Real-Time Lane Departure Detection Using Google Routes

Nafisa Zarrin Tasnim, Attiq Uz Zaman and M. I. Hayee

Department of Electrical Engineering, University of Minnesota Duluth, Duluth, MN 55812, U.S.A.

Keywords: Lane Departure Detection, Google Routes, Road Reference Heading, GPS Trajectory, Lane Departure

Warning System (LDWS).

Abstract: Our previously developed Lane Departure Detection and Warning System (LDWS) used a standard GPS

receiver and two algorithms to detect an unintentional lane departure. The first algorithm generated the Road

Reference Heading (RRH) from a vehicle’s past trajectories, while the second algorithm predicted lane

departure in real time using RRH. A significant limitation of this system is the dependency on past trajectories.

A vehicle must travel on the road at least once in the past to use that trajectory for RRH generation needed

for future lane departure detection. To avoid dependency on past trajectories, this work uses Google routes

instead of past trajectories to extract the RRH of any given road. We also compared the RRH generated from

a Google route with that of a past trajectory and found both RRHs to be comparable indicating that our LDWS

does not need to rely on RRH from past trajectories. To evaluate the accuracy of lane departure detection

using Google RRH, we performed many field tests on a freeway. Our field test results show that our LDWS

can accurately detect all lane departures on long straight sections of the freeway irrespective of whether the

RRH was generated from a Google route or past trajectory.

1 INTRODUCTION

According to the World Health Organization (WHO)

report, roughly 1.3 million people die yearly from

road traffic crashes (Road Traffic Injuries, n.d.). A

survey that examined the national sample of US

crashes from 2005 to 2007 identified driver error as

the critical reason contributing to 94 percent of

crashes (Singh, 2015). Advanced driver assistance

systems (ADAS) (Antony & Whenish, 2021)

facilitate drivers to make well-informed decisions and

consequently help them avoid collisions. The recent

advancements in ADAS technologies have increased

safety for both drivers and pedestrians. The analysis

of a study that estimated the safety benefits of in-

vehicle lane departure warning (LDW) and lane-

keeping aid (LKA) systems in reducing relevant car

injury crashes demonstrated that such systems

reduced head-on and single-vehicle injury crashes by

53% at a 95% confidence interval on Swedish roads

within a specific speed limit range and road surface

conditions (Sternlund et al., 2017). Most Lane

Departure Warning Systems (LDWSs) depend on

image processing and use cameras, infrared, or laser

sensors to estimate a vehicle’s lateral shift within a

lane to detect an unintentional lane departure (An et

al., 2006, Baili et al., 2017, Chen & Boukerche, 2020,

Hsiao and Yeh 2006, Jung et al., 2013, Leng & Chen,

2010, Lindner et al., 2009, Yu et al., 2008). Vision-

based LDWSs face challenges adapting to diverse

weather conditions, handling light changes, and

mitigating shadow effects. Blockage of lane markings

by other vehicles in LDWS images can also cause

detection failures. Real-time processing of captured

images is crucial for vision-based LDWS,

necessitating synchronous image processing speed to

ensure timely and safe detection (Chen et al., 2020).

Gamal et al. (2019) also propose a real-time,

calibration-free lane departure warning system

algorithm utilizing Gaussian pyramid preprocessing

and Edge Drawing Lines algorithm, achieving high

accuracy (99.36%) and fast processing (80 fps),

suitable for integration into self-driving systems in

OEM cars, outperforming existing algorithms.

Although advanced image processing techniques

work well in diminished lighting scenarios (Dobler et

al., 2000, McCall & Trivedi, 2006), many of today’s

commercially available image processing-based

LDWSs have performance issues under unfavorable

weather or road conditions like fog, snow, or worn out

road markings potentially leading to inaccurate lane

detection or overlooking genuine lane markings. To

238

Tasnim, N., Zaman, A. and Hayee, M.

Real-Time Lane Departure Detection Using Google Routes.

