloudy sky in the daytime,
and also at night with dry or wet pavement.
A total of 21 points, as shown in Figure 6, were
measured in the field in the daytime under clear sky for
their XYZ coordinates. Points 1 - 6 were the top left cor-
ners of the crosswalk strips. Points 7 - 13 were white
(a) (b)
Figure 2. The HDR photogrammetric measurement plat-
form used (a) on Jayhawk Boulevard, and (b) at the roun-
(a) (b)
(c) (d)
Figure 3. Four HDR images of Jayhawk Boulevard taken at
the same off-road location aiming at the same direction but
under different environmental conditions: (a) Dayt ime ,
clear sky, (b) Daytime, cloudy sky, (c) Nighttime, dry pave-
ment, and (d) Nighttime, wet pavement.
wood boards place d on the curb. Points 7 and 13 were on
roadside. Points 8 - 10 were on the central island. Points
11 and 12 were on a deflection island. Points 14 - 16
were the centers of two “STOP” signs. Points 17 - 21
were corners of deflection markings. Since the rounda-
bout was not flat, those 21 points were divided into three
groups in data treatment based on their Z coordinates. In
each group, the target points had very similar Z values.
The first group consists of Points 1 - 8, 17 - 19. Among
them, Points 1, 6, 7, and 9 were used as reference points
for data treatment. Other points were used as target
points to validate the photogrammetric measurement.
The second group consists of Points 10, 11, 12, and 16.
Points 10 - 12 were used as reference points, while Point
16 was used as a target point. The third group consists of
Points 13 - 15, 20 and 21. Points 13, 14, 15 and 20 were
used as reference points, while Point 21 was used as a
target point.
5.2. Results of Geometry Measurement
Of every single point measured in the two roadway en-
H. Y. CAI, L. J. LI
Figure 4. Locations of the 24 points measured on Jayhawk Boulevard in the daytime under clear sky.
(a) (b) (c) (d)
Figure 5. The four HDR images of the roundabout taken at the same location aiming at the same direction but under differ-
ent environmental conditions: (a) Daytime, clear sky, (b) Dayti me , cloudy sky, (c) Nighttime, dry pavement, and (d) Night-
time, wet pavement.
Figure 6. Locations of the 21 points measured at the roundabout in the daytime under clear sky.
vironments, the calibrated geometric coordinates Xcali,
Ycali, Zcali were calculated from its pixel location extracted
from the calibrated HDR image generated in the daytime
under clear sky. The calculated Xcali, Ycali, and Zcali were
then compared to the coordinates Xmeter, Ymeter, and Zmeter
actually measured in the field using the laser distance
meter or rangefinder mounted on the 3D angular tripod
head. Equation (4) was used to calculate the measure-
ment errors in percentage of the 3D geometric XYZ
coordinates. Since the measurement errors vary with the
distance, an absolute error in millimeter is not as useful
as the error percentage in reflecting the results of geome-
H. Y. CAI, L. J. LI
try measurement, thus, not included.
% 100
% 100
% 100
cali meter
cali meter
cali meter
= ×
= ×
= ×
Table 1 summarizes the measurement errors of the 24
points on Jayhawk Boulevard. Likewise, the geometric
errors of measuring the 21 points at the roundabout are
shown in Table 2. In Tables 1 and 2, the interquartile
range (IQR) covers the most common (the middle 50%)
measurement errors in practice (Cai, 2013). It was found
that the errors of geometry measurement on Jayhawk
Boulevard, regardless of X, Y, or Z coordinate, varied a
lot with the measuring distance, e.g., from 5.7% to 39.5%
with an interquartile r ange (IQR) of 14 .3% - 37.5% for X
coordinate, from 6.8% to 40.3% with an interquartile
range (IQR) of 21.1% - 38.6% for Y coordinate, and
from 0.6% to 61.6% with an interquartile range (IQR) of
4.4% - 39.4% for Z coordinate. In addition, the average
errors of measuring the XYZ coordinates were ranged
from 23.1% to 27.5%. The margin of errors for measur-
ing the geometry of the roundabout was also very large,
0.1% - 90.0% (X), 0 .2% - 52.3 % (Y), 0.0% - 13 7.5 % (Z),
due to its sloped pavement. The largest error (137.5%)
occurred at Z coordinate for approximation of the sloped
and curved roundabout pavement. The average errors of
the measured XYZ coordinates of the 21 target points
were in a range of 9.3% - 16.2% (for XYZ coordinates),
with IQR of 1.9% - 16.5% (X), 4.9% - 14.4% (Y), and
0.0% - 2.2% (Z), smaller than those of the longer Jay-
hawk Boulevard.
