Intersection with Highly Adaptive Traffic Lights

Can Still Be Suitable for C-ITS Service GLOSA

Thomas Otto

, Ina Partzsch and Michael Klöppel-Gersdorf

Fraunhofer Institute for Transportation and Infrastructure Systems IVI, Zeunerstraße 38, Dresden, Germany

Keywords: GLOSA, C-ITS, CCAM, Stability Indicators.

Abstract: The C-ITS service “Green Light Optimal Speed Advisory” (GLOSA) is a highly promising Day1.0 services

for efficient, environmentally friendly and safe cooperative transport. It will play a crucial role in the

increasing automation of assistance and driving functions in cooperative connected automated mobility

(CCAM). Over the past three decades, significant efforts have been made to make traffic light signalling as

adaptable as possible to traffic requests. However, there is a common belief that the more flexible the control,

the less reliable the forecast and thus the functionality of GLOSA. This paper introduces stability indicators

to demonstrate that this belief is only partially accurate. The proposed tool allows for the analysis of historical

data from existing systems to derive an indicator for the quality and suitability of the C-ITS GLOSA

application. We demonstrate the feasibility of the approach using real world data from the C-ITS corridor

Hamburg, Germany.

1 OBJECTIVES AN MOTIVATION

Safety and environmental compatibility within the

mobility sector are the main objectives of the coming

years and an essential building block for ensuring an

efficient mobility transition. With increasing

automation and cooperation between the

infrastructure and the vehicles via so-called Car2X

communication, the prerequisites for this are being

created. The need for C-ITS services will increase as

they can be directly integrated into driving functions

of automated and autonomous vehicles. All services

of cooperative intelligent transport systems (C-ITS)

serve to improve traffic safety, increase the efficiency

of the traffic network or ensure environmentally

friendly transport. On the one hand, these C-ITS

applications can be warning services (e.g., traffic jam

warning, road works warning, accident warning) but

also information services (e.g., traffic light phase

assistant, route recommendations, speed limits).

Traffic lights are and remain the bottlenecks of

urban traffic. Increasing automation and

digitalization will not change that. Where large traffic

volumes collide, the only option, also in the future,

will be to separate them temporally. This need will

https://orcid.org/0000-0003-0099-9363

https://orcid.org/0000-0001-9382-3062

remain as long as not all road users are connected and

automated. Since we always take pedestrians and

cyclists into account in urban areas - which also

contributes to the climate aims - there will be not

more intersections without traffic lights in the

foreseeable future. This makes it even more necessary

to efficiently utilize the previously unused potential

of signalized intersections. To this aim, cooperative

systems offer a range of services that can be used

accordingly. One of the most frequently discussed

services at the moment is the C-ITS service GLOSA.

This is often simply referred to as the traffic light

phase assistant. In addition to transmitting the signal

status information, the GLOSA service also transmits

information about the remaining red and green times.

The forecast algorithm and its accuracy have already

been examined in several research papers, for

example, as statistical prediction model for traffic

lights via Kalman filtering (Protschky et al., 2014),

via support vector machines (Weisheit et al., 2014),

via machine learning algorithm (Scheegans et al.,

2022) or via Markov chain (Otto et al., 2013;

Barthauer et al., 20214). Confidence intervals

regarding switching time difference in consideration

of the lead time to the signal change were first

148

Otto, T., Partzsch, I. and Klöppel-Gersdorf, M.

Intersection with Highly Adaptive Trafﬁc Lights Can Still Be Suitable for C-ITS Service GLOSA.

DOI: 10.5220/0013354500003941

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

In Proceedings of the 11th International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS 2025), pages 148-156

ISBN: 978-989-758-745-0; ISSN: 2184-495X

discussed in (Bodenheimer et al., 2015). The GLOSA

service also transmit speed recommendations for the

optimal approach to the intersection via the

SPATEM/MAPEM C-ITS message.

