Enhancing LiDAR Reliability Through Utilization of Premium Road Marking Materials

Reliable traffic lane recognition (TLR) plays a crucial role in automated vehicles (AVs), permitting precise positioning, which is critical for appropriate trajectory planning. Light detection and ranging (LiDAR), because of its proficient depth-estimation capabilities and independence from external lighting, serves as a common tool in TLR. However, the efficacy of LiDAR-based lane detection is notably affected by both road marking’s conditions and varying meteorological conditions. A definitive guideline for improving LiDAR reliability in adverse environmental conditions remains to be established. Within the scope of this study, the feasibility of using structured road markings (RM) reflectorized with glass beads (GB) characterized by refractive index (RI) increased from the typical 1.5 to 1.65 to promote LiDAR reliability was investigated. Visibility of two dissimilar types of RM was assessed across five distinct weather conditions (dry, wet roadway surface, and rainfall intensities of 6, 25, and 100 mm/h) using two LiDARs (128-layer and 16-layer). LiDAR intensity and the number of reflected points were used as indirect indicators for assessing the equipments’ reliability. The findings revealed that the use of structured RM resulted in enhanced LiDAR intensity and reduced intensity attenuation over distance under dry conditions. Comparison of the two LiDARs showed some advantages of using 128-layer sensor, even though principal results remained comparable. This outcome shed light on the possibility of improving LiDAR reliability from the perspective of upgrading road infrastructure.


I. INTRODUCTION
D EPENDABLE traffic lane recognition (TLR) is of paramount importance for advanced driver assistance systems (ADASs) and for the emerging technology of automated vehicles (AVs), providing immediate lane information to ensure accurate positioning, enhanced lane-keeping, and well-informed decisions within intricate driving contexts [1].Amongst sensors for ADAS, light detection and ranging Manuscript received 23 November 2023; accepted 16 January 2024.Date of publication 31 January 2024; date of current version 14 March 2024.This work was supported by TU Graz Open Access Publishing Fund.The associate editor coordinating the review of this article and approving it for publication was Prof. Rosario Morello.(Corresponding author: Yongqi Zhao.) Yongqi Zhao, Hexuan Li, and Arno Eichberger are with the Institute of Automotive Engineering, Graz University of Technology, 8010 Graz, Austria (e-mail: yongqi.zhao@tugraz.at;hexuan.li@tugraz.at;arno.eichberger@tugraz.at).Tomasz E. Burghardt, Friedrich Wiesinger, and Bernhard Helmreich are with M. Swarovski GmbH, 3363 Neufurth, Austria (e-mail: tomasz.burghardt@swarco.com;friedrich.wiesinger@swarco.com;bernhard.helmreich@swarco.com).
Improving materials utilized for LiDAR input holds weakly explored potential to augment the reliability of TLR, particularly concerning trajectory planning reliant on the accurate detection of road markings (RM).To understand the impact of RM on LiDAR response, one has to realize that all RM are dual-layer systems comprising a coating layer and a drop-on glass beads (GB) layer [17].The key property for RM is retroreflectivity, which is achieved through the GB; it depends on the used drop-on GB and their condition [18].GB with higher refractive index (RI) are capable of increasing the retroreflectivity of RM, which can promote LiDAR response [18].
Limited studies from the perspective of upgrading RM to improve LiDAR-based TLR reliability were presented so far.