5.3. Results of L ight Measurement
Luminance mapping the two roadway environments was
at pixel level. Figures 7 and 8 show the light distribution
maps plotted in MatLab measured on Jayhawk Boulevard
and at the roundabou t, respectively, in the daytime und er
clear sky or cloudy sky, or at nighttime with dry or wet
pavement. Each map contains 18 millions values. For the
benefit of direct comparison, the luminance scale of dif-
ferent maps is fixed in a range of 0.00519 cd/m2 (the
minimum luminance recorded in those eight HDR im-
ages)—117782.0 cd/m2 (the maximum luminance rec-
orded). Likewise, the luminance gradient magnitude is
set in a fixed range of 8.950 × 107 - 60834.900 cd/m2/
pixel. The luminance gradient direction is in a fixed an-
gular range of 180˚ - 180˚, measured anticlockwise in
light of the x axis of the photographic images (e.g., the
angle 0˚ is to the right hand, 90˚ is up, 180˚/180˚ is to
Table 1. Errors of measuring the XYZ coordinates of the 24
points on Jayhawk Boulevard.
Measurement Errors (%)
Max 39.5 40.3 61.1
Min 5.7 6.8 0.6
Average 25.0 27.5 23.1
Median 24.1 27.1 10.2
IQR 14.3 - 37.5 21.1 - 38.6 4.4 - 39.4
Table 2. Errors of measuring the XYZ coordinates of the 21
points at the roundabout.
Measurement Errors (%)
Max 90.0 52.3 137.5
Min 0.1 0.2 0.0
Average 16.2 12.1 9.3
Median 8.6 8.6 0.0
IQR 1.9 - 16.5 4.9 - 14.4 0.0 - 2.2
the left, 90˚ is down). Figures 7 and 8 visualize the per-
pixel lighting conditions across the two roadway envi-
The two roadway environments had the highest overall
light level in the daytime under clear sky, and the lowest
light level at rainy night. At nighttime, the luminance of
the wet pavement was lower and less uniform than that
of the dry pavement. In particular, puddles that reflected
the light of the road lamps to the camera appeared brigh-
ter. The overall light distribution in sunny days was less
uniform (the gradient magnitude was higher) than that in
cloudy days, and than that at night. In particular, the sky
on cloudy days was less uniform (the gradient magnitude
was higher) than that on sunny days, while the sky un-
iformity was the best at night independent of the pave-
ment conditions.
On Jayhawk Boulevard, the pavement, building façade
and plants had much higher change rate of luminance
than concrete sidewalks, glass windows and the sky. This
is because the pavement and building façade have unpo-
lished rough surface, and plants have complex geometric
shapes, which diffuse the reflected light very well in dif-
ferent directions. At the roundabout, the triangle YIELD
sign had more uniform luminance than surrounding ob-
jects, due to its well-maintained retro-reflective surface.
On the gradient direction maps, objects with polished
surfaces can be easily detected, like road lamp poles,
tripod, building façade, and signs. However, the unpo-
lished rough surfaces, like pavement and plants, were
barely visible, since their reflected light was well dif-
fused without a primary direction. The edges of objects,
H. Y. CAI, L. J. LI
H. Y. CAI, L. J. LI
Figure 7. The lighting conditions of Jayhawk Boulevard (a) in the daytime under clear sky, (b) in the daytime under cloudy
sky, (c) at nighttime with dry pavement, (d) at nighttime with wet pavement.
