Вы находитесь на странице: 1из 12

A methodological review of travel time: The role of driving behaviour

in heterogeneous traffic congestions

Correa, J.

ABSTRACT
The aim of this paper is twofold. In first place, it summarizes the pros and cons of existing
methods for predicting and modelling travel times at controlled intersections from the traffic
engineering viewpoint. Secondly, it illustrates the use of multi-agent systems to study travel times
and related metrics of traffic in controlled intersections by considering the role of driving
behaviour in heterogeneous traffic congestions. By introducing “motorcycle lane-sharing” as a
punctual case of heterogeneous traffic, the paper suggest the need to explore the relationship
between driving behaviour and traffic control in multi-agent traffic simulations.

Keywords: Travel time, driving behaviour, traffic heterogeneity; controlled intersections;


multiagent systems.

1. Introduction

Beyond existing reviews on the topic of travel time (Noland & Polak, 2002) and its
treatment in modelling methods (F. Zheng, 2011), little is known on the relationship among
travel time and driving behaviour in heterogeneous traffic congestions at controlled
intersections.

The relevance to study these relationships is threefold. In first place, these relationships can
be useful in broaden the understanding of traffic congestion as a complex system which can
be simulated from the multi-agent system paradigm (Bazzan, 2007). Secondly, it highlights
the role of driving behaviour in predicting travel time according to individual variables
such as “politeness” (Kesting, Treiber, & Helbing, 2007) or “courtesy” (Hidas, 2002).
Finally, it might have the potential to clarify the role of traffic control strategies to reduce
traffic congestions (and consequently travel times) when these variables might change over
time. The aim of this paper is to review the literature on the existent methods for predicting
travel time, and introduce multi-agent system as a suitable technique to explore these
relationships.

The paper is structured as follows. Section two summarizes existent methods to study travel
time in controlled intersections from the traffic engineering viewpoint. In such an overview
data collection procedures are distinguished from modelling methods. The role of driving
behaviour in multi-agent traffic control simulations is the aim of the third section in which
motorcycle lane-sharing is introduced as a punctual case of heterogeneous traffic to suggest
the relevance to explore the relationship between driving behaviour and the efficiency of
traffic control strategies.
2. Existent methods

Travel time can be studied in two realistic settings, namely, free traffic flows (i.e.,
freeways) and interrupted traffic flows (i.e., intersections in road networks). Most of the
current knowledge on travel time models come from the former, while the latter have only
been recently studied (Liu, 2008; Viti, 2006; F. Zheng, 2011). The relevance to study travel
time in interrupted traffic flows is its utility to differentiate its mechanisms from those in
freeways (F. Zheng, 2011). This section, as an introduction to the next one, is conveniently
subdivided in two parts, namely: procedures of data collection and modelling methods.

2.1. Procedures of data collection

Travel time can be empirically measured from two information sources, namely: via survey
methods (SM) to evaluate individual desired route choices based on perceptions of travel
time variability (Noland & Polak, 2002); and via traffic sensors (TS) to calculate
probabilistic estimations of travel time from the data collected with fixed sensors or mobile
sensors (F. Zheng, 2011).

One of the main limitations in both methods is the great amount of factors (e.g., density,
road network size, day of the week, measurement periods, etc.) that influence the precision
of probabilistic estimations of travel times (Srinivasan & Jovanis, 1996). Due to these
limitations, an active and on-going research in the field of intelligent transport systems
(Cherri, Nodari, Toffetti, & Information Society Tecnologies [IST], 2004) is producing a
vast amount of technological procedures to collect data that allow the probabilistic
estimation of travel times in urban networks with different devices, such as “probe
vehicles” (PV) (Bhaskar, Chung, & A.-G. Dumont, 2011), “automatic number plate
recognition” (ANPR) with traffic videos (Friedrich, Jehlicka, & Schlaich, 2008) and other
communication-based technologies (e.g., GPS, Bluetooth, mobile phones, etc.). The pros
and cons of existent procedures of data collection are summarized in Table 1.
Table 1 Comparisons of existent procedures of data collection
Method Pros Cons
 Sampling procedure for selecting representative road
SM (Noland &  Ease of implementation users
Polak, 2002)  Ease of analysis  Sensitivity to learning effects and credibility

