This document describes a study that estimates traffic flow parameters in Nairobi, Kenya using available road quality and camera speed data. The study develops a probabilistic model to account for noisy data and uncertainty. The model relates unobserved true traffic speeds over time to observed road quality conditions and camera-measured speeds. The model is evaluated on new data and achieves an average error of 3.63 kph in predicting speeds. The approach provides a way to estimate traffic conditions in developing cities that lack comprehensive traffic data.
The way built environment factors influence the success of BRT systems holds lessons for planners and policymakers. Read more: http://bit.ly/1mf3hUV
Transforming Transportation 2015: Smart Cities for Shared Prosperity is the annual conference co-organized by the World Resources Institute and the World Bank.
Presentation TRB Article:
Guarda, P., Velásquez J. M., Tun H. T., Chen, X. and Zhong, G. Comparing Chinese and non-Chinese Bus Rapid Transit: Evidence from evaluation of global BRT based on BRT design indicators. Transportation Research Board 96th Annual Meeting, January 8-12, 2017, Washington D.C, United States of America [link]
Sustainable Urban Transport Planning using Big Data from Mobile PhonesDaniel Emaasit
In the past decades, there has been rapid urbanization as more and more people migrate into cities. The World Health Organization (WHO) estimates that by 2017, a majority of people will be living in urban areas. By 2030, 5 billion people—60 percent of the world’s population—will live in cities, compared with 3.6 billion in 2013. Developing nations must cope with this rapid urbanization. Transportation and urban planners must estimate travel demand for transportation facilities and use this to plan transportation infrastructure. Presently, the technique used for transportation planning uses data inputs from local and national household travel surveys. However, these surveys are expensive to conduct, cover smaller areas of cities and the time between surveys range from 5 to 10 years. This calls for new and innovative ways for Transportation Planning using new data sources.
In recent years, we have witnessed the proliferation of ubiquitous mobile computing devices in developing countries. These mobile phones capture the movement of vehicles and people in near real time and generate massive amounts of new data. My PhD research investigates how we can utilize anonymized mobile phone data ( i.e. Call Detail Records) and probabilistic machine learning to infer travel/mobility patterns. One of the objectives of this research is to demonstrate that these new “big” data sources are cheaper alternatives for transport modeling and travel behavior studies.
The way built environment factors influence the success of BRT systems holds lessons for planners and policymakers. Read more: http://bit.ly/1mf3hUV
Transforming Transportation 2015: Smart Cities for Shared Prosperity is the annual conference co-organized by the World Resources Institute and the World Bank.
Presentation TRB Article:
Guarda, P., Velásquez J. M., Tun H. T., Chen, X. and Zhong, G. Comparing Chinese and non-Chinese Bus Rapid Transit: Evidence from evaluation of global BRT based on BRT design indicators. Transportation Research Board 96th Annual Meeting, January 8-12, 2017, Washington D.C, United States of America [link]
Sustainable Urban Transport Planning using Big Data from Mobile PhonesDaniel Emaasit
In the past decades, there has been rapid urbanization as more and more people migrate into cities. The World Health Organization (WHO) estimates that by 2017, a majority of people will be living in urban areas. By 2030, 5 billion people—60 percent of the world’s population—will live in cities, compared with 3.6 billion in 2013. Developing nations must cope with this rapid urbanization. Transportation and urban planners must estimate travel demand for transportation facilities and use this to plan transportation infrastructure. Presently, the technique used for transportation planning uses data inputs from local and national household travel surveys. However, these surveys are expensive to conduct, cover smaller areas of cities and the time between surveys range from 5 to 10 years. This calls for new and innovative ways for Transportation Planning using new data sources.
In recent years, we have witnessed the proliferation of ubiquitous mobile computing devices in developing countries. These mobile phones capture the movement of vehicles and people in near real time and generate massive amounts of new data. My PhD research investigates how we can utilize anonymized mobile phone data ( i.e. Call Detail Records) and probabilistic machine learning to infer travel/mobility patterns. One of the objectives of this research is to demonstrate that these new “big” data sources are cheaper alternatives for transport modeling and travel behavior studies.
Tutorial on AI-based Analytics in Traffic ManagementBiplav Srivastava
This is the tutorial on AI analytical techniques for traffic management presented at the IJCAI 2013 conference, Beijing, China presented by Biplav Srivastava and Akshat Kumar.
This is the tutorial on AI techniques for traffic management presented at the AAAI 2012 conference, Toronto, Canada presented by Biplav Srivastava and Anand Ranganathan.