H. Y. CAI, L. J. LI
H. Y. CAI, L. J. LI
Figure 8. The lighting condi tions of the round about (a) in the dayti me unde r cle ar sky, ( b) in the dayti me under cloudy sky, (c)
at nighttime with dry pavement, (d) at nighttime with wet pavement.
H. Y. CAI, L. J. LI
e.g., buildings, pavement, sidewalks, poles, plants, could
still be detected due to changed directions of light reflec-
tion on edge s.
Moreover, those luminance maps and luminance gra-
dient magnitude maps shown in Figures 7 and 8 were
further treated for identifying potential roadway hazards,
including unshielded roadway lamps, too bright pave-
ment puddles, extremely high roadway contrasts, and
large non-uniform pevement shadows, etc. Any electric
light sources brighter than a threshold value of 500 - 700
cd/m2 will cause glare at night [32]. This threshold value
could be much higher in the daytime due to increased
adaptation level. Extreme light contrasts can also be
identified by a threshold luminance gradient magnitude,
e.g., 500 - 700 cd/m2/pixel that is consistent with the
threshold luminance, whose polarity is shown by the
gradient direction. The distracting glare sources and large
and sharp pavement shadows will harm the drivers’ de-
tection of pavement markings and road targets, thus,
need to be identified and eliminated for roadway safety.
For demonstration, this study adopted the threshold val-
ues of 5000 cd/m2 and 5000 cd/m2/pixel in the daytime
and 500 cd/m2 and 500 cd/m2/pixel at nighttime. The
number of 5000 was assumed, since the threshold value
in the daytime is not a constant which varies with the
light adaptation level. The threshold values could be ad-
justed for more or less stringent needs. Figure 9 shows
the identified potential hazards of the two roadway envi-
ronments at their pixel lo cations, from which the geome-
tric locations could be calculated (but not actually calcu-
lated in this study due to tremendous workload). Figure
9 is useful to find the quantity, location and magnitude of
any potential lighting hazards in the two roadway envi-
ronments at first glance. It was found that the sky was a
potential glare source in cloudy days but not under clear
sky. At nighttime, the roadway electric lights were the
only potential glare sources.
6. Discussion
Capturing synchronous light and geometry data using the
HDR photogrammetric measurement method developed
in this study could speed up the measurement of roadway
environments. The measurement speed is much faster
than conventional point-by-point measurement using
meters and tools in that millions of target points visible
to the camera lens can be measured simultaneously. The
camera-aided measurement takes only 1 - 2 minutes for
photograph ing the entire scene with sequen tial exposures,
plus another short period of time for measuring reference
points in the field using the laser distance meter or
rangefinder mounted on the 3D angular measurement
tripod head. The average speed was approximately 3
minutes per point in this study. Often four reference
points are sufficient for measuring the pavement and the
sidewalks. For measuring other targets on different
planes, more reference points are needed. A short time of
HDR photographing is critical for the accuracy of lu-
minance measurement. Yet the measurement of steady
reference points in the field could take longer time with-
out jeopardizing the accuracy of their XYZ coordinates.
The HDR photogrammetric measurement can solve
some problems of existing roadway environment mea-
surements and facilitate the data treatment. The existing
measurements of light and geometry of the roadway en-
vironments are separated using different devices. The
separated measurements increase the field measurement
labor, slow the data collection, and cause difficulties in
follow-up data alignment due to inconsistent measuring
resolutions and displaced measuring points. These prob-
lems can be solved using the HDR photogrammetric
measurement, in which concurrent luminance and geo-
metry data can be acquired in extraordinarily high reso-
lution. The so acquired 18 millions of synchronous light
and geometry data embedded on an HDR image of the
roadway environment can also facilitate the evaluation of
the overall roadway lighting conditions. Example appli-
cations include 1) spatial and temporal light distributions
illustrated on 2D luminance map and 2D/3D luminance
gradient maps, 2) identification of potential road visual
hazards (e.g., highlight, non-uniformity, large shadow,
extremely large contrast, etc.) on the pavement and si-
dewalks together with their sizes and shapes, and 3)
identification of potential glaring sources and their geo-
metries and locations. These applications were tested
useful in this study in the roadway environments on Jay-
hawk Boulevard and at the roundabout. Moreover, the
acquired concurrent light and geometry data could be
used for derivation of some higher level judging metrics,
with which the measured geometry and light are jointly
involved, for timely roadway environment assessment.