TABLE I EVALUATED RM SYSTEMS
In a laboratory experiment, it was shown that the use of RM reflectorized with GB having higher RI resulted in better LiDAR response [18], [19].The necessity of employing a water drainage structure for RM for reliable LiDAR-based TLR during rainy conditions was recently emphasized based on static field measurements [19].
As a weakness of the prior studies, one can list static position of the sensors at a fixed distance to the measured RM.The purpose of this work was to eliminate the effects of such unrealistic prior approaches that missed the natural variations that occur during driving.Hence, the experiment described herein was conducted during actual driving, which should provide additional information and enhance the knowledge base through the estimation of the extent of distance attenuation.The evaluated conditions included dry, wet road without precipitation, and various rainfall intensities.
To compare the response of LiDAR to RM that are known in the industry to represent the two extremes from the perspective of visibility, the test site was marked with standard RM (flat lines, reflectorized with drop-on GB with RI 1.5) and premium RM (structured RM capable of improved moisture drainage, reflectorized with GB with RI 1.65).
The major contributions of this work are as follows.
1) The results presented herein focus on the influence of weather conditions.Enhancements in LiDAR response were investigated from a novel perspective: the improvement of input materials.This approach complements the advancements in software and hardware associated with LiDAR systems.2) Experimental validation was performed under realistic driving conditions with various weather conditions, effectively bridging the gap between controlled laboratory settings and the actual driving environments.3) A side-by-side evaluation of two different LiDARs, varying in the number of layers and the power.4) Reduced LiDAR response attenuation over distance was observed through the utilization of premium RM for the first time, offering valuable insight for the development of machine vision algorithms.The outcome would be useful for ADAS software developers.In addition, it should also provide information for road administrators seeking the optimum RM to prepare the roads for the introduction of AV.

II. METHODOLOGY A. Road Markings
The specificity of RM-a unique double layer industrial maintenance coating system comprising the bottom paint layer and the top drop-on GB layer-has to be emphasized [17].While the paint furnishes color contrasting with the roadway surface, the GB provides retroreflectivity that is enhancing visibility at night on unlit roads and simultaneously plays a profound role in protecting the paint layer from abrasion [20].Retroreflectivity is the key parameter of RM which is used to determine their functional service life.Per standard EN 1436 [21], retroreflectivity is measured as coefficient of retroreflected luminance (R L ) and expressed in mcd/m 2 /lx.The same unit is used for measurements of retroreflectivity under conditions of wetness (R W ) and for daytime visibility (Q d ), expressed as luminance coefficient under diffuse illumination.
The evaluated RM and their properties are described in Table I; the main differences were the selection of drop-on GB, which provided dissimilar R L and as such should result in different LiDAR response, and the presence or absence of 3-D structures that would enhance moisture drainage (which would affect R W ). The materials were procured commercially and applied by professional industrial application crew.Both static and dynamic methods for measuring of R L were utilized [22]; this was done not only to validate the measurement test methods, but also because the dynamic method was seen as more appropriate for a driving experiment.Small differences in GB diameters were considered, based on professional experience, irrelevant for the purpose of this experiment.External appearance of the evaluated RM is depicted in Fig. 1.

B. Test Site
The evaluation was conducted at a closed course proving ground situated in St. Valentin, Austria, pictured in Fig. 2; the yellow arrow indicates the driving lane and driving direction of the test vehicle.The superstructure visible in Fig. 2 is a rain simulator, which was used in the course of this study.The assessment was done at a traffic lane 3.1 m wide delimited with three stripes (each 6.0 m long and 0.12 m wide, 12 m apart), so they imitated the marking arrangement at motorways in Austria; the stripes on the left were marked with the premium RM and on the right with the standard RM that were described in Table I.The small sample size was deemed sufficient for

TABLE II DESCRIPTION OF EXAMINED CONDITIONS
reliable results, particularly because data from numerous test drives was collected.
During the measurement campaign, the test vehicle was traveling at 30-60 km/h.Another vehicle was driving in the same lane as the test vehicle in front of it, at dynamically changed distances.This was done to imitate a common situation of slightly constricted traffic and simultaneously to provide additional reference point for LiDAR.There were total 78 drives, distributed across five test conditions described in Table II.For each test drive, data from only one frame was considered, but then the results from all test drives at particular conditions were merged.Despite being rooted in data from a single frame, the analysis maintained its realism by considering vehicle motions.

