Register now After registration you will be able to apply for this opportunity online.
Reverse Engineering of traffic lights schedules from data
Modern traffic simulations have become valuable tools for studying traffic dynamics and developing
novel traffic management policies. Using tools such as SUMO and Aimsum NEXT, researchers can
realistically represent a traffic network’s response to a given policy. However, The fidelity of such
simulations is deeply impacted by factors such as the scheduling of the traffic lights, undermining
their utility as a tool. In this work, we aim to develop a scalable method to detect the scheduling of a traffic light based solely on loop-detector data. We will validate the designed algorithm using real
data and state-of-the-art micro-simulators.
In the past decades, researchers have developed many control techniques to mitigate traffic
congestion. Several of these works focus on dynamically influencing the flow of vehicles, for ex-
ample, via dynamic traffic lighting, variable speed limits, or ramp metering. Many of these
works utilize simulations to empirically validate their methodologies. When simulating real traffic
systems, such as the city of Zürich, the validity of these results depends on the alignment between
the simulation and reality. One of the factors that these simulations often fall short of reflecting is the
traffic light schedules. The information on these schedules is hard to come by and often contained
in unstandardized documents, making it difficult and unscalable to retrieve. In this project, we aim to develop a scalable methodology to detect traffic light schedules based
solely on publicly available data. In particular, we want to develop an algorithm that will iteratively
estimate the state of the traffic light using loop-detector data. This type of detector, the most com-
monly used in cities today, provides a measurement of occupancy based on the time for which it is
activated by a car. This problem is non-trivial due to multiple possible schedules being exhibited by
one traffic light based on the time of the day and the presence of humans in the system, which do
not always respect the scheduling causing noise in the measurement. To overcome these problems,
we want to develop an algorithm based on Particle Filtering to retrieve the correct state of the traffic
light in the presence of noise.
Particle filtering is an effective strategy to estimate the state of a system based on sequential
Monte Carlo sampling. It reproduces the function of a Kalman filter without any assumption of linear-
ity on the system dynamics and for non-Gaussian disturbances. This makes it a natural candidate
for obtaining the estimates of the traffic light schedules in a traffic system. As a result, we aim to
utilize Particle filtering to estimate the traffic like states, using data from loop-detectors scattered
across a city.
In the past decades, researchers have developed many control techniques to mitigate traffic congestion. Several of these works focus on dynamically influencing the flow of vehicles, for ex- ample, via dynamic traffic lighting, variable speed limits, or ramp metering. Many of these works utilize simulations to empirically validate their methodologies. When simulating real traffic systems, such as the city of Zürich, the validity of these results depends on the alignment between the simulation and reality. One of the factors that these simulations often fall short of reflecting is the traffic light schedules. The information on these schedules is hard to come by and often contained in unstandardized documents, making it difficult and unscalable to retrieve. In this project, we aim to develop a scalable methodology to detect traffic light schedules based solely on publicly available data. In particular, we want to develop an algorithm that will iteratively estimate the state of the traffic light using loop-detector data. This type of detector, the most com- monly used in cities today, provides a measurement of occupancy based on the time for which it is activated by a car. This problem is non-trivial due to multiple possible schedules being exhibited by one traffic light based on the time of the day and the presence of humans in the system, which do not always respect the scheduling causing noise in the measurement. To overcome these problems, we want to develop an algorithm based on Particle Filtering to retrieve the correct state of the traffic light in the presence of noise. Particle filtering is an effective strategy to estimate the state of a system based on sequential Monte Carlo sampling. It reproduces the function of a Kalman filter without any assumption of linear- ity on the system dynamics and for non-Gaussian disturbances. This makes it a natural candidate for obtaining the estimates of the traffic light schedules in a traffic system. As a result, we aim to utilize Particle filtering to estimate the traffic like states, using data from loop-detectors scattered across a city.
The goals of the project are as follows:
1. Learn about Particle Filtering;
2. Formalize the problem of traffic lights schedule estimation from data;
3. Develop an algorithm based ion Particle filtering to estimate the traffic light schedules
4. Validate the designed algorithm via micro-simulations and real data coming from the city of
Zürich.
Publications: If the final results are promising they can potentially be turned into a publication.
The goals of the project are as follows: 1. Learn about Particle Filtering; 2. Formalize the problem of traffic lights schedule estimation from data; 3. Develop an algorithm based ion Particle filtering to estimate the traffic light schedules 4. Validate the designed algorithm via micro-simulations and real data coming from the city of Zürich. Publications: If the final results are promising they can potentially be turned into a publication.
MSc_BSc_Proposal_traffic.pdf