Application of Fuzzy Logic in Transport Planningijsc
Fuzzy logic is shown to be a very promising mathematical approach for modelling traffic and transportation processes characterized by subjectivity, ambiguity, uncertainty and imprecision. The basic premises of fuzzy logic systems are presented as well as a detailed analysis of fuzzy logic systems developed to solve various traffic and transportation planning problems. Emphasis is put on the importance of fuzzy logic systems as universal approximators in solving traffic and transportation problems. This paper presents an analysis of the results achieved using fuzzy logic to model complex traffic and transportation processes.
APPLICATION OF FUZZY LOGIC IN TRANSPORT PLANNINGijsc
Fuzzy logic is shown to be a very promising mathematical approach for modelling traffic and
transportation processes characterized by subjectivity, ambiguity, uncertainty and imprecision. The basic
premises of fuzzy logic systems are presented as well as a detailed analysis of fuzzy logic systems developed
to solve various traffic and transportation planning problems. Emphasis is put on the importance of fuzzy
logic systems as universal approximators in solving traffic and transportation problems. This paper
presents an analysis of the results achieved using fuzzy logic to model complex traffic and transportation
processes.
CycleStreets: Our Story - presentation to Net2Camb eventCycleStreets
Here is our presentation at the Net2Camb event.
See:
http://www.cyclestreets.net/blog/2010/12/29/net2camb-meetup-building-cyclestreets/
http://net2camb.org/2011/01/january-net2camb-meetup-building-cyclestreets/
SUSTAINABLE TRANSPORATION PLANNING – A SYSTEMS APPROACHIAEME Publication
Chennai is the fourth largest metropolitan city of India which covers an area of 426 sq.km and recorded a population of 46.81 lakhs in 2011. The Chennai Metropolitan Area which extends over an area of 1189 sq.km recorded the population of 86.96 lakhs in 2011 and the density is 11,000 per sq.km. The population of Chennai in 1639 was 40,000 and today the city is estimated to have a population of 7.5 million, which gives a population density of about 6482 per sq. km. This rapid
increase in population leads to traffic congestion and imbalanced supply and demand of transport facilities. Thus it is important to develop a dynamic model which would exhibit the invention of various transportation facilities in Chennai and to estimate the travel demand for both present and future situation.
New Research Articles 2020 November Issue International Journal of Software E...ijseajournal
The International Journal of Software Engineering & Applications (IJSEA) is a bi-monthly open access peer-reviewed journal that publishes articles which contribute new results in all areas of the Software Engineering & Applications. The goal of this journal is to bring together researchers and practitioners from academia and industry to focus on understanding Modern software engineering concepts & establishing new collaborations in these areas.
VEHICULAR 2020 Presentation by Kohei HosonoKohei Hosono
Title:
Implementation and Evaluation of Priority Processing by Controlling Transmission Interval Considering Traffic Environment in a Dynamic Map
Author:
Kohei Hosono, Akihiko Maki, Yoichi Watanabe, Hiroaki Takada, Kenya Sato
Affiliation:
Computer and Information Science, Graduate School of Science and Engineering, Doshisha University
Fujitsu Limited
Institutes of Innovation for Future Society, Nagoya University
Mobility Research Center, Doshisha University
Conference:
The Ninth International Conference on Advances in Vehicular Systems, Technologies and Applications VEHICULAR 2020
Abstract:
Much attention has been attracted to the research of cooperative automatic driving that focuses on safety and efficiency by sharing the data obtained from sensor information of a vehicle. In addition, dynamic maps, a common information and communication platform for the integrated management of shared sensor information, are under consideration. A vehicle always sends data to a server that manages the dynamic map, and the server runs applications for driving support and control on the basis of the data, so fast information processing is required. However, congestion is a concern when data is continuously sent from vehicles to the server at high transmission intervals and when many vehicles are managed by dynamic maps on the server. In addition, the data transmission interval from the vehicle required by the road characteristics differs in actual traffic environments. Therefore, congestion can be alleviated by adjusting the transmission interval of data from the vehicle in consideration of road characteristics. In this paper, a platform for a dynamic map consisting of a server and a vehicle is constructed. We have also implemented a priority processing function that sets the priority for each section of a lane, and adjusts the transmission interval on the basis of the characteristics of the road around the vehicle.
Tutorial on AI-based Analytics in Traffic ManagementBiplav Srivastava
This is the tutorial on AI analytical techniques for traffic management presented at the IJCAI 2013 conference, Beijing, China presented by Biplav Srivastava and Akshat Kumar.
This is the tutorial on AI techniques for traffic management presented at the AAAI 2012 conference, Toronto, Canada presented by Biplav Srivastava and Anand Ranganathan.