DOI: 10.5220/0012557000003702

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 10th International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS 2024), pages 238-248

ISBN: 978-989-758-703-0; ISSN: 2184-495X

tackle the limitations posed by adverse weather

conditions and degraded road markings on the

performance of image processing-based LDWSs,

recent advancements have emerged. For instance,

Son et al. (2015) have introduced a lane departure

warning system specifically designed to combat

illumination challenges prevalent in driving

environments. This system demonstrates a 93%

detection rate, surpassing traditional approaches,

even in scenarios with blurred lane markings or low

sun angles. Furthermore, Sang et al. (2024) have

developed a self-tuned algorithm that integrates fuzzy

logic-based adaptive functions with edge

identification and line detection modules. This

algorithm aims to enhance image quality in

challenging weather conditions, although it may have

limitations in handling multiple and curved lines. To

address some of these performance issues, Global

Positioning System (GPS) technology is integrated

within vision-based LDWS. To estimate a vehicle's

lateral movement in its lane, such systems employ

differential GPS (DGPS) technology (Bajikar et al.,

1997) and/or inertial navigation and/or odometry

sensors (Toledo-Moreo & Zamora-Izquierdo, 2009)

in addition to high-resolution digital maps, making

them challenging to deploy and expensive (Clanton et

al., 2009). Weon et al. (2021) propose a lane

departure detection algorithm for vehicles using

Geographic Information System (GIS) and DGPS

data with high positioning accuracy under 20 cm. It

calculates lane segment distances based on vehicle

position, utilizing Bezier curves for curved sections.

Although GPS is a very helpful tool for navigation,

there are situations in which signal interference,

multipath effects, or complex road networks may

make it less reliable in streets and urban areas. Rose

et al. (2014) demonstrate how combining vision,

height measurements, and a lane map can

significantly reduce drift when a GPS signal is lost,

such as in dense foliage or urban environments. By

utilizing existing sensors in commercial vehicles, the

system achieves submeter accuracy in lateral distance

measurement.

Our previously developed LDWS system was

based on a standard GPS receiver without using any

image processing or optical scanning devices, or

high-resolution digital maps. To detect a lane

departure, our developed LDWS system estimated

lateral vehicle shift using standard GPS technology.

The lateral vehicle shift was estimated by comparing

the vehicle trajectory acquired by a standard GPS

receiver with a Road Reference Heading (RRH),

which was obtained using a newly developed

algorithm (Shahnewaz Chowdhury & Hayee, 2021).

Our Lane Departure Detection (LDD) algorithm

accumulated instantaneous vehicle lateral shifts to

detect an unintentional lane departure in real-time

(Faizan et al., 2019).

One major limitation of the previously developed

LDWS system is the dependency on past trajectories.

To detect an unintentional lane departure of any

vehicle in real-time on a given road, the vehicle must

have traveled on the same road at least once in the

past to use that trajectory for RRH generation. During

the subsequent trips on the same route, the system can

detect a potential lane departure using already

generated RRH to warn the driver. To avoid

dependency on past trajectories, this work utilizes

Google routes from Google Maps instead of past

trajectories to extract the RRH of any given road.

RRH can be extracted from various GPS navigation

systems offered by companies such as Waze, Garmin,

TomTom, Sygic, and Apple, among others. However,

we opted for Google Maps as it is a widely recognized

and popular option. Google routes are available for all

roads within the US through Google Maps as

navigational routes.We have used the navigational

route on any given road provided by Google Maps as

a Google route to extract the RRH of that road using

our RRH generation algorithm. The extracted RRH

from Google routes was used to detect an

unintentional lane departure in real-time and alert the

driver with an audible warning. We also compared the

lane departure detection results using Google RRH to

that of the past trajectories and found that lane

departure detection resulting from Google RRH and

RRH from the past trajectories are comparable. We

also performed field tests to evaluate the results

showing that our system can accurately detect lane

departure on long straight sections of the freeways.

This shows the potential of our LDWS to be used for

all US roads without the dependency on past

trajectories.

The rest of the paper is organized as follows.

Section 2 describes the architecture of the system,

followed by section 3, which describes RRH

generation from Google routes. Section 4 discusses

the field tests and results, followed by conclusions in

Section 5.