Such advanced metrics include visibility of objects on
the road, legibility of roadway signs, glare, uniformity,
and roadway surface reflectance, etc. Their calcultions
were not covered in this paper but will be tested in fur-
ther studies.
The HDR photogrammetric measurement of the road-
way environments is aided by affordable consumer-grade
digital cameras, personal computers, and free online
software programs (e.g., Photosphere, Radiance, and the
MatLab code shared online for data treatment and image
plotting (
Although more equipment is needed, such as a laser dis-
tance meter or rangefinder, the Minolta meters LS-100
and CL-200A, they can be shared with other uses. The
specially designed 3D angular measurment tripod head
that is mounted on the photogrammetric measurement
platform is still affordabe (recently quoted $1387 for a
H. Y. CAI, L. J. LI
H. Y. CAI, L. J. LI
Figure 9. The identified potential lighting hazards and their pixel locations on the HDR images, (a) on Jayhawk Boulevard in
the daytime under clear sky, (b) on Jayhawk Boulevard in the daytime under cloudy sky, (c) at the roundabout in the day-
time under c lear sky, (d) at the roundabo ut in t he day ti me under cloudy sky, (e) on Jayhawk Boulevard at nighttime with dry
pavement, (f) on Jayhawk Boulevard at nighttime with wet pavement, (g) at the roundabout at nighttime with dry pavement,
(h) at the roundabout at nighttime with wet pavement.
H. Y. CAI, L. J. LI
In addition, the HDR pho tog rammetric measurement is
non-contact, non-destructive, and could be off road. Thus,
it is safer and more efficient than on-road measurement
using meters and tools and the aforementioned remote
technologies equipped on vehicles moving on road.
However, off-road measurement is preferred only when
the approximation of the typical viewing directions of
drivers and pedestrians at a off-road location is valid. It is
challenging but theoretically possible to extend the
sightline of a driver on road to an off-road position given
that the measurement of luminance and geometry (e.g.,
size, shape, spacing, width, slope, etc.) of the targets is
indepen dent of the viewing distance .
On the other hand, since the luminance measurement
depends on the viewing direction of the camera lens (or
the aiming direction of a luminance meter), the measured
2D luminance maps and 3D/2D luminance gradient maps
can only be used to evaluate the roadway environments
viewed at that specific viewing direction at which the
camera lens aims. However, this problem is not new,
which also exists in the conventional meter measurement.
By taking multiple HDR images of the roadway envi-
ronments from different viewing directions of a typical
observer, this problem could be relieved. Multiple HDR
images of the same roadway environment can be taken
either simultaneously using multiple test rigs mounted at
different locations or sequentially using the same set of
test rig.
The HDR photogrammetric measurement can handle
many roadway elements in different shapes, sizes and
orientations, as long as they are visible to the camera lens.
Theoretically, an HDR image is embedded with millions
of light and geometry data of target points visible to the
camera lens. Thus, the HDR photogrammetric measure-
ment can cover any visible planes with available refer-
ence points. In this study, as two examples, dozens of
roadway points were measured not only on the straight
and flat Jayhawk Boulevard, but also on the oblique and
circular pavement and sidewalks of the roundabout.
Therefore, the HDR photogrammetric measurement is
deemed more useful than the existing remote methods
developed by Drakopoulos and Ornek [4], Tsai et al.
[5,6], Wu and Ts ai [7,8], and Nehate and Rys [12], which
are unable to measure oblique and circular planes.