Application of Fuzzy Logic in Transport Planningijsc
Fuzzy logic is shown to be a very promising mathematical approach for modelling traffic and transportation processes characterized by subjectivity, ambiguity, uncertainty and imprecision. The basic premises of fuzzy logic systems are presented as well as a detailed analysis of fuzzy logic systems developed to solve various traffic and transportation planning problems. Emphasis is put on the importance of fuzzy logic systems as universal approximators in solving traffic and transportation problems. This paper presents an analysis of the results achieved using fuzzy logic to model complex traffic and transportation processes.
APPLICATION OF FUZZY LOGIC IN TRANSPORT PLANNINGijsc
Fuzzy logic is shown to be a very promising mathematical approach for modelling traffic and
transportation processes characterized by subjectivity, ambiguity, uncertainty and imprecision. The basic
premises of fuzzy logic systems are presented as well as a detailed analysis of fuzzy logic systems developed
to solve various traffic and transportation planning problems. Emphasis is put on the importance of fuzzy
logic systems as universal approximators in solving traffic and transportation problems. This paper
presents an analysis of the results achieved using fuzzy logic to model complex traffic and transportation
processes.
CycleStreets: Our Story - presentation to Net2Camb eventCycleStreets
Here is our presentation at the Net2Camb event.
See:
http://www.cyclestreets.net/blog/2010/12/29/net2camb-meetup-building-cyclestreets/
http://net2camb.org/2011/01/january-net2camb-meetup-building-cyclestreets/
SUSTAINABLE TRANSPORATION PLANNING – A SYSTEMS APPROACHIAEME Publication
Chennai is the fourth largest metropolitan city of India which covers an area of 426 sq.km and recorded a population of 46.81 lakhs in 2011. The Chennai Metropolitan Area which extends over an area of 1189 sq.km recorded the population of 86.96 lakhs in 2011 and the density is 11,000 per sq.km. The population of Chennai in 1639 was 40,000 and today the city is estimated to have a population of 7.5 million, which gives a population density of about 6482 per sq. km. This rapid
increase in population leads to traffic congestion and imbalanced supply and demand of transport facilities. Thus it is important to develop a dynamic model which would exhibit the invention of various transportation facilities in Chennai and to estimate the travel demand for both present and future situation.
New Research Articles 2020 November Issue International Journal of Software E...ijseajournal
The International Journal of Software Engineering & Applications (IJSEA) is a bi-monthly open access peer-reviewed journal that publishes articles which contribute new results in all areas of the Software Engineering & Applications. The goal of this journal is to bring together researchers and practitioners from academia and industry to focus on understanding Modern software engineering concepts & establishing new collaborations in these areas.
VEHICULAR 2020 Presentation by Kohei HosonoKohei Hosono
Title:
Implementation and Evaluation of Priority Processing by Controlling Transmission Interval Considering Traffic Environment in a Dynamic Map
Author:
Kohei Hosono, Akihiko Maki, Yoichi Watanabe, Hiroaki Takada, Kenya Sato
Affiliation:
Computer and Information Science, Graduate School of Science and Engineering, Doshisha University
Fujitsu Limited
Institutes of Innovation for Future Society, Nagoya University
Mobility Research Center, Doshisha University
Conference:
The Ninth International Conference on Advances in Vehicular Systems, Technologies and Applications VEHICULAR 2020
Abstract:
Much attention has been attracted to the research of cooperative automatic driving that focuses on safety and efficiency by sharing the data obtained from sensor information of a vehicle. In addition, dynamic maps, a common information and communication platform for the integrated management of shared sensor information, are under consideration. A vehicle always sends data to a server that manages the dynamic map, and the server runs applications for driving support and control on the basis of the data, so fast information processing is required. However, congestion is a concern when data is continuously sent from vehicles to the server at high transmission intervals and when many vehicles are managed by dynamic maps on the server. In addition, the data transmission interval from the vehicle required by the road characteristics differs in actual traffic environments. Therefore, congestion can be alleviated by adjusting the transmission interval of data from the vehicle in consideration of road characteristics. In this paper, a platform for a dynamic map consisting of a server and a vehicle is constructed. We have also implemented a priority processing function that sets the priority for each section of a lane, and adjusts the transmission interval on the basis of the characteristics of the road around the vehicle.
1. Introduction Our Approach Results Conclusions Appendix
Context-based Traffic Flow Parameter
Estimation:
A Case Study of Nairobi, Kenya
Daniel Emaasit1,2
1Research Intern
Mobility Team
IBM Research | Africa, Nairobi, Kenya
daniel.emaasit@ke.ibm.com
2PhD Student
Civil and Environmental Engineering Department
University of Nevada, Las Vegas, USA
emaasit@unlv.nevada.edu
August 25 2016
1 / 27
5. Introduction Our Approach Results Conclusions Appendix
The Research Problem (1/3)
There’s no robust method that incorporates a wide range of
contextual factors local to Nairobi to estimate traffic flow
parameters.