2 SYSTEM ARCHITECTURE

The previously developed LDWS system has been

enhanced to work for both past vehicle trajectories

and Google routes as shown in Figure 1a. The

enhanced LDWS can either generate RRH using past

trajectories or from Google routes obtained from

Real-Time Lane Departure Detection Using Google Routes

239

Google Maps. In the updated LDWS, we have

incorporated necessary modifications and

enhancements of the RRH generation algorithm to

accommodate RRH generation from Google routes.

With this additional feature of extracting RRH, our

LDWS is capable of detecting lane departures of a

vehicle in real-time on any road, regardless of

whether the vehicle has traveled on that road in the

past. Our LDWS compares the vehicle trajectory in

real-time with the RRH of that road to detect a lane

departure using our LDD algorithm as depicted in

Figure 1b where a vehicle trajectory with a lane

departure (red dotted line) can be compared with the

RRH (blue dotted line) obtained from the Google

route (blue solid line) or from one of the past

trajectories (not shown in Figure 1b) to determine

lane departure using our LDD algorithm (Faizan et

al., 2019).

We have also made enhancements to our LDD

algorithm to detect lane departure more efficiently.

However, this paper is more focused on the details of

the new feature of extracting RRH from Google

routes obtained from Google Maps. To evaluate the

effectiveness of lane departure detection using RRH

from Google routes, we have compared the results of

lane departure detection using RRH from both

Google routes and past trajectories, which will be

described in more detail in the field tests and results

section.

3 RRH GENERATION FROM

GOOGLE ROUTE

We have developed a backend browser-based

application where we can specify the start and end

points of the desired route on a Google map to obtain

the RRH from the Google route on almost any road

within the US. The user specifies the start and end

location points in the Google map to define the

Google route necessary for RRH generation. Our

backend browser application accesses Google Maps’

API for turn-by-turn directions between the start and

Figure 1: (a) Architecture of the LDWS system. (b) Conceptual diagram demonstrating that RRH (blue dotted line) can be

generated from a Google route (blue solid line, where points A and B are the start and end points, respectively) to detect an

intentional lane change from right to left (red dotted line). The road illustrated in this figure is a 4-lane road with the Google

route shown in the middle.

(1b)

(1a)

VEHITS 2024 - 10th International Conference on Vehicle Technology and Intelligent Transport Systems

240

end points of the route and returns an array of

locations with latitudes and longitudes. In this array

of locations provided by Google Maps, there are more

location points on a curve segment of the road as

compared to a straight segment of the road. This array

of latitudes and longitudes represents the Google

route from which RRH is generated.

We preprocess the array of location points to

obtain a uniform distance resolution before applying

our RRH generation algorithm. To obtain a uniform

distance resolution, we have to access Google Maps

API a few times between intermediate location points

in the array until we get the desired resolution so that

every two consecutive points in the Google Maps

array of locations have the same distance as the

desired resolution typically in 1 to 3 m range. We

have to do more iterations on straight sections in the

array as compared to a curve section because there are

fewer number of points on a straight section as

compared to a curve section.

Our developed RRH generation algorithm

characterizes a typical road segment into straight and

curve sections, as any typical road can have a

combination of straight and curve portions. However,

the roads do not curve abruptly; therefore, the section

between a straight and a curve is characterized as a

transition section by the algorithm. Our algorithm

generates an RRH from the pre-processed Google

route (uniform array of locations) for any given

portion of the road in three major steps.

1) Identification: In the first step, all straight,

curve and transition sections are identified on that

road.

2) Characterization: After identification, each

section is characterized with a set of optimized

parameters that determines the RRH value at any

given point on that road. Every straight section is

characterized by a Path Average Heading (PAH), as

the heading remains constant for the entire length of

a straight section. Since the heading of a curve section

varies uniformly with distance, it is characterized by

a Path Average Heading Slope (PAHS) and an initial

heading (IH), where IH is the heading at the

Figure 2: (a) Browser application user interface showing the user-specified route on the Google map to initiate RRH

generation from the Google route. Points A and B indicate the start and end location of the desired route, respectively. (b)

Google route superimposed on the generated RRH with straight, curve and transition sections indicated by red, blue, and

green colors along the route.