Nonetheless, this study did not validate the accuracy of
the HDR photogrammetric measurement for luminance
mapping the two roadway environments, because of two
reasons. First, the accuracy of the HDR photography in
indoor and outdoor environments have already been
proven acceptable (1.5% - 10.1% for luminance mapping
gray, black, color, and light-emitting surfaces) in the li-
terature [26,27]. The two roadway environments do not
differ from the previous measured environments in light
of the dynamic range of luminance. Second, there is no
appropriate way to validate the accuracy of luminance
measurement on road using a luminance meter. With a
low resolution, the luminan ce meter is incapable to mea-
surement a small target point faraway on road. In addi-
tion, since luminance measurement depends on the
viewing direction, both the luminance meter and the
camera lens should aim at the target point at exactly the
same direction, which is impossible without blocking the
view of the camera lens for photographing. Then, why
not measure the luminance in sequence? Unfortunately,
meter measurement before or after the HDR photo-
graphing is useless since the roadway lighting often va-
ries over time.
On the other hand, it was found that the HDR photo-
grammetric measurement with current equipment had a
large margin of errors for g eometry measurement, which
varied with the measuring distance. Target points that
were farther from the camera had larger measuring er-
rors. For measuring the longer Jayhawk Boulevard, the
errors varied from 0.6% to 61.1%. For measuring the
closer roundabout, the margin of errors was from 0.0% to
137.5% that is larger due to its sloped pavement. The
average errors were calculated as 25.0% (for X coordi-
nate), 27.5% (Y), and 23.1% (Z) for measuring the 24
points on Jayhawk Boulevard, and 16.2% (X), 12.1%
(Y), 9.3% (Z) for measuring the 21 points at the rounda-
bout with smaller measurement distances. The large er-
rors were caused by some limitations of current mea-
surement equipment. First, the 3D angular measurement
tripod head has a precision of 0.1˚, which is insufficient
to measure targets that are very far from the camera, due
to the high sensitivity of angles in the photogrammetric
transformation. Measuring the XYZ coordinates of a
target point in a roadway environment is more difficult,
thus, less accurate, than the simple distance measurement
from the target to the camera. Second, the rangefinder
was used to replace the laser distance meter for measur-
ing faraway targets that were beyond its range of 650
feet. The rangefinder had an error of ±1 m, which con-
tributed to the large errors of measuring faraway targets.
Despite the large errors for geometric measurement,
the HDR photogrammetric measurement is still having
similar accuracy to most existing roadway measurement
methods, e.g., Drakopoulos and Ornek’s algorithm (the
measurement errors were up to 63.8%) [4], Tsai’s algo-
rithm (error was within 0.01 - 0.3 m) [5-8], Nehate and
Rysmethod using GPS data (error was ±0.2 m) [12],
Castro et al.’s Leica System 500 GPS (with maximum
error of 1 m) [13]. Nonetheless, the HDR photogramme-
tric measurement is useful for measuring synchronous
light and geometry data in different types of roadway
environments, which no previous measurement methods
could do in such a high resolution.
H. Y. CAI, L. J. LI
7. Conclusion
The HDR photogrammetric measurement developed in
this study is useful for evaluating roadway environments
but still needs improvements on the measurement accu-
racy and the speed of data treatment. Two measures can
be used in further studies to reduce the margin of errors
of geometry measurement. First, the angular resolution of
the 3D angular measurement tripod head could be in-
creased from 0.1˚ to 0.05˚, which is expected to largely
reduce the measurement errors. Second, deployment of
more accurate techniques for measuring the XYZ coor-
dinates of reference points in the field, e.g., using more
advanced remote sensing technologies, will also largely
reduce the workload in the field. Future studies will also
need to look into accurate compensation of the difference
in aiming angles between the camera and distance meters
when they are aiming at the same target due to their dif-
ferent optical structures. In addition, a program running
in MatLab is currently under development and testing in
the University of Kansas lighting research laboratory to
automatically conduct the photogrammetric transforma-
tion and calibrations. This program running on standard
PC can speed up the tremendous data treatment.
This research was supported by a Start-up Fund provided
by the School of Engineering and a New Faculty General
Research Fund provided by the Research and Graduate
Studies at the University of Kansas. The authors thank
Ms. Xiaomeng Su for her help on the MatLab code for
data treatment and image plotting.