Coordinated roads and thus the use of GLOSA is

subject to certain boundary conditions, which are

described in (Otto et. al., 2010; Genser, 2022). This is

the problem that is currently often discussed in

research papers. The hypothesis here is often that the

more flexible the control, the less reliable the forecast

and thus the functionality of GLOSA. The efforts of

the last 30 years to implement the most flexible

control possible, which adaptively takes into account

the needs of all road users, seem to be in extreme

contradiction to the requirement for a stable forecast

for the GLOSA service. Due to certain boundary

conditions in the transport network, there are

conditions that are ideal for the quality of the forecast.

These are, for example, coordination. This is exactly

where this hypothesis can be refuted. Research on this

was already carried out (Eckhoff et al., 2013; Eteifa

et al., 2021; Krumnow, 2023; Mellegård et al., 2020;

Suzuki et al., 2020).

Increasing digitalization is clearly accompanied

by increasing availability of data. This not only

happens spatially, but the requirements for temporal

access - such as real-time capability – also underwent

a complete change. Current requirements are not

comparable to the requirements on conventional

traffic management systems and applications. Until

recently, it was not necessary for traffic monitoring

and control to meet real-time requirements. However,

due to increasing digitalization, the necessities of

connected and automated driving, security

requirements, and new services in the C-ITS area,

cities and municipalities are facing completely new

challenges. This is one of the reasons for the current

focus is on transport infrastructure. There is an

enormous amount of catching up to do in terms of

open data provision while at the same time

implementing security mechanisms for the critical

infrastructure as well as dedicated user and rights

management.

Virtual traffic lighting is also often discussed in

the literature (Bazzi et al., 2016; Zhang et al., 2018).

In this case, physical signals could be completely

dispensed with. However, since a physical signal will

always be necessary for VRUs such as cyclists and

pedestrians, this approach will not be considered

further here.

For the reasons mentioned, so-called open data

access to data from the operational transport

infrastructure has so far been rare. The city of

Hamburg is one of the pioneers in Germany in this

regard. Real-time data from a variety of traffic lights

in Hamburg can be accessed via Hamburg’s urban

data platform or the Mobilithek as national access

point for ITS data in Germany. The data set contains

process data for a large number of intersections in

Hamburg. Among other things, current signal state

information can be accessed in real-time. There is

also data from detector inputs from pedestrians,

cyclists, vehicles, and public transport. An example

of this is shown in Figure. 1.

Figure 1: Open data example – no.537 [geoportal-

hamburg.de] upper: geometry of the intersection including

signal layout plan lower: current signal status of the LSA

including MAPEM and SPATEM information.

This paper introduces stability indicators based on the

publicly available traffic data. Here we build on

previous work (Otto et al., 2022), where we

introduced the theoretical background for the analysis

of signalized intersections. Furthermore, (Jeschor et

al., 2024) show results of a large-scale predictability

evaluation in Hamburg, which also provide

statements about consistency.

The derived tool enables the analysis of historical

data from existing systems to derive an indicator for

the quality and suitability of the C-ITS GLOSA

application. We can show that intersections with

Intersection with Highly Adaptive Trafﬁc Lights Can Still Be Suitable for C-ITS Service GLOSA

149

highly adaptive traffic control schemes can still be

suitable for GLOSA.

2 METHOD

Basis of our analysis is the data set provided by the

city of Hamburg. Our goal was to derive stability

indicators for a wide range of intersection topologies

and traffic volumes. For this reason, the

characteristics of the intersections in question should

be completely different. In the end, we settled on

three different intersections (277, 353, 835). The

structural design of these intersections can be seen in

the aerial photos in Figure 2.

Figure 2: Intersections no.277, no.353 and no.835 (up to

down) [maps.google.com] intersection with different

structural characteristics as well as different traffic volumes

and relations.

The chosen intersections can be described as follows:

 Intersection no.277: This is the largest

intersection in the selected subset. Traffic

volumes are high during peak hours and

signalization is relatively complex. The

systems are designed with traffic-adaptive

control. The signal programs are designed in

such a way that they can react to traffic

fluctuations during the day. Cycling and

pedestrian traffic are carried on almost all

routes.

 Intersection no.353: This is a relatively typical

intersection with a medium geographical

extent. The traffic load also differs throughout

the day, to which the implemented adaptive

control can react.