(2a)

(2b)

Real-Time Lane Departure Detection Using Google Routes

241

beginning point of the curve section. Each transition

section is characterized the same way as a curve

section.

3) Aggregation: Finally, all the individual

road sections are combined to obtain a composite

RRH for that road to be used with the LDD algorithm

for lane departure detection.

The process of generating RRH from the Google

route using the browser application is illustrated in

Figure 2. Our browser application lets the user specify

the start and end points on a given road on Google

Maps, as shown in Figure 2a, where points A and B

are the start and end points, respectively. After the

user specifies the start and end points, our browser

application runs our RRH generation algorithm and

converts the Google route to RRH consisting of all

sections of the road, as shown in Figure 2b, where the

red, blue, and green colors along the Google route

indicate straight, curve and transition sections,

respectively. Our browser application also generates

a corresponding datafile containing all the relevant

information of all the sections of RRH within the

Google route. The information in this datafile

contains the start and end points of each section

(straight, curve and transition) and the relevant

parameter values characterizing each section to

determine the RRH value for any point on the road.

Using this information, an RRH value can be

calculated on any given point of the road between the

start and end points specified by the user.

3.1 Algorithm and Parameter

Optimization

The implementation steps of our RRH generation

algorithm are as follows:

1) First, the Google route is obtained from the

browser application containing the array of

points in terms of longitude and latitude for

each point. The Google route is then pre-

processed for uniform distance resolution so

that each of the two consecutive points has a

constant distance between them (1-3 m

range), as explained earlier.

2) Then, the heading between two consecutive

points is calculated from the longitude and

latitude of those points (Shahnewaz

Chowdhury & Hayee, 2021). The heading

array is then filtered with a lowpass filter

having a corner frequency of 20 Hz

(cycles/m) and a total stop frequency of 125

Hz (cycles/m). The filtering is done to

smoothen the heading array obtained from

Google Maps. The choice of corner and stop

frequencies is made to ensure the removal of

undesired ripple effects from the heading

array.

3) Next, differential headings per meter are

calculated from the filtered heading array.

The differential heading is then smoothened

to further minimize undesired ripples. The

smoothening is obtained by averaging

differential heading at each point over 40

meters, i.e., ± 20m on either side of the given

point.

4) The average differential heading array is

then used to identify the straight sections

using a threshold of 0.002 deg/m, i.e., any

consecutive portion with a differential

heading of less than the threshold of 0.002

deg/m will be considered a straight section

(Shahnewaz Chowdhury & Hayee, 2021).

This step will generate the start and end

points of each of the straight sections present

in the Google route. Please note that we

combined two consecutive straight sections

that are not too far apart from each other

because a road cannot curve abruptly. For

that purpose, we are using a range of 75-100

m as a parameter to combine two

consecutive straight sections. Straight

sections are then characterized by

calculating a PAH between the start and end

points of each straight section.

5) After identifying and characterizing straight

sections, curve sections are identified and

characterized. For this purpose, any portions

of the Google route between two

consecutive straight sections are considered

curve sections, i.e., any contiguous portions

on the Google route having a differential

heading above the average differential

heading threshold of 0.002 deg/m. A PAHS

value for each of the curve sections between

two consecutive straight sections is then

calculated from the average differential

heading. The beginning and end points of

each of the curve sections are identified

where the average differential heading is

closest to the calculated PAHS on that curve

section. The heading at the beginning point

of each curve section becomes the IH for that

curve section.

6) The above method of identifying the start

and end points of the curve section tends to

make the transition sections longer for those

curves which have a higher value of PAHS.

Therefore, for such curves which have a

VEHITS 2024 - 10th International Conference on Vehicle Technology and Intelligent Transport Systems

242

PAHS value of >0.02 deg/m, we apply

selective thresholding to reduce the lengths

of the transition sections and increase the

lengths of the adjacent straight sections

while keeping the curve section length the

same. To achieve this, we increase the

threshold of 0.002 deg/m (as in step 4) to

0.01 deg/m (5 times the original threshold)

only for those curves which have a PAHS

value of >0.02 deg/m.