C. LiDARs
Time-of-flight LiDAR emits its own laser pulses and measures the time taken for the reflected light to return.The pulse intensity and directionality directly contribute to its capability to operate without being affected by external lighting conditions.This independence from external lighting is amongst the most important advantages of LiDAR; it guarantees accurate and reliable data acquisition at night time and even in the presence of glare, which was reported to bring the visibility by camera (and thus by human drivers) to nil [19], [23].Hence, conditions with various external illumination were not tested.
Two LiDARs, with specification as listed in Table III were utilized.The purpose of evaluating two different LiDARs was the enhancement of knowledge and clarification of some inconsistencies reported in various literature.
Output data included the relative position and intensity of LiDAR point clouds.In terms of intensity, the LiDAR manufacturer defined it as a scale to evaluate the ability of the obstacle to reflect light.Instead of reporting the actual strength of returned laser point, software of the tested equipment provided intensity values varying from 0 to 255, with 0 being the weakest intensity from a black, diffuse reflector and 255 representing the intensity of an ideal reflector.For the presentation of the results, the obtained raw intensities were not normalized, even though the scale in range 0-100 was linear and in the range 100-255 was logarithmic.

D. Data Acquisition
As presented in Fig. 3, the measurement system features a data logger with 50 Gbit/s bandwidth and object-based acquisition software.It employs an inertial measurement unit and real-time kinematic (RTK) corrected GPS for vehicle localization, achieving 1 cm accuracy.The system integrates 128-layer and 16-layer LiDARs.All data are synchronously recorded and timestamped using GPS time data.

E. Detection Algorithm and Data Processing
In the data postprocessing phase, a detection algorithm based on the prior studies was used [24], [25] were extracted based on a predefined intensity threshold, so the RM (the ROI) could be differentiated from the background (the roadway surface).It should be noted that LiDAR points of gaps between RM were not analyzed; they were excluded through the use of distance and intensity thresholds.

F. Data Presentation
The outcomes of this study was depicted in both lateral and longitudinal orientations through scatter pots and boxplots [26].For each direction, 15 figures are presented in three rows and five columns, where each column from left to right corresponds to distinct weather conditions-dry, wet, natural rain (NR), heavy rain (HR), and downpour (DP) (see Table II).In figures of each row, from top to bottom, is presented LiDAR intensity distribution in scatter plot, statistical distribution of LiDAR intensity in the boxplot, and statistical distribution of number of LiDAR points in boxplot.
To validate the observed disparities inferred from the boxplots, an unpaired two-tailed student's t-test was applied to ascertain the statistical significance of the observed distinctions within 95% confidence level; p-values are provided.