1. Traffic incidents (crashes, . . . )
2. Pavement conditions (potholes, . . . )
3. Weather conditions (rainy, flooding . . . )
4. Socio-political events (riots, sports, . . . )
5. Land use information (commercial, residential, . . . )
Related Work: Other studies by Box & Waterson3, Fowe et
al.4, Kwon & Murphy5 do not take into account local
contexts.
3
Method for state estimation of a road network, 2013.
4
Dynamic location referencing segment aggregation, 2015.
5
Modeling freeway traffic with coupled HMMs, 2000.
5 / 27
6. Introduction Our Approach Results Conclusions Appendix
The Research Problem (2/3)
Definition of traffic flow parameters/conditions (Speed,
Density, Flow):
Greenschield’s fundamental diagram(s) of traffic flow6
Figure 2: Flow Vs Density.
6
Foundations of Traffic Flow Theory: The Fundamental Diagrams,
Journal of the Transportation Research Board (2008).
6 / 27
7. Introduction Our Approach Results Conclusions Appendix
The Research Problem (3/3)
However, local transport authorities in Nairobi do not have
the necessary data (Typical of developing cities)
Look for alternative sources of data
(a) Speeds from "Twende,
Twende and Access Kenya".
(b) Road quality data from "IBM
StreetSense Project".
Figure 3: Available data sources.
7 / 27
8. Introduction Our Approach Results Conclusions Appendix
Research Question
How can we use the available road quality &
camera-speed data to estimate traffic flow
parameters (speed) in Nairobi?
We do not know the true traffic speeds. We only have speeds
(from cameras) with some unknown observation error.
This observation error must be accounted for when analyzing
the true speed dynamics.
8 / 27
10. Introduction Our Approach Results Conclusions Appendix
Step Process (1/2)
1. For this study, take a roadway as a case study
2. Then select a segment from that roadway
Figure 4: Roadway segment.
3. Then segment the 24 hour camera speed observations into
5-min time intervals (T = 288 time intervals)
10 / 27
11. Introduction Our Approach Results Conclusions Appendix
Step Process (2/2)
4. For each segment, develop a probabilistic model7 of the true
trajectory of traffic speeds.
Figure 5: Factor Graph of the
proposed model.
where:
θt = latent speed at timet
θ0 = initial latent speed
xt = road quality observations
yt = camera speed observations
P(θt|xt) = Prob. of latent speed
given the road quality
P(yt|θt) = Prob. of camera speed
given the latent speed
P(θt|θt−1) = Prob. of latent speed at
timet given latent speed at timet−1
7
Bishop, C. M. (2013). Model-Based Machine Learning.
11 / 27
12. Introduction Our Approach Results Conclusions Appendix
Data Preparation - speed data (1/2)
Speed Data Source: Access Kenya Data
55,511,370 speed observations, 5246 distinct road segments.
Case Study: Uhuru Highway
Figure 6: Speed distributions at study segment. 12 / 27
13. Introduction Our Approach Results Conclusions Appendix
Data Preparation - speed data (2/2)
Data was segmented into 5-min intervals/windows per day
The average speed in the interval was used as the observation
Figure 7: 5-minute time intervals
13 / 27
14. Introduction Our Approach Results Conclusions Appendix
Data Preparation - road quality
(a) Road quality categories in
overall dataset
(b) PMF for road quality at
study segment
Figure 8: Properties of road quality data.
14 / 27
15. Introduction Our Approach Results Conclusions Appendix
Learning & Model Evaluation
Condition the observations to their known quantities
Figure 9: Incorporate observed data.
Then perform exact inference8 to learn model parameters.
posterior probability distributions of latent speeds
8
Stan Development Team (2016a). The Stan C++ Library, version 2.10.0.15 / 27
17. Introduction Our Approach Results Conclusions Appendix
Discussion of Results (1/2)
Measured predictive accuracy: by predicting on new data9
Overall Average Error = 3.63 kph (Recall: T = 288
intervals/windows)
Table 1: Sample predicted mean latent speeds (kph) from new data
Predicted Observed Error
1 30.4 29.1 1.3
2 30.6 28.9 1.7
3 29.8 27.7 2.1
4 24.9 27.2 2.3
5 22.6 20.7 1.9
6 23.5 20.8 2.7
.. .... .... ....
