Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Case Studies in Construction Materials 18 (2023) e02048

Contents lists available at ScienceDirect

Case Studies in Construction Materials


journal homepage: www.elsevier.com/locate/cscm

Visibility of flat line and structured road markings for


machine vision
Tomasz E. Burghardt a, *, Oleksandr Chistov b, Thomas Reiter b, c, Roman Popp d,
Bernhard Helmreich a, Friedrich Wiesinger a
a
M. Swarovski GmbH, Wipark, 14 Straße 11, 3363 Neufurth, Austria
b
ZKW Group GmbH, Rottenhauser Straße 8, 3250 Wieselburg, Austria
c
Engineering Center Steyr GmbH & Co KG, Steyrer Straße 32, 4300 St. Valentin, Austria
d
ZKW Lichtsysteme GmbH, Scheibbser Straße 17, 3250 Wieselburg, Austria

A R T I C L E I N F O A B S T R A C T

Keywords: Road markings are equally needed for human drivers and for machine vision equipment; their
Retroreflectivity visibility demands high contrast ratio. Particularly difficult is achieving visibility under the
LiDAR conditions of wetness at night and in the presence of glare from an oncoming vehicle. A field
Camera
evaluation of visibility by camera and by LiDAR was done on two types of road markings that
Contrast ratio
Glare
were applied as pedestrian crossing: flat lines with typical retroreflectivity and structured lines
Pedestrian crossing with high retroreflectivity – thus, extreme cases were assessed. The measured camera contrast
Pavement markings ratio dropped meaningfully in the presence of moisture in case of flat line markings, but remained
high in case of structured markings that facilitated moisture drainage. Glare was detrimental to
visibility, bringing it to almost naught regardless of the type of road markings. Simultaneous
evaluation with LiDAR showed profound differences under the conditions of moisture: while the
response from flat line markings dropped to nil (recovery time >20 s after wetting), the structured
markings continuously provided meaningful response. This outcome proves that for reliable
guiding by machine vision, as well as for human drivers, structured road markings that facilitate
water drainage should be used. For dependable steering by machine vision equipment under
adverse conditions, a combination of LiDAR and camera is seen as necessary.

1. Introduction

With the recent advances in transport technology a vehicle starts to be seen more as a system than a machine; hence, the role of
machine vision (MV) is increasing. For successful introduction of automated vehicles (AV) and for assurance of reliability of some of
the advanced driver assistance systems, there is a necessity to provide infrastructure that the MV would be able to reliably read and use
for safe guidance. Simultaneously, human drivers should also benefit the same infrastructure. Both camera and Light Detection And
Ranging (LiDAR) equipment are frequently employed by the MV; radar is also used, but for slightly different purposes [30]. While the
progress within the field of the algorithm development for MV seems exponential, most of the sensors continue to rely on the detection
of road markings (RM) [25]. RM are advantageous from this perspective because of their usability for human drivers and for MV [9];
their current presence on the roads and the existence of reliable technologies emphasise their usefulness. In the foreseeable future,

* Corresponding author.
E-mail address: tomasz.burghardt@swarco.com (T.E. Burghardt).

https://doi.org/10.1016/j.cscm.2023.e02048
Received 6 December 2022; Received in revised form 21 March 2023; Accepted 5 April 2023
Available online 5 April 2023
2214-5095/© 2023 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
T.E. Burghardt et al. Case Studies in Construction Materials 18 (2023) e02048

there does not appear to be an alternative to their utilisation [32], even though substitutes are being sought and explored.
The key parameter in visibility of RM is contrast between the marking and the neighbouring roadway surface – not only human
vision depends on it [6,43], but also MV that is camera-based [40]. LiDAR-based MV also perceives RM; reflective intensity of RM
distinguishes them from the road surface background, which also can be considered as kind of a contrast ratio [33,47]. The presence of
glare (from man-made sources like oncoming vehicles or from sunshine), which reduces contrast ratio, capable even to reduce visibility
to nil, was consistently reported as one of the biggest challenges [7,45].
There is a plethora of research related to the MV and sensors [31], but the correlation of the quality and type of RM with their
visibility for MV remains weakly researched [9,34]. There appears to be an overall disconnection between engineers working in the
field of MV and the professionals involved with RM; the software developers of MV may seem to be assuming that RM are just part of
the road [29]. This apparent lack of mutual understanding and cooperation between engineers responsible for the development of road
features recognition software and those working in the field of road markings might be one of the issues hampering development or
ineffectively directing the attention of scientists. This may have led to a general impression that RM are not sufficiently reliable and
consistent sources of information for MV. Recently, it was hypothesised that several crashes caused by vehicles operating in
semi-automated mode could have been avoided if RM were in good condition [15]; nonetheless, it remains unknown what would
constitute sufficiently good conditions for the utilised MV sensors.
The purpose of this experiment was quantification of visibility of two commonly used types of RM that in the industry are known to
represent the extremes: ‘standard’ flat stripes with typical retroreflectivity and ‘premium’ structured stripes with high retroreflectivity;
these RM were applied side-by-side as a pedestrian crossing ‘zebra’ stripes. They were tested stationarily with a camera and a LiDAR in
the daytime and at night time, under both dry and wet conditions, and in the presence or the absence of glare from an oncoming
vehicle. The measured recovery of visibility from flooding is a new perspective, not addressed so far in the literature related to RM.
Evaluation of the visibility was done through evaluation of response from LiDAR and through assessment of contrast ratio measured
from pictures taken with a camera. Importantly, unlike in vast majority of current research, this evaluation combined the knowledge of
both road marking professionals familiar with the materials and their capabilities and data processing engineers working with the MV
equipment. The outcome is to serve not only as a demonstration of the differences between RM and their effect on the visibility for both
human drivers and MV sensors, but can also be utilised by the engineers developing MV algorithms, by scientists researching the
inadequacies of the current MV capabilities, and by road administrators who can select appropriate RM to maximise their visibility and
thus increase road safety. To maintain focus on the results from this case study, review of prior literature or discussions related to the
topics of the equipment, the algorithms for RM recognition, or the materials other than those that were evaluated were considered as
being out of scope.