 Intersection no.835: Within the specified

selection, this is a T-intersection that has the

lowest complexity.

Our analysis is based on the signaling parameters of

the intersections. Our research hypothesis is that these

alone are sufficient to find stability indicators,

regardless of intersection layout and size. The

following parameters were examined:

 Number of signal programs and activated times

of these programs,

 Cycle time of signal programs throughout the

day,

 Length of the green time and area of the green

time based on the cycle second,

 Sequences of green times in cycle and variation

of the sequences,

 Variation of green start and green end of the

signal groups in the cycle based on the signal

program,

 Relation via matrices of the start times and end

times of the signal groups to each other to

analyze the interval,

 Representation of the probabilities of the signal

groups for green in the cycle based on the data.

Based on the parameters mentioned above, it should

be determined which of these factors can be used

significantly to map the GLOSA suitability of

adaptive controls.

3 RESULTS

The results are evaluated based on the three

intersections described above. To simplify the

presentation, only the results of intersection 353 are

VEHITS 2025 - 11th International Conference on Vehicle Technology and Intelligent Transport Systems

150

presented and shown specific in this paper. This is

intended to facilitate general understanding and

provide an easy introduction to the scientific

evaluations of the tool.

3.1 Geo Data

The geographical coordinates of the individual signal

groups at the intersection were extracted from the

data and clearly displayed using the GIS program

QGIS and the Google satellite map. Fig. 3 shows

intersection 353 with all primary signal groups. The

signal groups are separated by color according to lane

type.

Figure 3: Visualization of geo data at intersection 353 – no

of intersection as well as no of signal controller

[maps.google.com].

In addition to vehicle and pedestrian signal groups,

there are also signal groups for bicycle and bus lanes

with the corresponding signaling systems at this

intersection. Mixed lane types are also available. In

total, there are 36 signal groups.

3.2 Length of Green and Cycle Time

First, the duration of the green state for each signal

group at the intersection was determined from the

data. For this purpose, the observation times for

vehicle, bus and bicycle signal groups were extracted

from the existing data. The selected data were sorted

according to its time in order to determine the time

difference in seconds between successive green and

amber, green and red or green and dark. The

following Figure 4 upper part shows an example of

the green phase duration for the vehicle signal group

353_10. The times were grouped on the x-axis by the

hours of the day and presented with boxplots. It can

be seen from the figure that the green times show a

large variance with the exception of 00:00 to 03:00,

17:00 to 19:00 and 22:00 to 23:00.

A similar procedure was used to determine the

cycle times. For this purpose, only times with signal

state green were extracted from the data. Regardless

of the lane type, the cycle time was then determined

from the sorted data using the time differences

between two consecutive green. The following Fig 4

lower part shows the cycle time using the example of

the vehicle signal group 353_10. Similar to the green

Figure 4: Intersection no.353 – upper: duration of green

phases for signal program 1 and signal group 353_10

throughout the day lower: duration of cycle time for signal

program 1 and signal group 353_10 throughout the day.

stages, the cycle times also show variances, with

the exception of the times from 00:00 to 03:00, 17:00

to 19:00 and 22:00 to 23:00. Furthermore, it becomes

clear that the average from 00:00 to 04:00 and 20:00

to 22:00 is 75 seconds, while at the other times with

the exception of the three hours from 09:00 to 10:00

a.m. as well as 1:00 p.m. to 2:00 p.m. and 11:00 p.m.

to midnight is 90 seconds. (see Figure 4 lower part)

3.3 Signal Programs

In the next step, the signal program data was

evaluated separately for each signal group.

For this purpose, all observation times were

recorded from the data using the respective signal

program and sorted by time. The observation time

indicates the time of switching to another program. In

the following Figure 5, the data for a period of three

days from February 1st, 2023 (Wednesday) to

February 3rd, 2023 (Friday) was presented.

Intersection with Highly Adaptive Trafﬁc Lights Can Still Be Suitable for C-ITS Service GLOSA

151

Figure 5: Intersection no.353 – current no of signal

programs throughout the day.