 Sensitivity to oversaturated traffic situations at


 Accurate calculations for intersections
ANPR travel time between two  It assumes that motorist drive their vehicles along the
(Friedrich et al., adjacent points adjacent points (which is not necessarily true due to
2008)  High recognition rate (i.e., 85- unexpected stops, accidents, etc.).
90%)  Difficulties in proving if it works well in road networks

 Positioning requires four satellites, not always available


PV with GPS
 The reflection of satellites signals has a significant
(Davidson,  Increasing popularization of
impact on positioning.
Hautamaki, Collin, GPS devices in vehicles
& Takala, 2009)  Accurate data requires high frequency transmission

PV with mobile  Low positioning accuracy ranging from 50m up to 300m


 Popularization of mobile
phones  Lacks discrimination whether the mobile phone is inside
phones inside vehicles near
(Wunnava et al., a vehicle or it belongs to a cyclist or a motorcyclist
mobile antennas
2007)
 Detection signals lacks according to vehicle speed
 Provide a good representative  For shorter lengths inappropriate detection moments can
of the actual traffic conditions cause large errors.
PV with Bluetooth
 The constant broadcast is  The error is larger is two or more Bluetooth devices are
(Yegor, Wu, Wang,
detectable maintaining placed in a short distance or close to intersections.
& Lee, 2010)
personal information  It assumes that motorist drive their vehicles along the
anonymously adjacent points

2.2. Modelling methods

There are two approaches in predicting travel time; namely “model based” and “data
driven” approaches (Liu, 2008). Model based approaches explain traffic processes on
physical mechanisms, while in data driven approaches the prediction of travel time is done
without explicitly addressing such processes, although in practice they are relatively easy to
implement and have shown good performances.

In model based approaches, there is another distinction in terms of “macroscopic models”


and “microscopic models”. In macroscopic models, travel time is related with the
mechanisms of traffic streams with macroscopic variables (i.e., speed, flow and density)
(Liu, 2008); while in microscopic models, travel time is related with the mechanisms of
individual vehicles trajectories by analysing microscopic variables (i.e., vehicle type, lane
changing, headways, driver’s characteristics, driving behaviour) (Brackstone & McDonald,
1999).

Liu (2008) distinguishes three types of model based approaches, namely: “simulation
models” (SimM), “delay formulas” (DF), and “queuing based models” (QBM); and just
recently Zheng (2011) incorporates “cell transmission models” (CTM), and “cumulative
vehicle plots” (CVP) as another pair of approaches. The pros and cons of these models are
summarized in Table 2.

Table 2 Comparisons of existent travel time model based approaches


Method Pros Cons
 Easy implementation based on
 Highly computational complexity
origin-destination (OD) matrices
SimM  Highly expertise required
 Evaluate travel time in different
(Bierlaire &  Intensive model parameter and calibration
scenarios under several traffic
Crittin, 2004)
conditions

 Only valid when average flow rate does not exceed


average capacity rate (not able to deal with congested
traffic situations)
 The assumption that the probabilistic distribution of
arrivals (i.e., poisson) does not change but over time, is
 Widely used because of its
DF (Dion, hardly met in realistic situations.
simplicity (macroscopic
Rahka, &  It assumes a known distributions (with a constant means)
viewpoint)
Kang, 2004) for headways between vehicles from the stop line
 It assumes to run long enough to have a steady
condition.
 The validity is constrained in comparison with the large
variety of cases observed in realistic situations

 It requires the measurement of a queue length which is


QBM (Takaba,
difficult to collect directly
Morita, Hada,  It is applicable for congested
 It is sensible to saturation flow and jam density, because
Usami, & situations
Yamaguchi, these variables are not constants over time as it assumes,
but they tend to vary
1991)
 It assumes that queue formation is deterministic which
 It can be applicable for real traffic
may not be realistic for urban networks, since vehicle
situations by analysing
arrivals and departures are not deterministic but rather
CTM macroscopic variables.
stochastic following following certaing statistical
(Daganzo,  Easy to implement for its
distributions (e.g., Poisson, binomial).
1994, 1995) transparency, and straightforward
 It does not consider microscopic variables in the
rapid calculations for congested
analysis.
and non-congested situations
CVP (Bhaskar,  It is applicable for in between two
 It assumes that there is no counting error at the
Chung, & A. G. road sections, based on fixed
downstream intersection which is unrealistic
Dumont, 2009) sensors and probe vehicles