2. Background

RM are a special type of industrial coatings with a unique feature of comprising two vastly dissimilar layers: a paint layer and a glass
beads layer that must work together to furnish the functional product [39]. The paint layer adheres to the roadway surface, gives the
desired colour, provides adhesion surface for glass beads and simultaneously the surface for retroreflectivity; on the top of it is applied
a layer of partially embedded drop-on glass beads, which provide retroreflection and simultaneously protect the underlaying paint
layer from abrasion [13]. There are many materials that can be used for the paint layer, some of them not even being paints but
polymeric or non-polymeric masses [2] – all of them must be reflectorised with glass beads; their various types were recently described
[46].
The paint layer can be applied in form of flat lines (known in the industry as Type I markings); they are the most commonly used
type of RM, easiest to prepare, to apply, and often with the lowest unit price. Unfortunately, under wet conditions the surface of flat
lines may become submerged, which leads to significant decrease in visibility caused by the loss of contrast against the roadway
surface and in most cases also low retroreflectivity. To alleviate this issue, RM of Type II were developed; these are applied as
structures, either random or regular, with relatively high vertical profile. The structures provide vibroacoustic effect warning drivers
about the departure from traffic lane, shelter some of the glass beads from abrasion by passing vehicles, and simultaneously facilitate
water drainage to improve visibility during wet conditions. While all of these parameters of structured RM were reported as increasing
road safety [24], their effect on MV equipment reliability has not been assessed so far; this article is partially filling this knowledge gap.
Amongst many parameters of RM, the most important for visibility are retroreflectivity (measured through coefficient of retro­
reflected luminance, RL – under dry conditions or RW – under wet conditions) and daytime visibility (measured through luminance
coefficient under diffuse illumination, Qd). It is known for more than a decade that RL of RM plays critical role in the reliability of
steering by MV [21,22], particularly in the darkness on unlit roads when RL is enhancing the visibility. The effects of the RM choice on
visibility for MV through camera and LiDAR contrast ratio under various conditions were recently analysed under laboratory con­
ditions [10]; RL was again shown as the one of key properties, even if Qd played important role under some conditions. Nonetheless,
there is no agreement as to what minimum RL, RW, and Qd would be needed for reliable steering; it can be hypothesised that the main
issue in the lack of consistency is the influence of illumination, the necessity to additionally consider the edge detection, and the
continuity of preview in dynamic experiments.

2
T.E. Burghardt et al. Case Studies in Construction Materials 18 (2023) e02048

3. Methodology and materials

3.1. Pedestrian crossing

Six pedestrian crossing ‘zebra’ stripes, each 0.5 m wide, 3.0 m long, with 0.5 m gaps between them were installed on a newly laid
asphalt road. Three of them, on one side of the street, were ‘standard’ flat lines and another three ‘premium’ structured markings;
selected pertinent details about the materials and the measured properties are provided in Table 1. All of the utilised products were
procured commercially; their application methods and the used quantities were done as customarily in the industry for this types of
jobs. The properties of the tested RM (RL, RW, and Qd) were measured by the authors according to the procedures described in the
standard EN 1436 [20]. Whereas generalisation based on assessment of only three stripes of RM can bring concerns related to small
sample size, one should keep in mind that even smaller samples are utilised for official homologations in Germany or Austria; hence,
the robustness and reliability of the testing procedures were deemed reliable. Based on professional experience, uncertainties of the
measurements, quantified through standard deviations, were typical for these types of RM and represent natural variations that can be
measured in the field [3]. Analysis was done only on the ‘zebra’ stripes; the edge lines were excluded from the analyses due to different
visibility angles, which could skew the results; they were marked with the ‘standard’ RM.
The application of moisture on the pedestrian crossings was done through manual rapid flooding of all of the stripes simulta­
neously, each with 10 ± 1 dm3 of water. This was to imitate the worst-case scenario of a sudden downpour and permitted for mea­
surements of the recovery of visibility as the applied water drained from the markings. As an additional confounder was added glare
from an oncoming car equipped with LED headlights, which was positioned 20 m from the analysed area. Approximately 0.5 h after the
first measurements were taken (surface was permitted to partially dry), flooding was repeated but without the glare source. Illumi­
nation was not measured.
The tests were done in a static mode, which is the first step to compare the effects of dissimilar materials on MV response. One must
note that the purpose of this experiment was not evaluation of algorithms for object detection, but testing the influence of various
materials on the sensors’ responses. Hence, to make data comparable and the results repeatable, confounders associated with
movement of vehicle had to be minimised. While for LiDAR the existence of a difference in static or driving mode could be argued, in
case of a camcorder factors like simple vibration of the vehicle could cause blurring of the image and therefore affect the performance
of the algorithms. With the vehicle moving, the number of necessary tests would have to be significantly increased in order to
compensate for the random influences caused by the vehicle movements, slight variations in direction, surrounding illumination, etc.
For data collection, the MV equipment was installed at an external rack mounted in front of a vehicle positioned 20 m from the
analysed pedestrian crossing and throughout the test it had low beam LED headlights turned on. The distances of 20 m were selected
based on the engineering knowledge related to the MV algorithms as being typical to establish target visibility and trajectory planning.
Data collection was done from single frames extracted from the continuous recordings collected with the following equipment:

• LiDAR: 1550 nm, 0.07◦ horizontal and 0.03◦ vertical resolution, 120◦ horizontal and 0 − 30◦ vertical field of view, up to 640 lines
per second; recording was done at 10 Hz from the equipment positioned 0.6 m above ground, and
• Camcorder: 2048 × 1536 pixels giving 34◦ field of view, with each pixel size 3.45 × 3.45 µm, sensor CMOS IMX265 7.1 × 5.3 mm;
recording was done at 36 frames per second from the equipment positioned 0.6 m above ground.

3.2. Data processing

Processing of the data was done by manually separating the regions of interest (ROI) that comprised each ‘zebra’ stripe (marked
area) against the neighbouring road surface (background area). Since the images were taken stationarily, the ROI were the same for all
of the images. The ROI comprised only areas that were solely within the marked area or a background, with exclusion of the edges to
minimise bias. Simultaneously, disregarding the edge areas was an additional confounder related to visual recognition as it strongly
relies on sharp edges.