3.4 Sequences

In the first step of the evaluation, the sequence was

determined. This indicates the order in which the

signal groups switch to green, taking into account the

current signal program. For this purpose, only the

observation times in which the signal color was green

were extracted from the data. These dates were sorted

by their time. The entire day's data was then assigned

to the current signal program according to its time.

The first signal group in the series is the one with the

lowest time in the signal program data group. This

signal group is fixed at position 1. The sequences are

then recorded according to their time. A new

sequence starts with the next recurrence of the fixed

signal group. The following figures show examples

of the signal group sequences for two signal

programs.

The frequency of the sequences was made

visually visible using transparent dots. Sequences

shown in very pale blue occur only rarely, while the

main strands are marked in strong blue. The

representations show significant differences in the

variance of the sequences. While the signal programs

1 have a large number of sequences and sometimes

have several main sequences, a clearly fixed sequence

can be seen in the signal programs 4.

3.5 Start and end Times of Green

Signal

To evaluate the specific green start and green end

times within the signal group sequence, the

observation times were extracted from the data,

which contained the signal colors green to amber for

Figure 6: Intersection no.353 – sequence of signal groups -

upper: program 1 – lower: program 4.

the vehicle signal groups or green to red for the

pedestrian signal groups. The data was assigned to the

current signal program according to its time.

Analogous to the procedure for determining the signal

group sequence, the first signal group to appear was

fixed and the sequences were then determined based

on the data with the signal color green. The following

figures show the results for the signal programs.

The time difference in seconds to the fixed signal

group at position 1 is plotted on the x-axis. The

transparency of the points indicates the frequency of

occurrence of the green start (shown in green) and

VEHITS 2025 - 11th International Conference on Vehicle Technology and Intelligent Transport Systems

152

green end times (shown in orange). As with the

determination of the sequences, it is also clearly

visible from these figures that signal programs 1 has

a clear variance in the green start and green end times,

while signal programs 4 largely has fixed time

differences as well as fixed green start and green end

times.

Figure 7: Intersection no.353 – start/end green upper:

program 1 – lower: program 4.

3.6 Matrix Interval Between Signal

Groups

An essential factor for the evaluation is the time

intervals between the signal groups. Before the

matrices could be set up, the specific times between

the green starts (GA) and ends (GE) in all four

combinations (GAGA, GEGA, GAGE, GEGE) as

well as their variance had to be determined. The

following figure shows an example of the result of the

GAGA time differences for the vehicle signal group

353_10 in signal program 2. All signal groups are

plotted on the x-axis.

Figure 8: GAGA-matrix signal program 2 – example of

duration between greenstart and next greenstart.

In this case (GAGA time difference), the cycle time

is shown for the signal group 353_10 to itself. It can

be seen that without exception the signal group

353_10 switches to green at the same time as the

signal groups 353_1, 353_11, 353_17, 353_18,

353_19, 353_6 and 353_9, since the time difference

here is 0 in each case. Fixed time differences can also

be seen for the signal groups 353_15, 353_39 and

353_49, although there are some outliers. This gives

a clear indication of the interval in the stage

transition. For all other signal groups, time

differences with significant variances can be seen.

The box plots were drawn up for each individual

signal group and for all four combinations (GAGA,

GEGA, GAGE, GEGE). They form the data basis for

the matrices.

The intervals of the box plots from the bottom to

the top whiskers were first determined in the matrices.

To assess and visualize the forecast of the green starts

and ends of a signal group relative to the times of the

other signal groups. The intervals were classified into

the groups 0 to 3 seconds (green, well predictable), >

3 to 7 seconds (yellow, moderately predictable) and >

7 seconds (red, poorly predictable). The intervals and

their respective classification in this color scale are

shown in the matrices. Example matrices can be seen

in the following figures.

Intersection with Highly Adaptive Trafﬁc Lights Can Still Be Suitable for C-ITS Service GLOSA

153

Figure 9: Example matrices of GAGA, GEGA, GAGE and

GEGE – signal program 2.