7) Similarly, some of the curves with a small

PAHS may be falsely accounted for as curve

sections because their PAHS value is only

slightly higher than the threshold (0.002

deg/m) to be identified as curve sections to

begin with. To identify and eliminate such

false curve sections, the PAHS value for

each of the curve sections is recalculated by

dividing the difference of the PAH of the

two surrounding straight sections by the

distance between them. If such a revised

PAHS turns out to be smaller than 0.002

deg/m (original threshold), then that curve is

absorbed in the surrounding straight

sections.

8) The characterized parameters for each

straight and curve section are then optimized

to minimize the differential heading error

(Shahnewaz Chowdhury & Hayee, 2021)

between the RRH and Google route heading

over the entire portion of each of the straight

and curve sections.

9) In the end, transition sections are

identified and characterized. Any portion of

the Google route between a curve and a

straight section is marked as a transition

section. The heading at the start point of the

transition section becomes the IH of the

transition section. The PAHS of the

transition section is calculated from the

difference in headings between the start and

end points of the transition section divided

by the distance between them.

To see the effectiveness of the RRH generated

from the Google route, we have plotted the Google

route heading (green solid line) and the RRH (black

solid line) with respect to distance as shown in Figure

3. To identify straight, curve and transition sections

on the RRH, a mask with a black dotted line is also

shown in Figure 3. The mask has fixed but different

heading values for straight, curve and transition

sections to distinguish them from each other. The

straight sections have a fixed mask value of 252.5

(almost in the middle of Figure 3). For each of the

curve and transition sections, two different mask

values are attributed to each curve or transition

section depending on the PAHS. The mask value for

the curve section is either 275

or 230

, depending on

whether the PAHS is positive or negative,

respectively. Similarly, the mask value for the

transition section is 263.75

or 241.25

, depending on

whether the PAHS is positive or negative. Please note

that specific mask values are chosen to mark the

difference between the straight, curve and transition

sections within the RRH for the Google route in

Figure 3.

The RRH values (black solid line) match pretty

well with the Google route heading (green),

especially for straight sections (Figure 3). The match

between the RRH and the Google route varies for

curve sections and we noticed that for one of the

sharper curve sections between 1500 m and 2200 m,

the match is not as good as for the other portions

(Figure 3). This can be improved by tweaking the

conditions of the RRH generation algorithm in the

Figure 3: Google route heading (green) and corresponding RRH (solid black) vs. distance. A mask (black dotted line) is drawn

to indicate the straight, curve and transition sections.

Real-Time Lane Departure Detection Using Google Routes

243

Figure 4: A past trajectory heading (green) and corresponding RRH (solid black) vs. distance. A mask (black dotted line) is

drawn to indicate the straight, curve and transition sections.

future. However, our main goal is that the RRH

values for the straight sections are reliable so that

unintentional lane departure on long stretches of the

straight sections can be successfully detected.

We also wanted to compare the accuracy of the

Google RRH with the RRH obtained from the past

trajectories. Therefore, we applied the modified RRH

generation algorithm to one of the past vehicle

trajectories obtained on the Interstate I-35

southbound which is the same portion of I-35 for

which the Google route was used to extract RRH. The

trajectory length was a little shorter than 12 km

covering almost the same portion as the Google route

in Figure 3. The trajectory heading vs. distance is

plotted in Figure 4. For reference, an RRH obtained

from this trajectory is also shown in Figure 4. The

same mask as in Figure 3 is also used in Figure 4 to

differentiate among straight, curve and transition

sections of the RRH. Figure 4 shows that the

generated RRH (black solid line) follows the vehicle

trajectory heading (green) fairly well, indicating that

the modified RRH generation algorithm also works

well for the past trajectories. In fact, for a much

sharper curve present in the trajectory (between 1500

m and 2200 m range), the match between RRH and

the past trajectory is much better (Figure 4), as

compared to the similar match for the Google route

(Figure 3). These discrepancies in Google RRH exist

because sometimes the road width changes, i.e., the

number of lanes cause abrupt heading deviations in

the Google route. However, a more rigorous

comparison of the RRH obtained from the Google

route and that of the past vehicle trajectory will be

made later by detecting lane departures using our

LDD algorithm.