Table 1
Applied road marking systems.
Road marking type Flat lines (‘standard’); Type I Structured markings (‘premium’); Type II

Application method Manual spraying to obtain uniform markings Hand-operated roller to obtain random structure
Material Sprayed cold plastic Cold plastic
Applied material quantity (paint layer) 0.5 ± 0.05 kg/m2 (uniform dry layer thickness 3.0 ± 0.05 kg/m2 (dry layer thickness varying, circa
circa 0.45 mm) 0.5 − 2.5 mm)
Glass beads type Standard, refractive index 1.50; diameter range Premium, refractive index 1.65; diameter range 300 −
200 − 800 µm 850 µm
Glass beads drop-on quantity 0.4 ± 0.05 kg/m2 0.4 ± 0.05 kg/m2
Retroreflectivity under dry conditions (RL) 236 (29) 708 (64)
[mcd/m2/lx]a
Retroreflectivity under wet conditions (RW) 29 (10) 130 (47)
[mcd/m2/lx]a
Daytime visibility (Qd) [mcd/m2/lx]a 192 (14) 141 (13)
a
Average values for all stripes; for each stripe, five measurements were taken. Standard deviations given in parentheses.

3
T.E. Burghardt et al. Case Studies in Construction Materials 18 (2023) e02048

In case of camera images, the Y′ component (directly representing luma, which for all of the practical purposes in this manuscript
can be considered as tantamount to luminance) from Y′ CbCr colour space was extracted and its intensities for each value in the 0 − 255
range were used for analysis. The intensities were normalised, so in each case their sum was 1.0. To obtain average luminance for each
ROI, Y′ intensities of the individual 256 values were averaged according to Eq. 1, where: L – average luminance of ROI (later split into
Lm – marked area ROI and Lb – background area ROI), n – given value, and Y′ – intensity at the given value.
( )
1 ∑
i=255
L= ′
ni Yi (1)
256 i=0

Surprisingly, the choice of the optimum equation to represent contrast ratio for the use of MV in AV was not researched so far.
Simple contrast ratio, used to measure the difference between the maximum and minimum luminance, which is used in photography,
was deemed inappropriate because in some cases it can give unrealistically high values and inadequately represent the perceptual
recognition; nonetheless, some authors working in the field of MV and RM were using simple contrast and reported that values higher
than 2.0 (preferably over 3.0) were needed [16]. A better alternative seemed to be Weber contrast, which reasonably well represents
human vision as it measures the ratio of maximum luminance to its difference, or Michelson contrast (a peak-to-peak, or a modulation
contrast, describing a sinusoidal stimulus) providing signal-to-noise information. One has to keep in mind that while equal Weber
contrast represents equal absolute magnitude of change from the background luminance, equal Michelson contrast represents equal
percentage changes; this affects human perception [5], and was also reported to create difficulties for MV algorithms [35]. Still, other
contrast equation or their combination may give yet better representation of the MV needs [38]. In addition to contrast, one must never
forget about edge recognition, which may permit for reliable steering in cases of low contrast but also may cause erroneous assign­
ments for MV [8].
Contrast ratio was calculated according to two equations: Eq. 2 (Weber contrast) and Eq. 3 (Michelson contrast), where CRW –
Weber contrast ratio, CRM – Michelson contrast ratio, Lm – average luminance of the marked area ROI, Lb – average luminance of the
background ROI. The outcome from the three stripes of each RM was averaged. No corrections were applied to account for noise.
Lm − Lb
CRW = (2)
Lb

Lm − Lb
CRM = (3)
Lm + Lb
Processing of both LiDAR and camcorder data was done with an in-house proprietary software. Exemplary images used for the
analyses are furnished as Figs. 1–5 (in all of the cases, ‘standard’ flat line RM comprise the three stripes on the left and the ‘premium’
structured RM the three stripes on the right). One should note in Fig. 5 small areas (on the leftmost and on the rightmost stripes) that
were not fully flooded; these areas were excluded from the ROI and as such did not affect the measured contrast ratios. Exemplary
LiDAR points cloud image is provided in Fig. 6; noted should be intensity of the edge lines similar to furnished by the flat lines stripes –
this can serve as confirmation that they were the same indeed.

4. Results

4.1. Camera contrast ratio

The measured average effects of glare intensity (no glare, low beams, and high beams of the oncoming vehicle) are given in Table 2
and are charted in Fig. 7. With dry surface, the glare played only minimal role because for the analysed scenario it was at an angle that
did not cause blinding; the topic of the influence of glare source, its location and strength on visibility of RM has not been explored so
far – no references could be found in the literature.
Upon wetting of the surface, one must first note the increase in contrast measured with wet surface (in the absence of glare) – this is
a known phenomenon caused by the decrease in the luminance of the asphalt background due to the presence of water and its

Fig. 1. Daytime, dry conditions. No difference visible or measured through camera contrast ratio.

4
T.E. Burghardt et al. Case Studies in Construction Materials 18 (2023) e02048

Fig. 2. Night time, dry conditions. No difference visible or measured through camera contrast ratio.

Fig. 3. Night time, wet conditions (recovery for 30 s). The difference in RW is clearly visible and was measured through camera contrast ratio.

Fig. 4. Night time, dry conditions, glare from high beams of oncoming vehicle positioned 20 m away. No difference visible or measured through
camera contrast ratio.

Fig. 5. Night time, wet conditions (recovery for 30 s), glare from high beams of oncoming vehicle positioned 20 m away. The difference in RW is
clearly visible and was measured through camera contrast ratio.

5
T.E. Burghardt et al. Case Studies in Construction Materials 18 (2023) e02048

Fig. 6. LiDAR points cloud. Dry conditions. Visible difference in intensity caused by retroreflectivity (yellowish lines − higher intensity, teal lines −
lower intensity).