According to Zheng (2011) these methods “are very complex to implement in practice due
to the estimation or prediction requirements to traffic demand and supply at the model
boundaries as inputs” (p. 21). In contradistinction, data driven approaches represent another
viewpoint which consist in assuming that travel time is linearly and/or non-linearly
dependent of influencing factors or their combination. Liu (2008) summarized three types
of models in this second category, namely: “regression based models” (RBM), “K-nearest
neighbour models” (KNNM) and “Markov Chain Models” (MCM); and Zheng (2011)
incorporates “Time series methods” (TSM), “Fuzzy logic based approaches” (FL) and
“Neural networks methods” (NNM). The pros and cons of these methods are summarized
in Table 3.

Table 3 Comparisons of existent travel time data driven approaches


Method Pros Cons
 It is used for smaller segments of road networks, limiting
 Ease of implementation by travel time prediction for larger routes.
RBM
employing multivariate regression  The parameters of the models are static, which raises
(Sisiopiku,
formulas and fixed traffic sensors doubts on their utility in evaluating dynamic traffic
Rouphail, &
to know arrival times processes as time goes by.
Santiago, 1994)

 There are no standard rules to select the variables which


constitute the state vector in the analysis.
 It is difficult to estimate the spatial and temporal seize of
the variables.
 An insufficient temporal size of the variables can cause
 The concept of “nearest-
KNNM an incomplete image of the traffic searched in historical
neighbour” is straightforward and
(Robinson & databases, which leads to poor estimations and
its ease of implementation.
Polak, 2005) predictions.
 It requires a large amount of historical data.
 If the input state vector is located at the boundary of the
state space of the existing observations, the results will
be biased.

 It assumes homogeneous traffic conditions for a single


vehicle type, when in fact, traffic composition is a
MCM (van  It deals with dynamics traffic
stochastic variable.
Zuylen & variables (i.e., queues lengths)
 It assumes saturation flow and vehicles arrivals as a
Hoogendoorn, under the assumption of constant
constant with a certain known probabilistic distribution
2007) parameters
over the time period under analysis.

 It captures general trends, having difficulties for


TSM (Szeto,  Speed, flow, density or travel time
capturing non-recurrent traffic behaviour or the
Ghosh, Basu, & in previous time intervals can be
transition from congestion to free traffic flow.
O’Mahony, included to predict travel time in
2009) next time intervals
 It requires the determination of the number of
membership functions and there are no single rules for
 It allows imprecise input data for
such
FL (P. Zheng & predicting travel time
 It requires large historical database, covering different
McDonald,  It might be helpful when there is
traffic patterns and a sufficient expertise knowledge for
2009) no simple mathematical model for
the formulation of different rules for fuzzification and
highly non-linear processes
defuzzification.

 It has the potentials to learn


 The prediction error on the training set is small, but it
complex non-linear relationships
grows as long as new data is added.
between variables by identifying
 It tends to have poor generalizability (e.g., the data for
the patterns in the data
NNM (You & morning peak hours cannot be used to predict travel
 It does not require extensive
Kim, 2000) times in other time periods.
expertise on traffic flow
 The results of these models are location-dependent (i.e.,
modelling, being fast and easy to
the model developed for one route or link, cannot be
implement with “ready-to-use”
applied to other routes).
available software
As can be seen from Table 3 the difficulty in estimating travel time for road networks is
basically the same for that of the model based approaches; namely, the assumption of static
values for estimated parameters without a clear understanding of how such parameters
might change over time. A recent overview of the existing methods for the prediction of
travel time was prepared by Liu (2008) and Zheng (2011) and these overviews are
reproduced in Table 4 and 5.