4 FIELD TEST AND RESULTS

We performed many field tests to evaluate the RRH

obtained from the Google route using our LDD

algorithm. All the field tests were performed by

driving a test vehicle multiple times on the same 12

km segment of Interstate I-35 southbound, for which

an RRH was extracted from the Google route as

discussed earlier. Each of the test runs covered a

portion of the 12 km road segment to ensure that the

test vehicle remained on the same portion of the road

for which an RRH was extracted earlier from the

Google route. The test vehicle was driven at about the

speed limit (70 MPH) on the 4-lane freeway (2 lanes

each way) and many lane changes were made

intentionally during the field tests. For safety reasons,

intentional lane changes were made to test the

accuracy of lane departure detection. The vehicle

trajectory data for each of the test runs were collected

and evaluated with our LDD algorithm to detect the

start and end of the lane changes present in each

trajectory. Four such trajectories (red, blue, purple &

orange colors) from the test runs are shown in Figure

5a, where vehicle heading vs. vehicle traveled

distance is plotted. For reference, Google RRH (black

solid line) is also shown on the same scale in Figure

5a. In the first two trajectories (red and blue) shown

in Figure 5a, a total of 4 lane changes were made, and

in the third trajectory (purple), only 2 lane changes

were made. There was no lane change made for the

fourth trajectory (orange) to ensure that no false alarm

was detected by our LDD algorithm. All lane

changes were made only on the straight portions of

the road as unintentional lane departure is more

relevant on longer stretches of straight portions of the

road as discussed earlier.

VEHITS 2024 - 10th International Conference on Vehicle Technology and Intelligent Transport Systems

244

Figure 5: (5a) RRH resulting from the Google route (black) along with the headings of four different trajectories (red, blue,

purple & orange) on the same route vs. distance. A zoomed in portion of the Google RRH and 4 trajectories between the 3000

m to 3500 m distance range is shown as an in-set in the figure. (5b) Vehicle accumulative lateral distance vs. traveled distance

for four different driven trajectories (red, blue, purple & orange). The vehicle accumulative lateral distance is calculated using

Google RRH.

However, from the trajectories shown in Figure

5a, lane changes cannot be visibly identified because

there is a very small difference in the heading of a

vehicle trajectory and the Google RRH. An example

of a lane change is highlighted in a zoomed portion of

Figure 5a between the 3000 m and 3500 m range to

show that there is a lane change in two of the four

trajectories (red and purple) in opposite directions

indicated by the vehicle heading deviation in the

opposite directions as compared to the Google RRH.

Similarly, for the other two trajectories (blue and

orange), there was no lane change for this portion, so

no noticeable deviation was seen in vehicle heading

as compared to the Google RRH.

To evaluate the accuracy of the lane change

timing and duration, the accumulative lateral distance

of the vehicle for each trajectory vs. traveled distance

is shown in Figure 5b. The colors of the accumulated

lateral distance for each trajectory are kept the same

as in Figure 5a. When the accumulative lateral

distance exceeds a certain threshold (~1m) value on

either side, it is considered a lane departure. All lane

Zoomed-in

portion

(5a)

(5b)

Real-Time Lane Departure Detection Using Google Routes

245

Figure 6: (6a) RRH resulting from one of the past trajectories (black) along with the headings of three different trajectories

(red, blue & purple) vs. distance. (6b) Vehicle accumulative lateral distance vs. traveled distance for three different driven

trajectories (red, blue & purple). The vehicle accumulative lateral distance is calculated using the RRH extracted from a past

trajectory.

changes were accurately identified by our algorithm

indicating that the RRH obtained from Google route

works well for detecting lane departures, either

intentional (lane change) or unintentional. The start

and stop of each of the lane changes are clearly

identifiable in Figure 5b. For the orange trajectory,

where there is no lane change, the accumulative

lateral distance never exceeds the threshold of 1 m.