Table 2
Effect of night time glare intensity on average camera contrast ratio.
Glare intensity No glare Present, low intensity (low beams) Present, high intensity (high beams)1

Road marking type Flat line Structured Flat line Structured Flat line Structured

Contrast ratio equation (2) (3) (2) (3) (2) (3) (2) (3) (2) (3) (2) (3)

Dry surface 0.7 0.25 0.9 0.29 0.6 0.23 0.7 0.25 0.6 0.22 0.7 0.27
Wet surface, < 10 s recovery 1.0 0.32 2.1 0.48 1.1 0.28 1.5 0.36 0.1 0.05 0.2 0.08
1
Two centre stripes, which were obscured by glare, were excluded from the given averages.

Fig. 7. Effect of glare intensity on (a) Weber contrast ratio, and (b) Michelson contrast ratio.

capability to reflect light [27]. One should note that this effect may not occur in case of concrete roadways or the use of asphalt with
very low porosity, so generalisations are not advised. Low intensity glare (low beams) played only limited role and this scenario was
not explored any further. The high intensity glare (high beams) had really detrimental effect: the reflection from wet surface caused the
centre stripes to be completely obscured, as can be seen in Fig. 4 and in Fig. 5. These stripes were excluded from the measured averages
given in Table 2 because neither camera-based MV nor humans can benefit from perception in that area; in addition, because it was
reported that drivers tend to look away from glare source to avoid discomfort [45], elimination of the obscured area in the centre
would give more weight to the peripheral stripes that could still be perceived.
Interesting and at the first sight contradictory situation was encountered in analysis of the situation shown in Fig. 5 (wet surface
during glare): despite very low contrast ratio measured with the structured RM, they were visible while the flat lines remained un­
differentiated from the background. This was solely caused by using a fixed selection of the ROI, which excluded the edges that can
play very important role for human and machine perception [44]. As shown exemplarily in Fig. 8 (which is Fig. 5 after performing edge
detection according to Sobel algorithm), mostly the edges of the structured markings were perceived; in their absence, the flat line
markings could not be distinguished from the background. Hence, contrast ratios of the edges should be investigated as they became
more prominent visibility cues while contrast ratio of the larger marked areas became relatively irrelevant. Because for this experiment

6
T.E. Burghardt et al. Case Studies in Construction Materials 18 (2023) e02048

Fig. 8. Night time, wet conditions (recovery for 30 s), glare; edge detection according to Sobel. Edges in ‘standard’ flat lines stripes were not visible,
but were clear with ‘premium’ structured stripes.

we were not concerned with edge detection, excluding them was appropriate; however, this case emphasises the necessity to use
information from many sources and processed in various methods if one is to achieve reliable operation of MV.
Contrast ratios measured during recovery from flooding, calculated according to the Eq. 2 and Eq. 3, are shown in Table 3. In the
presence of glare from oncoming vehicle (Cf. Fig. 5), both standard and premium materials performed poorly, which emphasises the
critical importance of this confounder on the perception of RM for both human drivers and camera-based MV equipment. Importantly,
no meaningful recovery was observed. The measured differences between flat line and structured RM were not statistically significant
within 95% confidence level (Student’s t-test, paired), with p-values 0.3442 and 0.2725 correspondingly for Weber and Michelson
contrast.
In the absence of the glare, the structured RM facilitated water drainage, which resulted in contrast ratio higher than was measured
under dry conditions: the moist asphalt surface decreased the luminance of the background [27], effectively affecting the contrast. This
phenomenon was not happening in case of flat lines RM. Statistical analysis (Student’s t-test, paired) has shown the differences be­
tween the ‘standard’ flat line and the ‘premium’ structured RM to be significant within 95% confidence level, with p-value 0.0093 and
0.0078 correspondingly for Weber and Michelson contrast; the difference in visibility can be seen in Fig. 3.
One should observe some differences in data shown in Table 2 as compared to Table 3 (also the absence of data for dry conditions at
night); this occurred because images subjected to the analyses were taken on different occasions when illumination, despite being
subjectively perceived as the same, was sufficiently different to cause dissimilar contrast ratio values. This measured and reported
differences emphasise the uncertainty associated with using solely contrast ratios as they strongly depend on both the illumination and
the background surface characteristics [43]; surprisingly, in numerous publications related to MV this natural phenomenon tends to be
ignored.
The exploratory calculation of contrast ratio according to Michelson formula (Eq. 3) demonstrated that under certain conditions it
may be more useful for MV than Weber contrast (Eq. 2). As shown in Fig. 3, the visual distinction between the background and the
surfaces marked with structured markings was clear; nonetheless, while CRW= 8.7 (11 times larger than measured for flat lines)
appears exaggerated, CRM= 0.81 (3 times larger than for flat lines) seems more reasonable. Because such unreasonably high Weber
contrast value was caused by the background becoming darker, this effect was somewhat moderated in this case through the
employment of Michelson contrast. Nonetheless, this is a broad and complex issue that cannot be solved without further extensive
evaluation that belongs to the field of recognition algorithms and not the materials science presented herein; at present, let this
provided data serve as an example of another uncertainty in the field of MV.

Table 3
Average camera contrast ratio.
Glare Present, high intensity1 Absent

Road marking type Flat line Structured Flat line Structured

Contrast ratio equation (2) (3) (2) (3) (2) (3) (2) (3)

Daytime – dry – – – – 1.3 0.39 1.2 0.38


Night time – dry 3.0 0.56 2.1 0.50 –2 –2 –2 –2
Night time – moist – – – – 0.6 0.21 7.5 0.78
Night time – wet (10 s recovery) 0.1 0.01 0.1 0.06 0.7 0.25 5.9 0.74
Night time – wet (20 s recovery) 0.1 0.02 0.1 0.06 0.6 0.21 7.5 0.78
Night time – wet (30 s recovery) 0.1 0.05 0.2 0.06 0.8 0.27 8.7 0.81
Night time – wet (40 s recovery) 0.1 0.06 0.2 0.07 1.0 0.33 9.3 0.82
Night time – wet (50 s recovery) 0.1 0.06 0.2 0.07 1.2 0.37 9.1 0.82
1
The stripes obscured by glare were included.
2
The data is not available because of the experimental set-up with measurements in the absence of glare done on surface that was previously
wetted. Note that data furnished in Table 2 for the same conditions was collected on different occasion; hence, apparent differences.