9
Andrew Gelman (2013). Understanding predictive information criteria
for Bayesian models. arXiv:1507.04544v4 [stat.CO]
17 / 27
18. Introduction Our Approach Results Conclusions Appendix
Discussion of Results (2/2)
Probability distributions of (some) parameters
to show the uncertainities
(a) Latent speed at t = 1
(b) Latent speed at t = 259
Figure 10: Distributions of some parameters.
18 / 27
20. Introduction Our Approach Results Conclusions Appendix
Contributions
Journal papers:
1. Emaasit D., Walcott-Byrant A., Tatsubori M., Byrant R.
E., Osebe S., & Wamburu J. (2016). “Context-based Traffic
Flow Parameter Estimation: A Case Study of Nairobi,
Kenya”. Journal of Transportation Research Part B. (In
Progess)
2. Walcott-Byrant A., Byrant R. E., Tatsubori M., Emaasit
D., Osebe S., Wamburu J., & Fobi S.(2016).“The Living
Roads Project: Giving a Voice to Roads in Developing
Cities”. 96th Annual Meeting of the Transportation Research
Board(TRB). (Under Review)
Patent:
3. Emaasit D., Walcott-Byrant A., Tatsubori M., Byrant R.
E., & Amayo P. (2016). System and Method for Intelligent
Decision Making in a Threatening Event for a Self Driving
Car. (In Progress)
20 / 27
21. Introduction Our Approach Results Conclusions Appendix
Potential Benefits
Road users provided with an accurate view of the current
& future traffic flow conditions.
make travel plans in advance of commutes
motorists divert to avoid congestion, thereby reducing the
period of congestion.
Traffic operators can use this information to make traffic
control plans:
allowing traffic to be controlled to reduce the impact of
congestion,
operators can prioritize on “crutial” road segments
21 / 27
22. Introduction Our Approach Results Conclusions Appendix
Future Work
Estimate other parameters of traffic flow (Density &
Flow)
Expand to more roadways.
possibly the entire roadway network in Nairobi
Incorporate more local context factors.
to improve accuracy of traffic flow estimation
22 / 27
24. Introduction Our Approach Results Conclusions Appendix
Other Contributions
Road quality affects driver behavior10
(a) Swerving Vs speed bumps (b) Swerving Vs potholes
Figure 11: Road quality Vs driver behavior.
10
Walcott-Byrant A., Byrant R. E., Tatsubori M., Emaasit D., Osebe S.,
Wamburu J., & Fobi S.(2016).“The Living Roads Project: Giving a Voice to
Roads in Developing Cities”. 96th Annual Meeting of the Transportation
Research Board(TRB). (Under Review)
24 / 27
25. Introduction Our Approach Results Conclusions Appendix
Methodology Used: Model-Based Machine Learning
A different viewpoint for machine learning proposed by
Bishop (2013)11, Winn et al. (2015)12
Goal:
Provide a single development framework which supports the
creation of a wide range of bespoke models
The core idea:
all assumptions about the problem domain are made
explicit in the form of a model
11
Bishop, C. M. (2013). Model-Based Machine Learning. Philosophical
Transactions of the Royal Society A, 371, pp 1–17
12
Winn, J., Bishop, C. M., Diethe, T. (2015). Model-Based Machine
Learning. Microsoft Research Cambridge. http://www.mbmlbook.com.
25 / 27
26. Introduction Our Approach Results Conclusions Appendix
Misc - Notes
Why probabilistic modeling?
to account for noisy data & uncertainity
to incorporate our domain assumptions
26 / 27
27. Introduction Our Approach Results Conclusions Appendix
Model Development
The model can be described as follows:
"data {
int<lower=0> T; // the number of time steps
vector[T] y; // the observation vector for camera speeds
int<lower=0> N; // the number of categories/classes in road quality
int<lower=0,upper=2> x; // the observation vector for road quality
simplex[N] beta_x; // simplex for the prob dbn for road quality
real<lower=0,upper=80> theta_init; // Initial state speed
}
parameters {
real<lower=0> sigma_theta; // SD of state process
real<lower=0> sigma_y; // SD of observation process for camera speeds
}
transformed parameters {
vector<lower=0>[T] theta;
}
model {
theta_init ~ uniform(0, 10); // Priors
sigma_theta ~ uniform(0, 10);
sigma_y ~ uniform(0, 10);
for (n in 1:N)
x[n] ~ multinomial(beta_x); // Likelihood for road quality observation
y ~ normal(theta, sigma_y); // Likelihood for camera speed observation
theta[1] ~ normal(theta_init, sigma_theta); // State process
for (t in 2:T)
theta[t] ~ normal(theta[t-1], sigma_theta); 27 / 27