Table 4 An overview of existing methods for predicting travel time


Method Spatial Data for Real-time Signals Turning Variance
scope validation application traffic
SimM Route Both Yes Yes Yes Yes
DF Segment Both No Yes No No
QBM Segment Empirical Yes Yes No No
RBM Segment Empirical Yes Yes No No
KNNM Segment Empirical Yes No Yes Yes
MCM Route Simulation No No No No
TSM Segment Empirical Yes Yes Yes Yes
FL Segment Empirical Yes No No Yes
NNM Segment Empirical Yes Yes Yes Yes

Table 5 Input data of existing methods for predicting travel time


Method Input data
SimM
Individual driver behaviour, OD and route choice
(Macro)
SimM
Speed, OD and route choice,
(Micro)
DF Saturated flow rate, signal timing, arrival flow, free flow travel time
RBM Occupancy, signal time
KNNM Flow, occupancy
MCM Delay proabability
TSM Large historical average travel time
FL Large historical average travel time
NNM Large historical average travel time

As can be seen from Table 4 and 5, the estimation and prediction of travel time for
controlled intersections resides in the limitations of the methods employed for such
purposes, since none of the methods are capable to include all relevant factors that occur
concomitantly in the real world from a dynamical viewpoint. In this sense, the multi-agent
systems paradigm (Wooldridge, 2002) might be promising in surpassing the
aforementioned limitations.
3. The role of driving behaviour in multi-agent traffic control

Most of the methods in the previous section come from the traffic engineering literature
(Dowling, Skabardonis, & Alexiadis, 2004). These methods are somehow limited in terms
of the spatial scope they deal with (i.e., focused only in segments of road networks).
Computational methods (i.e., SimM or MCM), in contradistinction, have the potential value
to study travel times and related metrics of traffic in broader spatial scopes, allowing the
researcher to design experiments in order to test the effectiveness of traffic control
strategies by considering traffic control devices (i.e., traffic lights) as adaptive intelligent
agents which can be able to learn “optimal solutions” to reduce traffic congestions at large
scales according to traffic volumes of vehicles approaching to several intersections of road
networks.

Although some computational approaches from the framework of “Markov Decision


Processes” (MDP) and “Multi-Agent Reinforcement Learning” (MARL) (Bakker,
Whiteson, Kester, & Groen, 2010) seem to overcome some of the problems and challenges
to reduce traffic congestions at large scales (Bazzan, 2009), it should be noticed that almost
all of these approaches lack to grasp the role of driving behaviour as another component of
the whole system by making few assumptions, if any, on the relationship between driving
behaviour and the efficiency of traffic control at large scales. In the next section,
“motorcycle lane-sharing”, as a punctual exemplar of heterogeneous traffic (i.e., in which
there are several types of vehicles on the road) is introduced to suggest the importance to
explore whether or not driving behaviour could play a key role in controlling or reducing
traffic congestions at large scales.

3.1. Motorcycle lane-sharing and driving behaviour: Is it important to consider driving


behaviour in multi-agent traffic control simulations?

Motorcycle lane-sharing (lane-splitting or filtering) is referred to the movement of


motorcycles between lanes of stopped or slow-moving vehicles (Sperley & Joy Pietz,
2010). This phenomenon seems to be increasing in California, UK and some other Latin-
American countries, which suggests that motorcycles are indeed a full mode of
transportation in congested urban areas wherein this practice is allowed by the local
authorities (Wigan, 2002). The literature on this topic is limited, due to the absence of
empirical studies focused on the benefits or safety concerns of this practice and because of
the scarce availability of quantitative data. Nonetheless, this situation can be illustrating in
clarifying whether or not driving behaviour could play an important role in controlling or
reducing traffic congestions at large scales.
In previous works, some behavioural factors such as “the driver courtesy” (Hidas, 2002) or
“the politeness factor” (Kesting et al., 2007) has been included as individual parameters for
modelling lane-changing, but there is no clear evidence on the relevance of these
behaviours on traffic congestions or whether these behaviours might change over time.

In congested situations, when motorcyclists are allowed to drive in between queues of


vehicles, it could be expected that the average speed for motorcyclists would be higher in
comparison with the average speed for motorists (the higher the speed, the lower the travel
time), due to the benefits of motorcyclists in these situations. Nonetheless, it is not clear
whether or not motorcycle lane-sharing is also related with motorists’ performance in terms
of their average speed, their frequency of brakes use and/or their frequency of lane-
changing, and if it does what would it be the net impact of motorcycle lane-sharing on these
individual behaviours. Even in the case that motorcyclists should drive as motorists (i.e.,
inside the lane and behind the in-front vehicle) it is not clear whether or not the “fatigue” of
drivers, as a psychological consequence of inefficient traffic control strategies in
oversaturated traffic conditions, could be an important factor that might explain a
significant decrease in drivers’ performances, although the experimental evidence suggest
so (Philip et al., 2003).