[1] X. Su, M. Mahoney, M. I. Saifan and H. Cai, “Mapping
Luminance Gradient across a Private Office for Day-
lighting Performance,” The Illuminating Engineering So-
ciety (IES) Annual Conference, Minneapolis, 11-13 Nov-
ember 2012, pp. 1-15.
[2] A. Ekrias, M. Eloholma and L. Halonen, “Analysis of
Road Lighting Quantity and Quality in Varying Weather
Conditions,” LEUKOS, Vol. 4, No. 2, 2007, pp. 89-98.
[3] A. Ekrias, M. Eloholma, L. Halonen, X. Song, X. Zhang
and Y. Wen, “Road Lighting and Headlights: Luminance
Measurements and Automobile Lighting Simulations,”
Building and Environment, Vol. 43, No. 4, 2008, pp. 530-
[4] D. Drkopoulos and E. Ornek, “Use of Vehicle-Collected
Data to Calculate Existing Roadway Geometry,” Journal
of Transportation Engineering, Vol. 126, No. 2, 2000, pp.
[5] Y. C. Tsai, J. P. Wu, Y. C. Wu and Z. H. Wang, “Auto-
matic Roadway Geometry Measurement Algorithm Using
Video Images,” Image Analysis and Processing-ICIAP,
Vol. 3617, 2005, pp. 669-678.
[6] Y. C. Tsai, Z. Hu and Z. Wang, “Vision-Based Roadway
Geometry Computation,” Journal of Transportation En-
gineering, Vol. 136, No. 3, 2009, pp. 223-233.
[7] J. P. Wu and Y. C. Tsai, “Enhanced Roadway Geometry
Data Collection Using an Effective Video Log Image Pro-
cessing Algorithm,” Journal of the Transportation Re-
search Record, Vol. 1972, No. 1, 2006, pp. 133-140.
[8] J. P. Wu and Y. C. Tsai, “Enhanced Roadway Inventory
Using a 2-D Sign Video Image Recognition Algorithm,”
Computer-Aided Civil and Infrastructure Engineering,
Vol. 21, No. 5, 2006, pp . 369-382.
[9] B. W. He, X. L. Zhou and Y. F. Li, “A New Camera
Calibration Method from Vanishing Points in a Vision
Sys te m ,” Transactions of the Institute of Measurement
and Control, Vol. 33, No. 7, 2011, pp. 806-822.
[10] R. Cipolla, T. Drummond and D. Robertson, “Camera
Calibration from Vanishing Points in Images of Architec-
tural Scenes,” British Machine Vision Conference, British
Machine Vision Association, London, 7-10 September
1999, pp. 224-233.
[11] B. Caprile and V. Torre, “Using Vanishing Points for
Camera Calibration,” International Journal of Computer
Vision, Vol. 4, No. 2, 1990, pp. 127 -140.
[12] G. Nehate andM. Rys, “3D Calculation of Stopping-Sight
Distance from GPS Data,” Journal of Transportation En-
gineering, Vol. 132, No. 9, 2006, pp. 691-698.
[13] M. Castro, L. Iglesias, R. Rodriguez-Solano and J. A.
Sanchez, “Geometric Modelling of Highway s Using Glo-
bal Positioning System (GPS) Data and Spline Approxi-
mation,” Transportation Research Part C, Vol. 14, No. 4,
2006, pp. 233-243.
[14] W. Koc, “Design of Rail-Track Geometric Systems by
Satellite Measurement,” Journal of Transportation En-
gineering, Vol. 138, No. 1, 2012, pp. 114-122.
[15] Y. C. Tsai and F. Li, “Critical Assessment of Detecting
Asphalt Pavement Cracks under Different Lighting and
Low Intensity Contrast Conditions Using Emerging 3d
Laser Technology,” Journal of Transportation Engineer-
ing, Vol. 138, No. 5, 2012, pp. 649 -659.