A. 128-Layer LiDAR Response
1) Lateral Direction: In Fig. 5 are depicted outcomes in the lateral direction.In case of dry conditions [Fig.5(a)], the intensity peaks matched RM line positions, with notably higher intensities measured for premium RM (left side) as compared to standard RM (right side).The differences are clearly visible in box plot shown in Fig. 5(f): median intensity for premium RM (95) was 56% higher than for standard RM (61).This was expected given the dissimilar R L of these RM (see Table I).Additionally, as shown in Fig. 5(k), more LiDAR points for premium RM were observed than for the standard RM, which was surprising.While at present no fully tenable explanation can be provided, it could be theorized that the vertical positioning of the LiDAR sensor and the existence of vertical profile in the structured RM could be playing a role.This issue requires additional investigation.These differences persisted also when the road surface (and the RM) were wet [Figs.5(b) and (g), and 6(l)], albeit the intensity decreased significantly.
Expectedly, during rainfall, the measured LiDAR intensity decreased; it was about an order of magnitude lesser than in dry conditions.No significant differences in intensity peaks between the tested RM were observed in scatter plots regardless of the rain intensities, as shown in Fig. 5(c)-(e).However, the boxplots of LiDAR intensity revealed meaningful differences, with median values for the premium RM 10%-13% higher than for the standard RM [Fig.5(h)-(j)].The number of LiDAR points detected from both types of RM were comparable in NR and DP conditions [Fig.5(m) and (o)].More LiDAR points detected from standard RM than premium RM in HR conditions [Fig.5(n)] could be attributable to false positives resulting from raindrop interference [27].
The confirmation of above observations is given in Table IV.Very importantly, compared to standard RM, premium RM consistently showed statistically significant superiority regarding enhanced LiDAR intensity across all of the test conditions, with p-values < 0.001.
In Table V is presented the response attenuation over all examined conditions.Statistically significant negative influence on visibility of RM by LiDAR was confirmed for the nth time, but this time also the effect of dissimilar RM was assessed.In such case, a more robust solution for detecting RM could be a camera.
2) Longitudinal Direction: In Fig. 6 are shown the results in longitudinal direction.In dry conditions [Fig.6(a), (f), and (k)], the superior reflective characteristic of premium RM was observed at all sampled distances; LiDAR intensities from premium RM were 53% (0-15 m), 72% (15-30 m), and 98% (30-45 m) higher than obtained from standard RM.In Table VI, LiDAR intensity comparison between premium and standard RM over distance and test conditions is presented.Premium RM consistently provided higher LiDAR intensity than standard RM across all range intervals.When compared to the range interval 0-15 m, in intervals 15-30 m and 30-45 m, the intensities from premium RM decreased by 4% and 14%, respectively (these decreases were not statistically significant).In contrast, distance attenuation measured with the standard RM was correspondingly 15% and 34% (statistically significant decrease).
When the road surface was wet but there was no precipitation [Fig.6(b)], LiDAR point intensity in all observed intervals became lower as compared to dry conditions [Fig.6(a)], which aligned with the previous finding-moisture decreased intensities of detected LiDAR points [5]; basic statistical analysis also confirmed the differences (see Table VII).As shown in Fig. 6(g), the use of premium RM lead to Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.greater LiDAR intensity than was obtained from standard RM in intervals 0-15 m and 30-45 (correspondingly, median enhancements of 62% and 69% were measured).Lower intensity in 15-30 m range remains unexplained, but the presence of residual moisture could be the reason for inconsistency; these differences were statistically significant (see Table VI).The plural of LiDAR points detected from premium and standard RM, as shown in Fig. 6(l), were comparable, which was confirmed through statistical analysis (see Table VI).Under rainy conditions, the intensity peaks corresponding to RM decreased significantly, by approximately an order of magnitude at all ranges, as can be seen in scatter plots [Fig.6(c)-(e)]; these observations were statistically depicted in Table VII.As observed in boxplots Fig. 6(h) and (i), premium RM still provided higher LiDAR intensity than standard RM in 0-15 m and 15-30 m but at some conditions, the intensity in 30-45 m range dropped to nil (see Table VI).Generally, at longer distances, either type of RM could barely generate any LiDAR response, which was due to blocking of the laser by raindrops combined with covering of the GB with moisture.
case of DP conditions [Fig.6(e)], intensities at all observation distances were higher than in NR [Fig.6(c)] and HR [Fig.6(d)]; this unusual result could be explained as being a noise caused by raindrops (particularly their size and density), but the exact reason remains unexplained.This observation was confirmed in Table VII.As depicted in boxplot Fig. 6(j), premium RM demonstrated higher intensity than standard RM only in short distance (0-15 m) in the longitudinal direction (see Table VI); the differences were statistically significant (see Table VII).