After setting up the four individual matrices, all

matrices were overlayed. For this purpose, the

individual cell values of the four matrices were

compared with each other. The smallest cell value

from the four matrices is given in the overlay matrix.

The following figure shows the overlay matrix for

signal program 2.

Figure 10: Example matrix green start to green end

minimum span visualization.

The representation shows which signal groups can be

predicted well or poorly with regard to the light signal

changes of another signal group.

3.7 Probability of Green per Cycle

As a final step in the evaluation, the green times per

cycle time were evaluated. For this purpose, the main

strands of the signal group sequence as well as the

green start and green end times were used as a basis.

The green times resulting from the time difference

between the start of green and the end of green were

used to determine the frequency with which the signal

group is switched to green and at what second of

switching. The following figure shows an overview

of the green phases for signal program 1.

Figure 11: Intersection no.353 – Probability of green per

cycle- program 1.

The most frequently occurring sequence plus all

sequences with at least 80% similarity of the

respective sequence to the main sequence was used as

the data basis. The intensity of the green or red cells

visualizes the probability with which the signal group

shows the red or green status at the respective cycle

seconds.

Although the intersections in question show a

certain variability in signal programs, cycle time, etc.

our analysis has shown that due to interrelation

between the different signals there follows a certain

predictability of the signal program. This becomes

apparent in Fig. 10, where most cells in the

predictability matrix are zero, i.e., there is at least one

relation (green start/end, red start/end) between the

signals based on a fixed time period. This allows a

high-quality forecast based on current signal states or

more specifically signal state changes. In addition to

this short-term information, historical analysis also

yielded certain signal groups and time periods within

the cycle, where a certain status of the signal can be

assured (comp. Fig. 11).

VEHITS 2025 - 11th International Conference on Vehicle Technology and Intelligent Transport Systems

154

4 CONCLUSIONS

The commonly suggested thesis in research work and

in traffic engineering practice that the more flexible

the control, the less reliable the forecast and thus the

functionality of GLOSA could be refuted in this

research. The efforts of the last 30 years to implement

the most flexible control possible, which adaptively

takes into account the needs of all road users,

apparently contradict the requirement for a stable

forecast for the GLOSA service. In this article, a tool

chain was presented that shows indicators that at least

partially refute this thesis. The tool enables the

analysis of historical data from existing systems to

derive an indicator of the quality and suitability of the

C-ITS GLOSA application.

Based on the presented and differentiated

systems, it is clear that there are defined indicators

that can provide information about the stability of the

forecast both in the planning process and in the

operation of the system. The prediction is then easily

possible for most signal groups despite adaptive

control over the intermediate times. This leads to very

good results within the overall forecast, especially

when superimposing a long-term forecast from

historical data with a short-term forecast using

sequence, cycle time, green time and intermediate

time matrices.

It should be said here that the toolchain mentioned

is evaluated by traffic lights of only one city. The

systems correspond to the typical planning principles

in Germany and Europe. In most cases, rule-based

logic is stored, which has a fixed circulation time.

Furthermore, the traffic light programs are phase-

related controlled, where fixed phase transitions,

minimum and maximum release times and other

boundary conditions such as coordination - so-called

green waves - exist. Obviously, these boundary

conditions do not apply worldwide, but they are easily

applicable throughout Europe.

Finally, it should be noted that the detector inputs

are of course also of crucial importance, especially for

short-term forecasts. This influence has currently

only been taken into account indirectly. The aim of

further investigations will be to determine

correlations between the detector inputs and the

corresponding signal groups. The prioritization in

particular, the registration for the prioritization of

public transport, is often done very early and can

therefore be easily taken into account in the forecast.

This influence will be taken into account in further

analysis and research.

ACKNOWLEDGEMENTS

We would like to thank Julia Erdmann for her

visualizations of the traffic lights data. We also thank

Per-Arno Plötz and Henning David (City of

Hamburg) for valuable input interpreting the traffic

lights data set.

This study was financially supported by the

European Union.

REFERENCES

Barthauer, M., & Friedrich, B. (2014). Evaluation of a

signal state prediction algorithm for car to

infrastructure applications. Transportation Research

Procedia, 3, 982-991.