[16] S. Amarasiri, M. Gunaratne and S. Sarkar, “Modeling of
Crack Depths in Digital Images of Concrete Pavements
Using Optical Reflection Properties,” Journal of Tran-
sportation Engineering, Vol. 136 , No. 6, 2010, pp. 489-
[17] S. Amarasiri, M. Gunaratne and S. Sarkar, “Use of Digital
Image Modeling for Evaluation of Concrete Pavement
Macrotexture and Wear,” Journal of Transportation En-
H. Y. CAI, L. J. LI
gineering, Vol. 138, No. 5, 2012, pp. 589-602.
[18] D. Arditi, M. A. Ayrancioglu and J. Shi, “Effectiveness of
Safety Vests in Nighttime Highway Construction,” Jour-
nal of Transportation Engineering, Vol. 130, No. 6, 2004,
pp. 725-732.
[19] A. Zatari, G. Dodds, K. McMenemy and R. Robinson,
“Glare, Luminance, and Illuminance Measurements of
Road Lighting Using Vehicle Mounted CCD Cameras,”
LEUKOS, Vol. 1, No. 2, 2004, pp. 85-106.
[20] F. Aktan, T. Schnell and M. Aktan, “Development of a
Model to Calculate Roadway Luminance Induced by Fix-
ed Roadway Lighting,” Journal of the Transportation Re-
search Board, Vol. 1973, 2006, pp. 130-141.
[21] J. Armas and J. Laugis, “Increase Pedestrian Safety by
Critical Crossroads: Lighting Measurements and Analy-
sis,” The 12th European Conference on Power Elec-
tronics and Applications, Aalborg, 2-5 September 2007,
pp. 1-10.
[22] J. Armas and J. Laugis,. “Road Safety by Improved Road
Lighting: Road Lighting Measurements and Analysis,
[23] TechnoTeam, n.d., “LMK Mobile Advanced Specifica-
tion,” 2012.
[24] A. Ylinen, L. Tahkamo, M. Puolakka and L. Halonen,
“Road Lighting Quality, Energy Efficiency, and Mesopic
Design-LED Street Lighting Case Study,” LEUKOS, Vol.
8, No. 1, 2011, pp. 9-24.
[25] S. K. Nayar and V. Branzoi, “Adaptive Dynamic Range
Imaging: Optical Control of Pixel Exposures over Space
and Time,” Proceedings of the 9th IEEE International
Conference on Computer Vision, Nice, 13-16 October
2003, pp. 1168-1175.
[26] H. Cai and T. M. Chung, “Improving the Quality of High
Dynamic Range Images,” Lighting Research & Technol-
ogy, Vol. 43, No. 1, 2011, pp. 87-102.
[27] M. N. Inanici, “Evaluation of High Dynamic Range Pho-
tography as a Luminance Data Acquisition System,”
Lighting Research & Technology, Vol. 38, No. 2, 2006,
pp. 123-136.
[28] H. Zhou, F. Pirinccioglu and P. Hsu, “A New Roadway
Lighting Measurement System,” Transportation Research
Part C, Vol. 17, No. 3, 2009, pp. 274-284.
[29] J. A. Gutierrez, D. Ortiz de Lejarazu, J. A. Real, A.
Mansilla and J. Vizmanos, “Dynamic Measurement of
Traffic Sign Luminance as Perceived by a Driver,” Light-
ing Research & Technology, Vol. 44, No. 3, 2011, pp.
[30] H. Cai, “High Dynamic Range Photogrammetry for Syn-
chronous Luminance and Geometry Measurement,”
Lighting Research & Technology, Vol. 45, No. 2, 2013,
pp. 230-257.
[31] T. M. Chung and H. Cai, “High Dynamic Range Images
Generated under Mixed Ty pes of Lighting,” The Illumin-
ating Engineering Society (IES) Annual Conference, To-
ronto, 7-9 November 2010, pp. 1-13.
[32] Commission International de l’Eclairage (CIE), “Discom-
fort Glare in the Interior Working Environment,CIE
Publication 055, CIE, Paris, 1983.