B. 16-Layer LiDAR Response
1) Lateral Direction: In Fig. 7 and Table VIII are presented responses in the lateral direction.In dry conditions [Fig.7(a) and (f)], the premium RM showed significantly higher intensity than the standard RM, with its median intensity being 10% higher (106 versus 96).This aligned with 128-layer LiDAR findings and is attributable to different R L of the RM (see Table I).The measured less pronounced disparity remains puzzling, but it is believed to be a combined effect of much lower power, lower positioning of the sensor and characteristics of the structured RM where some GB could be sheltered from the laser pulses through the vertical profile.In addition, with lower positioning of the light source (and sensor) and higher RI of the GB, the diffusion cone could have moved to furnish lesser usable reflection [28].The number of LiDAR points for both RM types was equal [Fig.7(k)].Under wet conditions [Fig.7(b) and (g)], despite intensity reduced by an order of magnitude-a contrast to 128-layer LiDAR where such magnitude reduction occurred only in rainfall-this pattern persisted.This difference was attributed to the low power of 16-layer LiDAR, which resulted   in lower intensity of received signals [23].Unexpectedly, more LiDAR points for premium RM were recorded than for the standard RM [Fig.7(l)]; a phenomenon yet to be fully explained.
During precipitation, recorded LiDAR intensities of both types of RM diminished, showing a reduction by approximately an order of magnitude as compared to dry conditions.Aligned with 128-layer LiDAR, no evident intensity disparity was observable between the premium and the standard RM in scatter plots [Fig.7(c)-(e)], but differences were confirmed in box plots [Fig.7(h)-(j)].A comparable number of LiDAR points were detected from premium and standard RM in all rainfall conditions [Fig.7(m)-(o)].These observations were confirmed through statistical analysis, with results given in Table VIII.
In Table IX, the 16-layer LiDAR attenuation across various conditions, with comparisons of intensity and number of detected points for both RM types under different conditions is presented; the results aligned with Table V.This confirms significant negative impact of moisture on RM visibility by LiDAR, regardless of the number of LiDAR layers and vertical position of the sensor.Under rainy conditions, LiDAR points were solely recorded in the range of 0-15 m, regardless of the rainfall volume.Although higher LiDAR intensities recorded with premium RM than with standard RM were not clearly visible in the plots, statistical analysis shown in Table X confirmed their existence.
In Table XI, the 16-layer LiDAR response attenuation over moisture in the longitudinal direction is presented.Analogous to 128-layer LiDAR (see Table VII), the presence of moisture reduced the response substantially.Consequently, the investigation of response attenuation over distance using the 16-layer LiDAR was not possible.
In summary, the following conclusions were obtained from the side-by-side equipment comparison experiment.
1) Premium RM consistently showed higher intensity levels in lateral observations across various meteorological conditions.This was more pronounced in the 128-layer LiDAR system as compared to the 16-layer.2) In the longitudinal direction, enhanced intensity due to premium RM was only observed in the 128-layer LiDAR system.
3) The intensity superiority of premium RM was less evident in the 16-layer LiDAR system than in the 128-layer LiDAR system, being noticeable primarily in lateral measurements.