7
T.E. Burghardt et al. Case Studies in Construction Materials 18 (2023) e02048

4.2. LiDAR response

The response of LiDAR is given in Table 4; the intensity changes upon flooding the tested surface under glare conditions are
visualised in Fig. 9. Whereas the number of detected points was paralleling the changes in their intensity, one must keep in mind that
the initial differences in the measured intensity were most likely caused by higher RL of the structured stripes; using LiDAR to detect RL
was already reported as usable for the assessment of RM [17,42]. The differences between the evaluated materials are obvious: while
after flooding it took 30 s for the flat line stripes to provide measurable LiDAR response again, the response recorded from the
structured stripes never dropped to nil. Basic statistical analysis (Student’s t-test, two-tailed distribution, paired) indicated statistical
significance at 95% confidence level, with p-values 0.0033 and 0.0141, correspondingly for the number of detected points and the
measured intensity.
Because glare generally does not affect LiDAR response, it was quite surprising to measure the differences between the conditions of
its presence and absence, even if the relative outcome remained constant. The explanation, however, was simple and associated with
the instrument’s software and with the test protocol. The test was done firstly in the presence of glare and on road surface that was dry;
upon the drop of measured intensity, the LiDAR internal software automatically adjusted itself to deliver the highest contrast. This
internal recalibration persisted for the test without glare; hence, the LiDAR output level was different and indicated the effect of glare,
which in reality was only an artefact associated with the instrument set-up. Whereas it was possible to re-adjust the LiDAR software to
override the automatic adjustment, the authors have chosen to give this as an example for the necessity of not only appropriate se­
lection of test conditions, but foremost to emphasise the obligation of any scientist to doubt any results and search for alternative
plausible explanations of the measured effects.

5. Discussion

Outcome from this study matched the expectations of professional knowledge; in this experiment these anticipations were
quantified for the first time. Whereas obtention of adequate contrast ratio under dry conditions was not difficult, the conditions of
wetness resulted in its dramatic lowering to nil in case of flat line RM (Type I). Much better visibility was measured with structured RM
(Type II), where moisture drainage facilitated by higher vertical profile allowed for obtention of higher contrast ratio and improved
edge visibility. Additional role played slightly higher refractive index of glass beads utilised in the ‘premium’ structured RM: their use
not only resulted in higher RW [41], but also augmented response of the LiDAR.
Amongst main advantages of using LiDAR is its capability to overcome the issue of glare. A photosensitive sensor in a camera can
easily become oversaturated (too much energy is transferred to neighbouring pixels), which causes the features of the objects to be
indistinguishable. Because LIDAR sensor is being able to store a much larger charge and is time sensitive to the signal, the data is much
cleaner. Nonetheless, while LIDAR is a reliable solution during the conditions of glare, it has its weaknesses, mainly quite low reso­
lution and the type of data it produces. Consequently, while either camera alone or LiDAR alone may provide sufficient input to the AV
algorithms, their combination seems to be the preferred solution because input from each data sensor may be used to verify and
supplement the other data and to provide a different perspective. This redundancy is needed to minimise the risk of accidents caused by
AV due to incorrect or missing data. Processing of such simultaneous data inputs is not a topic of this paper.
The reliability of AV must be exquisite for their full acceptance [4] and RM are amongst the key road infrastructure features
potentially affecting the reliability. The results presented herein, particularly the use of ‘premium’ structured RM that provided high RL
and facilitated water drainage with a simultaneous employment of camera and LiDAR, may be a feasible solution to the issue of RM
recognition by MV under the conditions of wetness and in the presence of glare. A qualification test of RM for MV was postulated [18];
nonetheless, simple tests that are already performed – measurements of RL, RW, and Qd – should be adequate from the perspective of
RM maintenance. It is important to keep in mind that correlations between these parameters and the visibility of RM are not simple as
they strongly depend on the external factors like lighting, glare, weather, etc., but also the sensor selection; this was also confirmed in
our related prior laboratory study [10]. Furthermore, different sensors produce different results and these results cannot be compared
with each other directly. While contrast was measured only for camera data, reflectivity is a value coming from LIDAR data.
Nonetheless, one must always keep in mind the ‘ironies of automation’ [36]; the analysed visibility situation can serve as an
excellent example: while humans could steer their vehicles because of the multitude of additional perceptual cues, the MV system
could become blind and as such unreliable. To facilitate transfer of some of the driving tasks to AV, appropriate infrastructure features
must be installed for consistent recognition by MV. While the MV is nowadays installed as a support system in most of modern vehicles,
the human driver must be able to take over steering at any time; during such situations, exquisite visibility of RM is even more critical.
A human driver disengaged from steering must be able to immediately collect all of the perceptual cues necessary for appropriate
trajectory planning upon taking over control of the vehicle.
Amongst apparent limitations of this study one could consider the choice of MV equipment and only basic post-processing;
simultaneously this can be considered as advantageous because of the simplicity and ready availability of the technique. Testing of
additional equipment or algorithms was beyond the scope of this research; similarly, any modifications of the hardware or software
were never considered. Because this work was intended to provide the results from two extremes related to visibility and not to analyse
scenarios in-between, only the selected materials were tested. Nonetheless, testing of the combination of flat line RM with ‘premium’
glass beads would depart from the reality because such combinations are used in the field very seldom, if ever (based on information
from the industry, such RM comprise less than 5% of all applied). Based on results from prior laboratory experiment [10], it could be
anticipated that during rain the structured RM would remain better visible regardless of the used glass beads because of their capability
to facilitate moisture drainage; LiDAR response would be lower due to lesser RW.