We also wanted to compare the accuracy of lane

changes being identified correctly when the RRH was

generated from a Google route as well as a past

trajectory. We have used Trajectory 4 (orange) of

Figure 5 to generate an RRH from it because it does

not have any lane changes. The remaining three

trajectories (red, blue & purple) from the test runs

were used to detect lane changes in them using the

RRH extracted from one of the past trajectories

(Trajectory 4). The heading of the three trajectories

(red, blue and purple) are shown again in Figure 6a,

where vehicle heading is plotted with respect to

traveled distance. As stated earlier, in two of these

three trajectories (red and blue), a total of 4 lane

changes were made in each, and in the third trajectory

(purple), only 2 lane changes were made. For

(6a)

)

VEHITS 2024 - 10th International Conference on Vehicle Technology and Intelligent Transport Systems

246

reference, RRH extracted from Trajectory 4 has also

been shown in Figure 6a.

To evaluate the accuracy of the lane change

timing and duration, the accumulative lateral distance

of the vehicle for each of the three trajectories using

the RRH from Trajectory 4 is calculated and is shown

in Figure 6b vs. traveled distance. The colors of the

accumulative lateral distance for each trajectory are

kept the same as in Figure 6a. All lane changes were

accurately identified by our algorithm showing the

accuracy of the RRH obtained from a past trajectory

using the modified RRH generation algorithm. The

start and stop of each of the lane changes are clearly

identifiable in Figure 6b. These results indicate that

the lane departure can be accurately detected

irrespective of whether the RRH is generated from a

Google route or a past trajectory for straight portions

of the road.

5 CONCLUSIONS

We have successfully developed and implemented

the algorithm to extract RRH from a Google route to

work with our previously developed LDWS to detect

unintentional lane departures. We have evaluated the

effectiveness of the RRH from the Google route by

performing field tests and comparing the results with

that of the RRH from a past trajectory. Our results

indicate that our LDWS can accurately detect a lane

departure irrespective of whether the RRH is

generated from a Google route or a past trajectory for

straight portions of the road. However, to ensure

accurate lane departure detection on curved road

sections, further refinement of the RRH generation

algorithm is necessary to align the RRH with the

trajectory. Although results have been reported from

fourtrips of many along the same 12 km segment of

the Interstate I-35 southbound route, it is worth

mentioning that this is an ongoing work, and we are

in the process of validating this approach with more

data from different routes.

ACKNOWLEDGEMENTS

The authors wish to acknowledge those who made

this research possible. This work was made possible

by Minnesota cities and counties by the Local Road

Research Board with support from MnDOT’s Office

of Research & Innovation.

REFERENCES

An, X., Wu, M., & He, H. (2006). A novel approach to

provide lane departure warning using only one forward-

looking camera. International Symposium on

Collaborative Technologies and Systems (CTS’06),

356–362.

Antony, M. M., & Whenish, R. (2021). Advanced driver

assistance systems (ADAS). In Automotive Embedded

Systems: Key Technologies, Innovations, and

Applications (pp. 165–181). Springer

Baili, J., Marzougui, M., Sboui, A., Lahouar, S., Hergli, M.,

Bose, J. S. C., & Besbes, K. (2017). Lane departure

detection using image processing techniques. 2017 2nd

International Conference on Anti-Cyber Crimes

(ICACC), 238–241.

Bajikar, S., Gorjestani, A., Simpkins, P., & Donath, M.

(1997). Evaluation of in-vehicle GPS-based lane

position sensing for preventing road departure.

Proceedings of Conference on Intelligent

Transportation Systems, 397–402.

Chen, W., Wang, W., Wang, K., Li, Z., Li, H., & Liu, S.

(2020). Lane departure warning systems and lane line

detection methods based on image processing and

semantic segmentation: A review. Journal of Traffic

and Transportation Engineering (English Edition),

7(6), 748–774.

Chen, Y., & Boukerche, A. (2020). A novel lane departure

warning system for improving road safety. ICC 2020-

2020 IEEE International Conference on

Communications (ICC), 1–6.