The distinction between “cooperative drivers” and “competitive drivers” (Vanderhaegen,


Chalmé, Anceaux, & Millot, 2006), might be useful in clarifying how driving behaviour in
oversaturated traffic conditions could be related with traffic control strategies. The degree
in which “cooperative drivers” or “competitive drivers” might affect such strategies in road
networks, could be an initial step to explore whether or not driving behaviour should be
included as an important factor in reducing traffic congestions at intersections.
4. Concluding remarks

This paper has presented a methodological review of existent methods for measuring and
modelling travel times from the traffic engineering viewpoint. It was illustrated that these
methods are somehow limited in terms of the requirements of data collection procedures.
Multi-agent systems, as a computational method, can overcome many of these limitations,
although its use in modelling traffic control strategies at large scales of road networks is
also limited from the computational viewpoint. An important point is whether driving
behaviour can play a key role in affecting the efficiency of these strategies. The motorcycle
lane-sharing, as a punctual case of heterogeneous traffic, was presented to suggest the need
to explore the relevance of drivers’ performances in the efficiency of traffic control
strategies, since the relationship between driving behaviour and traffic control has not been
directly studied so far.
REFERENCES

Bakker, B., Whiteson, S., Kester, L. J. H. M., & Groen, F. C. A. (2010). Traffic Light
Control by Multiagent Reinforcement Learning Systems. Traffic. Retrieved from
http://staff.science.uva.nl/~whiteson/pubs/bakkericis10.pdf

Bazzan, A. L. C. (2007). Traffic as a complex system : Four challenges for computer


science and engineering. Anais do XXXIV Seminario Integrado de Software e
Hardware (SEMISH) (pp. 2128-2142). SBC.

Bazzan, A. L. C. (2009). Opportunities for multiagent systems and multiagent


reinforcement learning in traffic control. Autonomous Agents and Multi-Agent
Systems, 18(3), 342-375.

Bhaskar, A., Chung, E., & Dumont, A. G. (2009). Integrating cumulative plots and probe
vehicle for travel time estimation on signalized urban network. 9th swiss Transport
Research Conference. Monte Varità, Switzerland.

Bhaskar, A., Chung, E., & Dumont, A.-G. (2011). Fusing loop detector and probe vehicle
data to estimate travel time statistics on signalized urban networks. Computer-Aided
Civil and Infrastructure Engineering, 26(6), 433-450.

Bierlaire, M., & Crittin, F. (2004). An efficient algorithm for real-time estimation and
prediction of dynamic OD tables. Operations Research, 52(1), 116-127.

Brackstone, M., & McDonald, M. (1999). Car-following: a historical review.


Transportation Research Part F, 2(4), 181-196.

Cherri, C., Nodari, E., Toffetti, A., & Information Society Tecnologies [IST]. (2004).
Review of existing tools and methods ( No. IST-1-507674-IP) (p. 109). Retrieved from
http://www.aide-eu.org/pdf/sp2_deliv_new/aide_d2_1_1.pdf

Daganzo, C. F. (1994). The cell transmission model: A dynamic representation of highway


traffic consistent with the hydrodynamic theory. Transportation Research Part B,
28(4), 269-287.

Daganzo, C. F. (1995). The cell transmission model. Part II: Network traffic.
Transportation Research Part B, 29(2), 79-93.

Davidson, P. J., Hautamaki, J., Collin, J., & Takala, J. (2009). Improved vehicle positioning
in urban environment through integration of GPS and low-cost inertial sensors.
Proceedings: ENCGNSS. Naples, Italy.
Dion, F., Rahka, H., & Kang, Y. S. (2004). Comparison of delay estimates at
undersaturated and over-saturated pre-timed signalized intersections. Transportation
Research Part B, 38, 99-122.

Dowling, R., Skabardonis, A., & Alexiadis, V. (2004). Traffic Analysis Toolbox Volume
III: Guidelines for Applying Traffic Microsimulation Software ( No. Publication
Number: FHWA-HRT-04-040) (p. 146). McLean, VA, USA.