8
T.E. Burghardt et al. Case Studies in Construction Materials 18 (2023) e02048

Table 4
LiDAR response.
Glare Present, high beams Absenta

Road marking type Flat line Structured Flat line Structured

Measurement type Detected points Intensity Detected points Intensity Detected points Intensity Detected points Intensity

Initial (dry) 77 40.5 61 85.3 – – – –


Initial (moist) – – – – 52 1.6 41 15.7
Flooding: 0 s 0 0.0 33 1.3 14 0.8 17 8.0
Recovery: 10 s 0 0.0 38 2.6 0 0.0 38 2.6
Recovery: 20 s 0 0.0 40 4.0 0 0.0 38 4.5
Recovery: 30 s 0 0.0 42 9.4 5 0.3 39 6.6
Recovery: 40 s 0 0.0 42 9.4 11 0.4 41 8.0
Recovery: 50 s 1 0.1 43 8.4 11 0.4 41 8.0
Recovery: 60 s 3 0.2 86 9.7 n/a n/a n/a n/a
Recovery: 70 s 16 0.2 42 11.9 n/a n/a n/a n/a
Recovery: 200 s 40 1.9 41 28.3 n/a n/a n/a n/a
a
See text for the explanation of the reason for the differences as compared to the glare conditions.

Fig. 9. LiDAR intensity (glare conditions) upon flooding of surface.

Additional work related to the correlation between the choice of materials for RM and the response of MV equipment under various
lighting conditions and climatic conditions is needed, so the selection of materials for the benefit of both human and machine drivers
could be done based on quantified data. Of particular interest should be identification of the thresholds for detection – by human
drivers of different ages and by various MV; selected findings related to Visibility Level were recently reviewed [6]. The effects of
structure types and colour of RM has not been evaluated so far; for example, it is unknown whether regularly spaced dots with very
high vertical profile would provide additional benefits [11]. Recently, the effect of diffuse sunshine glare and wet roadway surface was
shown to be bringing contrast ratio to nil even in case of structured RM [8]; it would be interesting to know whether under such
conditions LiDAR would provide adequate response. The effect of concrete roadway surface, which tends to be much lighter than
asphalt and as such capable of furnishing only mediocre contrast ratio, must be thoroughly assessed to estimate the limits of the
equipment’s response in unfavourable conditions. The tests presented herein were performed on newly applied RM, but one must
always consider that RM are a deteriorating system that requires periodic renewals (not replacement); hence, prolonging of the period
when the properties are adequate (i.e. within their intended functional service life) should play profound role. It was reported that
initial RL cannot be reliably used as a predictor of long-term performance [12,23] and that both the choice of the paint and the drop-on
glass beads significantly affected the properties [14].
Whereas the main focus of this article was visibility of RM for MV, one should never forget about human drivers – they are he first to
benefit from increased contrast ratio and improved visibility. It is therefore quite strange that the recent legislative push in many
countries to prepare road infrastructure for AV appears to have minimised the role of the primary beneficiaries. While observing that
structured RM with high RL furnished better results in the experiment described herein, one must consider that increase in RL was
correlated with a decrease in accident rate [1] and increased ease of driving [19,26]. Because of better visibility, they may also
constitute a factor improving the mobility of elderly drivers, whose sight acuity decreased [28,37]. These important social factors
should never be forgotten.

9
T.E. Burghardt et al. Case Studies in Construction Materials 18 (2023) e02048

6. Conclusions

The presented analysis has shown that the choice of RM type played profound role on their visibility, particularly under the
conditions of moisture at night. The critical issue remains glare, which brings the visibility by camera (which is almost tantamount to
visibility by human drivers) to almost nil, regardless of the used road marking system. Glare conditions can be overcome by LiDAR, as
long as the RM remain retroreflective, which was confirmed: while ‘premium’ structured lines (Type II RM) remained visible to LiDAR
due to their surface facilitating moisture drainage and higher RW, recognition of the ‘standard’ flat lines (Type I RM) dropped to zero.
Hence, a combination of MV equipment comprising at least a camera and a LiDAR, supplementing each other, is at present seen as
necessary for reliable steering of AV. Even though the assessment was done only on two classes of RM and in stationary conditions,
these limitations were not seen as affecting the overall picture, but instead permitted on concentrating on the important visibility
parameters. It must be emphasised again that RM comprise dissimilar materials and performance classes; as was demonstrated, there
are commercially available RM solutions that can furnish improved visibility for both human and machine drivers.
Because the effectiveness of advanced driver assistance systems and the success of the anticipated introduction of AV depend on the
quality of RM, it is of utmost importance to maintain them at high standards. A co-operation and mutual understanding of needs,
capabilities, and limitations is necessary between all of the stakeholders: manufacturers and applicators of RM, engineers developing
the hardware and software for AV, and the road administrators responsible for RM maintenance. Clearly visible RM are equally
necessary for human drivers, so this should be a current action item that cannot be delayed. Everybody ought to strive to furnish high
quality RM for the ultimate goal of their installation – road safety.

Declaration of Competing Interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing in­
terests: Tomasz E. Burghardt reports article publishing charges was provided by M. Swarovski GmbH. Tomasz E. Burghardt, Bernhard
Helmreich, and Friedrich Wiesinger report a relationship with M. Swarovski GmbH that includes: employment.

Data availability

Data will be made available on request.