Bazzi, A., Zanella, A., & Masini, B. M. (2016). A

distributed virtual traffic light algorithm exploiting

short range V2V communications. Ad Hoc Networks,

49, 42-57.

Bodenheimer, R., Eckhoff, D., & German, R. (2015,

October). GLOSA for adaptive traffic lights: Methods

and evaluation. In 2015 7th International Workshop on

Reliable Networks Design and Modeling (RNDM) (pp.

320-328). IEEE.

Eckhoff, David, Bastian Halmos, and Reinhard German

(2013). Potentials and limitations of green light optimal

speed advisory systems. IEEE Vehicular Networking

Conference. IEEE, 2013.

Eteifa, S., Rakha, H. A., Eldardiry, H. M., & Center, E.

(2021). Estimating Switching Times of Actuated

Coordinated Traffic Signals: A Deep Learning

Approach (No. UMEC-039). Urban Mobility & Equity

Center.

Genser, A. (2022). Machine learning for traffic

management in urban transportation networks

(Doctoral dissertation, ETH Zurich).

Jeschor, D., Matthes, P., Springer, T., Pape, S., & Fröhlich,

S. (2024, June). Cloudy with a Chance of Green:

Measuring the Predictability of 18,009 Traffic Lights in

Hamburg. In 2024 IEEE Intelligent Vehicles

Symposium (IV) (pp. 2882-2889). IEEE.

Krumnow, M. (2023). Prädiktion von Signallaufzeiten

verkehrsadaptiver Lichtsignalanlagen zur

Unterstützung von C-ITS Anwendungen.

Mellegård, N., & Reichenberg, F. (2020). The day 1 C-ITS

application green light optimal speed advisory—A

mapping study. Transportation Research Procedia, 49,

170-182.

Otto, T., & Hoyer, R. (2010). Operating conditions of on-

board displayed green wave speeds via V2I-

communication. Proceedings fovus-Network for

Mobility.

Otto, T., Klöppel-Gersdorf, M., & Partzsch, I. (2022,

August). A Framework for Urban C-ITS GLOSA

Evaluation. In Conference on Sustainable Urban

Intersection with Highly Adaptive Trafﬁc Lights Can Still Be Suitable for C-ITS Service GLOSA

155

Mobility (pp. 382-392). Cham: Springer Nature

Switzerland.

Otto, T., & Weichenmeier, F. (2013). Self-learning

algorithm and signal state prognosis at traffic lights for

V2I applications. In Proceedings 20th ITS World

Congress ITS Japan.

Protschky, V., Wiesner, K., & Feit, S. (2014, June).

Adaptive traffic light prediction via Kalman filtering. In

2014 IEEE Intelligent Vehicles Symposium

Proceedings (pp. 151-157). IEEE.

Scheegans, L. E., Heckmann, K., & Hoyer, R. (2022,

September). Exploiting Stage Information for

Prediction of Switching Times of Traffic Actuated

Signals Using Machine Learning. In 2022 12th

International Conference on Advanced Computer

Information Technologies (ACIT) (pp. 544-548). IEEE.

Suzuki, H., & Marumo, Y. (2020). Safety evaluation of

green light optimal speed advisory (GLOSA) system in

real-world signalized intersection. Journal of Robotics

and Mechatronics, 32(3), 598-604.

Weisheit, T., & Hoyer, R. (2014). Prediction of switching

times of traffic actuated signal controls using support

vector machines. In Advanced Microsystems for

Automotive Applications 2014: Smart Systems for

Safe, Clean and Automated Vehicles (pp. 121-129).

Springer International Publishing.

Zhang, R., Schmutz, F., Gerard, K., Pomini, A., Basseto, L.,

Hassen, S. B., ... & Tonguz, O. (2018, August). Virtual

traffic lights: System design and implementation. In

2018 ieee 88th vehicular technology conference (vtc-

fall) (pp. 1-5). IEEE.

VEHITS 2025 - 11th International Conference on Vehicle Technology and Intelligent Transport Systems

156