Friedrich, M., Jehlicka, P., & Schlaich, J. (2008). Automatic number plate recognition for
the observance of travel behaviour. 8th International Conference on Survey Methods
in Transport: Harmonisation and Data Comparability (pp. 1-17). Annecy, France:
International Steering Committee for Travel Survey Conferences. Retrieved from
http://isctsc.let.fr/papiers/workshop final version/21 B2 Friedrich et al.pdf

Hidas, P. (2002). Modelling lane changing and merging in microscopic traffic simulation.
Transportation Research Part C: Emerging Technologies, 10(5-6), 351-371.
doi:10.1016/S0968-090X(02)00026-8

Kesting, A., Treiber, M., & Helbing, D. (2007). General Lane-Changing Model MOBIL for
Car-Following Models. Transportation Research Record, 1999(1), 86-94.
doi:10.3141/1999-10

Liu, H. (2008). Travel time prediction for urban networks. Delft University of Technology.
PhD thesis.

Noland, R. B., & Polak, J. W. (2002). Travel time variability: a review of theoretical and
empirical issues. Transport Reviews, 22(1), 39-54.

Philip, P., Taillard, J., Klein, E., Sagaspe, P., Charles, A., Davies, W. L., Guilleminault, C.,
et al. (2003). Effect of fatigue on performance measured by a driving simulator in
automobile drivers. Journal of Psychosomatic Research, 55(3), 197-200.
doi:10.1016/S0022-3999(02)00496-8

Robinson, S., & Polak, J. W. (2005). Modelling urban link travel time with inductive loop
detector data by using the k-NN method. Transportation Research Record: Journal of
the Transportation Research Board, 1935, 47-56.

Sisiopiku, V., Rouphail, N., & Santiago, A. (1994). Analysis of correlation between arterial
travel time and detector data from simulation and field studies. Transportation
Research Record: Journal of the Transportation Research Board, 1457, 166-173.

Sperley, M., & Joy Pietz, A. (2010). Motorcycle lane-sharing (Vol. OR-RD-10-2, p. 27).
Oregon Department of Transportation. Oregon, USA. Retrieved from
http://ntl.bts.gov/lib/33000/33000/33048/Motorcycle_Lane-Sharing.pdf
Srinivasan, K. K., & Jovanis, P. P. (1996). Determination of number of probe vehicles
required for reliable travel time measurement in urban network. Transportation
Research Record: Journal of the Transportation Research Board, 1537(1), 15-22.

Szeto, W. Y., Ghosh, B., Basu, B., & O’Mahony, M. (2009). Multivariate traffic
forecasting technique using cell transmission model and SARIMA model. Journal of
Transportation Engineering, 135(9), 658-667.

Takaba, S., Morita, T., Hada, T., Usami, T., & Yamaguchi, M. (1991). Estimation and
measurement of travel time by vehicle detectors and license plate readers. Vehicle
Navigation and Information Systems Conference (pp. 257-267). Society of
Automotive Engineers.

Vanderhaegen, F., Chalmé, S., Anceaux, F., & Millot, P. (2006). Principles of cooperation
and competition: application to car driver behavior analysis. Cognition, Technology &
Work, 8(3), 183-192. doi:10.1007/s10111-006-0037-9

Viti, F. (2006). The dynamics and the uncertainty of delays at signals. Delft University of
Technology. PhD thesis.

Wigan, M. (2002). Motorcycles as full mode of transportation. Transportation Research


Record: Journal of the Transportation Research Board, (1818), 39-46.

Wooldridge, M. (2002). An introduction to multiagent systems. West Sussex, UK: John


Wiley & Sons.

Wunnava, S. V., Yen, K., Babij, T., Zavaleta, R., Romero, R., & Archilla, C. (2007). Travel
time estimation using cell phones for highways and roadways. Florida, USA.

Yegor, M., Wu, Y., Wang, Y., & Lee, U. K. (2010). Field experiments on bluetooth-based
travel time data collection. 89th TRB annual meeting. Washington, DC; USA.

You, J., & Kim, T. J. (2000). Development and evaluation of a hybrid travel time
forecasting models. Transport Research Part C, 8(1-6), 231-256.

Zheng, F. (2011). Modelling Urban Travel Times. Delft University of Technology. PhD
thesis.

Zheng, P., & McDonald, M. (2009). Estimation of travel time using fuzzy clustering
method. Institution of Engineering and Technology Intelligent Transport Systems,
3(1), 77-86.

van Zuylen, H. J., & Hoogendoorn, S. P. (2007). A probabilistic calculation of queue


lengths at signalized intersections. ITS World Congress. Beijing, China.

Вам также может понравиться