References

[1] R.E. Avelar, P.J. Carlson, Link between pavement marking retroreflectivity and night crashes on Michigan two-lane highways, Transp. Res. Rec.: J. Transp. Res.
Board 2404 (2014) 59–67, https://doi.org/10.3141/2404-07.
[2] D. Babić, T.E. Burghardt, D. Babić, Application and characteristics of waterborne road marking paint, Int. J. Traffic Transp. Eng. 5 (2015) 150–169, https://doi.
org/10.7708/ijtte.2015.5(2).06.
[3] D. Babić, M. Fiolić, D. Žilioniene, Evaluation of static and dynamic method for measuring retroreflection of road markings, Građevinar 69 (10) (2017) 907–914,
https://doi.org/10.14256/JCE.2010.2017.
[4] F. Becker, K.W. Axhausen, Literature review on surveys investigating the acceptance of automated vehicles, Transportation 44 (6) (2017) 1293–1306, https://
doi.org/10.1007/s11116-017-9808-9.
[5] A. Beghdadi, M.A. Qureshi, S.A. Amirshahi, A. Chetouani, M. Pedersen, A critical analysis on perceptual contrast and its use in visual information analysis and
processing, IEEE Access 8 (2020) 156929–156953, https://doi.org/10.1109/ACCESS.2020.3019350.
[6] R. Brémond, Visual performance models in road lighting: a historical perspective, Leukos 17 (3) (2021) 212–241, https://doi.org/10.1080/
15502724.2019.1708204.
[7] T.E. Burghardt, A. Pashkevich, Contrast ratio of road markings in Poland: evaluation for machine vision applications based on naturalistic driving study, in:
A. Akhnoukh, K. Kaloush, M. Elabyad, B. Halleman, N. Erian, S. Enmon II, S. Henry (Eds.), Advances in Road Infrastructure and Mobility. Proceedings of the
18th International Road Federation World Meeting & Exhibition, Dubai 2021, Springer, Cham, Switzerland, 2021, pp. 676–690, https://doi.org/10.1007/978-
3-030-79801-7_49.
[8] Burghardt, T.E., Pashkevich, A., 2022. Camera contrast ratio of road markings at dual carriageway roads. Transportation Research Procedia, in press. Presented
at Transportation Research Arena 2022; Lisbon, Portugal, 14–17 November 2022.
[9] T.E. Burghardt, H. Mosböck, A. Pashkevich, M. Fiolić, Horizontal road markings for human and machine vision, Transp. Res. Procedia 48 (2020) 3622–3633,
https://doi.org/10.1016/j.trpro.2020.08.089.
[10] T.E. Burghardt, R. Popp, B. Helmreich, T. Reiter, G. Böhm, G. Pitterle, M. Artman, Visibility of various road markings for machine vision, Case Stud. Constr.
Mater. 15 (2021), e00579, https://doi.org/10.1016/j.cscm.2021.e00579.
[11] T.E. Burghardt, E. Maki, A. Pashkevich, Yellow thermoplastic road markings with high retroreflectivity: demonstration study in Texas, Case Stud. Constr. Mater.
14 (2021), e00539, https://doi.org/10.1016/j.cscm.2021.e00539.
[12] T.E. Burghardt, D. Babić, A. Pashkevich, Performance and environmental assessment of prefabricated retroreflective spots for road marking, Case Stud. Constr.
Mater. 15 (2021), e00555, https://doi.org/10.1016/j.cscm.2021.e00555.
[13] T.E. Burghardt, A. Pashkevich, D. Babić, H. Mosböck, D. Babić, L. Żakowska, Microplastics and road markings: the role of glass beads and loss estimation,
Transp. Res. Part D: Transp. Environ. 102 (2022), 103123, https://doi.org/10.1016/j.trd.2021.103123.
[14] T.E. Burghardt, D. Babić, A. Pashkevich, Sustainability of thin layer road markings based on their service life, Transp. Res. Part D: Transp. Environ. 109 (2022),
103339, https://doi.org/10.1016/j.trd.2022.103339.
[15] S.C. Calvert, G. Mecacci, B. van Arem, F.S. de Sio, D.D. Heikoop, M. Hagenzieker, Gaps in the control of automated vehicles on roads, IEEE Intell. Transp. Syst.
Mag. 13 (2020) 146–153, https://doi.org/10.1109/mits.2019.2926278.
[16] Carlson, P.J., Poorsartep, M., 2017. Enhancing the roadway physical infrastructure for advanced vehicle technologies: a case study in pavement markings for
machine vision and a road map toward a better understanding. In: Proceedings of Transportation Research Board 96th Annual Meeting; Washington, District of
Columbia, United States, 8-12 January 2017, paper 17–06250.
[17] E. Che, M.J. Olsen, C.E. Parrish, J. Jung, Pavement marking retroreflectivity estimation and evaluation using mobile LiDAR data, Photogramm. Eng. Remote
Sens. 85 (8) (2019) 573–583, https://doi.org/10.14358/PERS.85.8.573.
[18] J. Claybrook, S. Kildare, Autonomous vehicles: no driver… no regulation? Science 361 (6397) (2018) 36–37, https://doi.org/10.1126/science.aau2715.