Clanton, J. M., Bevly, D. M., & Hodel, A. S. (2009). A low-

cost solution for an integrated multisensor lane

departure warning system. IEEE Transactions on

Intelligent Transportation Systems, 10(1), 47–59.

Dobler, G., Rothe, S., Betzitza, P., & Hartlieb, M. (2000).

Vehicle with optical scanning device for a lateral road

area. Google Patents.

Faizan, M., Hussain, S., & Hayee, M. I. (2019). Design and

development of in-vehicle lane departure warning

system using standard global positioning system

receiver. Transportation Research Record, 2673(8),

648–656.

Gamal, I., Badawy, A., Al-Habal, A. M., Adawy, M. E.,

Khalil, K. K., El-Moursy, M. A., & Khattab, A. (2019).

A robust, real-time and calibration-free lane departure

warning system. Microprocessors and Microsystems,

71, 102874.

Hsiao, P.-Y., & Yeh, C.-W. (2006). A portable real-time

lane departure warning system based on embedded

calculating technique. 2006 IEEE 63rd Vehicular

Technology Conference, 6, 2982–2986.

Jung, H., Min, J., & Kim, J. (2013). An efficient lane

detection algorithm for lane departure detection. 2013

IEEE Intelligent Vehicles Symposium (IV), 976–981.

Leng, Y.-C., & Chen, C.-L. (2010). Vision-based lane

departure detection system in urban traffic scenes. 2010

11th International Conference on Control Automation

Robotics & Vision, 1875–1880.

Real-Time Lane Departure Detection Using Google Routes

247

Lindner, P., Richter, E., Wanielik, G., Takagi, K., & Isogai,

A. (2009). Multi-channel lidar processing for lane

detection and estimation. 2009 12th International IEEE

Conference on Intelligent Transportation Systems, 1–6.

McCall, J. C., & Trivedi, M. M. (2006). Video-based lane

estimation and tracking for driver assistance: Survey,

system, and evaluation. IEEE Transactions on

Intelligent Transportation Systems, 7(1), 20–37.

Road traffic injuries. (n.d.). Retrieved December 11, 2023,

from https://www.who.int/news-room/fact-sheets/deta

il/road-traffic-injuries

Rose, C., Britt, J., Allen, J., & Bevly, D. (2014). An

integrated vehicle navigation system utilizing lane-

detection and lateral position estimation systems in

difficult environments for GPS. IEEE Transactions on

Intelligent Transportation Systems, 15(6), 2615–2629.

Sang, I. C., & Norris, W. R. (2024). A Robust Lane

Detection Algorithm Adaptable to Challenging

Weather Conditions. IEEE Access.

Shahnewaz Chowdhury, M. T. H., & Hayee, M. I. (2021).

Generation of Road Reference Heading using GPS

Trajectories for Accurate Lane Departure Detection.

Singh, S. (2015). Critical reasons for crashes investigated

in the national motor vehicle crash causation survey.

Son, J., Yoo, H., Kim, S., & Sohn, K. (2015). Real-time

illumination invariant lane detection for lane departure

warning system. Expert Systems with Applications,

42(4), 1816-1824.

Sternlund, S., Strandroth, J., Rizzi, M., Lie, A., & Tingvall,

C. (2017). The effectiveness of lane departure warning

systems—A reduction in real-world passenger car

injury crashes. Traffic Injury Prevention, 18(2), 225–

229.

Toledo-Moreo, R., & Zamora-Izquierdo, M. A. (2009).

IMM-based lane-change prediction in highways with

low-cost GPS/INS. IEEE Transactions on Intelligent

Transportation Systems, 10(1), 180–185.

Weon, I. S., Lee. S.G. & Woo, S. H. (2021). Lane Departure

Detecting with Classification of Roadway Based on

Bezier Curve Fitting Using DGPS/GIS. Tehnički

vjesnik, 28(1), 248-255.

Yu, B., Zhang, W., & Cai, Y. (2008). A lane departure

warning system based on machine vision. 2008 IEEE

Pacific-Asia Workshop on Computational Intelligence

and Industrial Application, 1, 197–201.

VEHITS 2024 - 10th International Conference on Vehicle Technology and Intelligent Transport Systems

248