IV. DISCUSSION
The findings of this study are aligned with expectations [19]; the utilization of premium RM, which were not only facilitating moisture drainage through a 3-D structure (Type II RM), but also through being reflectorized with GB characterized by RI 1.65 significantly promoted LiDAR response regarding intensity, even in the presence of moisture.Very importantly, the use of premium RM resulted in significantly lesser attenuation with increased distance (albeit only under dry conditions).No prior reports related to this aspect were found.This observation introduces an innovative approach toward establishing dependable TLR within moisture-laden environments-the use of better RM instead of better algorithms or sensors.This finding should be very advantageous for machine vision algorithms that demand extended look-ahead for reliable trajectory planning to ensure smooth and safe execution of ADAS decisions.Through the calculation of optimal paths derived from real-time data and predictive analytics, improved preview time for trajectory planning should significantly reduce abrupt maneuvers, mitigate collision risks, and promote passenger comfort.This enhancement is critical for fostering public trust in AV and is vital for the advancement of mobility.
The findings presented herein not only underscore the paramount importance of road infrastructure upgrades, but also establish a foundation for further investigations into LiDAR technology to handle real-world environmental complexities, thus propelling the development of safer and more reliable ADAS.
The improved recognition of RM by LiDAR was possible because of the use of GB with increased RI, which delivered higher retroreflectivity of RM under both dry and wet conditions.One has to note that the use of such GB has been documented as contributing to the longevity and sustainability of RM [29]: high-end RM, including those reflectorized with such "premium" GB were found to offer an extended functional service life as compared to standard RM.This extension was due to delivering of higher R L over an extended timeframe; consequently, better detectability for LiDAR technology would also be maintained.Higher visibility of such RM also for human drivers should additionally translate into improved road safety.
Certain constraints and ambiguities of this investigation should be acknowledged.An algorithm for detecting RM was employed, entailing the utilization of empirical threshold parameters for the extraction of LiDAR points corresponding to RM.This approach carries the inherent risk of incorporating points unrelated to RM while potentially overlooking some points that indeed pertain to them.Furthermore, during data collection, a leading vehicle was present, potentially affecting the visibility of RM for LiDAR because of water spray [30].
In this experiment, LiDARs operating at 905 nm were assessed.While the use of such frequency may be advantageous during rainy conditions [31], at other conditions, there may be benefits from utilizing more powerful LiDARs, operating at up to 1550 nm.Additionally, the experimental setup involved the mounting of 128-layer and 16-layer LiDAR systems at distinct heights on the test vehicle.Whereas, the optimal LiDAR mounting height for the detection of RM was reported as 0.5 m [32], in this study the sensors were positioned higher.Future experiments should align with this optimal height to improve data accuracy.
V. CONCLUSION Throughout this investigation, LiDAR response to two distinct types of RM with varying retrorelfectivity was analyzed across diverse weather conditions, encompassing dry, wet road (no precipitation), natural moderate intensity rain, HR, and DP.The research revealed that utilizing Type II structured RM (reflectorized with GB having RI 1.65) resulted in consistently higher LiDAR intensities as compared to Type I standard flat line RM (reflectorized with GB characterized by RI 1.5), even in moisture-laden environments.This holds the potential to enhance LiDAR reliability for TLR, thus positively influencing the trajectory planning of AV.Moreover, the elevated R L also holds promise for advancing road safety by augmenting RM visibility, benefiting both AV and human drivers.These findings bear relevance for road infrastructure planners and administrators, as the adoption of LiDAR technology continues to gain prominence in machine vision applications.The insights derived from this study offer valuable guidance in the selection and implementation of RM that effectively amplify LiDAR sensing capabilities.
. The algorithm is composed of three phases: data preprocessing, RM line position detection, and RM line LiDAR points acquisition.During the first phase [see Fig. 4(a)], LiDAR points reflected from the ground plane were obtained by defining the Region of Interest (ROI).In the second phase, the acquired LiDAR points were divided into several bins along the lateral direction of the test vehicle.Then, the confidence score of each bin was computed by an equal-weighted sum of intensities and the number of LiDAR points.Subsequently, the histogram of the confidence scores was visualized, where the confidence scores of the RM line positions appear as two peaks, representing the right and the left lines delimiting the traffic lane [see Fig. (4b)].Finally, LiDAR points detected from RM line positions [see Fig. (4c)]

2 )
Longitudinal Direction: In Fig. 8 response of the 16-layer LiDAR in the longitudinal direction are shown.In dry [Fig.8(a) and (f)] and wet [Fig.8(b) and (g)] conditions, the outstanding reflective property of premium RM was evident only in the range of 0-15 m; median LiDAR intensities of premium RM were 10% [Fig.8(k)] and 55% [Fig.8(l)] higher than obtained from standard RM.In Table X are provided the values and statistical analysis confirming the validity of the observations.In case of ranges 15-30 m and 30-45 m, no points were observable or their number was minuscule.This meaningful disadvantage of the 16-layer as compared to the 128-layer LiDAR was most likely caused by the shorted detection range (150 m versus 250 m), and lower installation height (0.74 versus 1.8 m) of the 16-layer LiDAR.

TABLE IV 128
-LAYER LiDAR RESPONSE IN LATERAL DIRECTION

TABLE V 128
-LAYER LiDAR RESPONSE ATTENUATION DUE TO MOISTURE IN LATERAL DIRECTION, STATISTICAL ANALYSIS

TABLE VII 128
-LAYER LiDAR RESPONSE ATTENUATION DUE TO MOISTURE IN LONGITUDINAL DIRECTION, STATISTICAL ANALYSIS

TABLE VIII 16
-LAYER LiDAR RESPONSE IN LATERAL DIRECTION

TABLE IX 16
-LAYER LiDAR RESPONSE ATTENUATION DUE TO MOISTURE IN LATERAL DIRECTION, STATISTICAL ANALYSIS