10
T.E. Burghardt et al. Case Studies in Construction Materials 18 (2023) e02048

[19] K. Diamandouros, M. Gatscha, Rainvision: the impact of road markings on driver behaviour-wet night visibility, Transp. Res. Procedia 14 (2016) 4344–4353,
https://doi.org/10.1016/j.trpro.2016.05.356.
[20] European standard EN 1436. Road marking materials — road marking performance for road users and test methods. European Committee for Standardization:
Brussels, Belgium, 2018.
[21] M. Hadi, P. Sinha, Effect of pavement marking retroreflectivity on the performance of vision-based lane departure warning systems, J. Intell. Transp. Syst. 15 (1)
(2011) 42–51, https://doi.org/10.1080/15472450.2011.544587.
[22] M. Hadi, P. Sinha, J. Easterling, Effect of environmental conditions on performance of image recognition-based lane departure warning system, Transp. Res.
Rec.: J. Transp. Res. Board, 2000 (2007) 114–120, https://doi.org/10.3141/2000-14.
[23] M.H. Harun, S. Rosdi, M. Rosmani, High performance thermoplastic and cold applied plastic road markings: how long do they last, IOP Conf. Ser.: Mater. Sci.
Eng. 512 (2018), 012002, https://doi.org/10.1088/1757-899X/512/1/012002.
[24] J. Hatfield, S. Murphy, R.F.S. Job, W. Du, The effectiveness of audio-tactile lane-marking in reducing various types of crash: a review of evidence, template for
evaluation, and preliminary findings from Australia, Accid. Anal. Prev. 41 (3) (2009) 365–379, https://doi.org/10.1016/j.aap.2008.12.003.
[25] A.B. Hillel, R. Lerner, D. Levi, G. Raz, Recent progress in road and lane detection: a survey, Mach. Vis. Appl. 25 (3) (2014) 727–745, https://doi.org/10.1007/
s00138-011-0404-2.
[26] T. Horberry, J. Anderson, M.A. Regan, The possible safety benefits of enhanced road markings: a driving simulator evaluation, Transp. Res. Part F: Traffic
Psychol. Behav. 9 (1) (2006) 77–87, https://doi.org/10.1016/j.trf.2005.09.002.
[27] H.W. Jensen, J. Legakis, J. Dorsey, Rendering of Wet Materials, in: D. Lischinski, G.W. Larson (Eds.), Rendering Techniques’, 99, Springer, Vienna, 1999,
pp. 273–281, https://doi.org/10.1007/978-3-7091-6809-7_24.
[28] J.A. Kimlin, A.A. Black, N. Djaja, J.M. Wood, Development and validation of a vision and night driving questionnaire, Ophthalmic Physiol. Opt. 36 (4) (2016)
465–476, https://doi.org/10.1111/opo.12307.
[29] L. Kunze, T. Bruls, T. Suleymanov, P. Newman, Reading between the lanes: road layout reconstruction from partially segmented scenes. In: Proceedings of 21st
International Conference on Intelligent Transportation Systems; Maui, Hawaii, United States, 4-7 November 2018, pp. 401–408, https://doi.org.10.1109/
ITSC.2018.8569270.
[30] S. Kuutti, S. Fallah, K. Katsaros, M. Dianati, F. Mccullough, A. Mouzakitis, A survey of the state-of-the-art localization techniques and their potentials for
autonomous vehicle applications, IEEE Internet Things J. 5 (2) (2018) 829–846, https://doi.org/10.1109/JIOT.2018.2812300.
[31] E. Marti, M.A. de Miguel, F. Garcia, J. Perez, A review of sensor technologies for perception in automated driving, IEEE Intell. Transp. Syst. Mag. 11 (4) (2019)
94–108, https://doi.org/10.1109/MITS.2019.2907630.
[32] Mosböck, H., Burghardt, T.E., 2018. Horizontal road markings and autonomous driving – back from the future. In: Proceedings of 37th Annual Southern African
Transport Conference; Pretoria, South Africa, 9–12 July 2018, pp. 557–568; http://hdl.handle.net/2263/69562.
[33] S. Muckenhuber, K. Softic, A. Fuchs, G. Stettinger, D. Watzenig, Sensors for automated driving, in: S. Van Uytsel, D. Vasconcellos Vargas (Eds.), Autonomous
Vehicles. Perspectives in Law, Business and Innovation, Springer, Singapore, 2021, pp. 115–146, https://doi.org/10.1007/978-981-15-9255-3_6.
[34] Najeh, I., Bouillaut, L., Daucher, D., Redondin, M., 2020. Maintenance strategy for the road infrastructure for the autonomous vehicle. In: ESREL 2020-PSAM
15–30th European Safety and Reliability Conference and the 15th Probabilistic Safety Assessment and Management Conference; Venice, Italy, 1–5 November
2020.
[35] S. Nercessian, S.S. Agaian, K.A. Panetta, 2012. Multi-scale image enhancement using a second derivative-like measure of contrast. In: Proceedings SPIE Vol.
8295, Image Processing: Algorithms and Systems X and Parallel Processing for Imaging Applications II, 82950Q, https://doi.org.10.1117/12.906494.
[36] I.Y. Noy, D. Shinar, W.J. Horrey, Automated driving: safety blind spots, Saf. Sci. 102 (2018) 68–78, https://doi.org/10.1016/j.ssci.2017.07.018.
[37] R. Payyanadan, J. Lee, L. Grepo, Challenges for older drivers in urban, suburban, and rural settings, Geriatrics 3 (2) (2018) 14, https://doi.org/10.3390/
geriatrics3020014.
[38] E. Peli, Contrast in complex images, J. Opt. Soc. Am. A 7 (10) (1990) 2032–2040, https://doi.org/10.1364/JOSAA.7.002032.
[39] B.W. Pocock, C.C. Rhodes, Principles of glass-bead reflectorization, Highw. Res. Board Bull. 57 (1952) 32–48.
[40] B. Ranft, C. Stiller, The role of machine vision for intelligent vehicles, IEEE Trans. Intell. Veh. 1 (1) (2016) 8–19, https://doi.org/10.1109/TIV.2016.2551553.
[41] S.Y. Shin, J.I. Lee, W.J. Chung, Y.G. Choi, Correlations between refractive index and retroreflectance of glass beads for use in road-marking applications under
wet conditions, Curr. Opt. Photonics 3 (5) (2019) 423–428, https://doi.org/10.3807/COPP.2019.3.5.423.
[42] M. Soilán, D. González-Aguilera, A. del-Campo-Sánchez, D. Hernández-López, S. Del Pozo, Road marking degradation analysis using 3D point cloud data
acquired with a low-cost Mobile Mapping System, Autom. Constr. 141 (2022), 104446, https://doi.org/10.1016/j.autcon.2022.104446.
[43] R.M. Spieringhs, K. Smet, I. Heynderickx, P. Hanselaer, Road marking contrast threshold revisited, Leukos 18 (4) (2022) 493–512, https://doi.org/10.1080/
15502724.2021.1993893.
[44] R. Szeliski, Computer Vision: Algorithms and Applications, Springer Nature,, Cham, Switzerland, 2022, https://doi.org/10.1007/978-3-030-34372-9.
[45] J. Theeuwes, J.W. Alferdinck, M. Perel, Relation between glare and driving performance, Hum. Factors 44 (2002) 95–107, https://doi.org/10.1518/
0018720024494775.
[46] K.M. Wenzel, T.E. Burghardt, A. Pashkevich, W.A. Buckermann, Glass beads for road markings: surface damage and retroreflection decay study, Appl. Sci. 12 (4)
(2022) 2258, https://doi.org/10.3390/app12042258.
[47] Y. Xing, C. Lv, L. Chen, H. Wang, H. Wang, D. Cao, E. Velenis, F.-Y. Wang, Advances in vision-based lane detection: algorithms, integration, assessment, and
perspectives on ACP-based parallel vision, IEEE/CAA J. Autom. Sin. 5 (3) (2018) 645–661, https://doi.org/10.1109/JAS.2018.7511063.

11

You might also like