International Conference on Artificial Intelligence and Smart Vehicle(ICAISV)

24 & 25 May 2023, Amirkabir University of Technology, Tehran, lran

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Maryam Esmaeili*�Dr. Ehsan Nazerfard*�*Department of Computer Engineering, Amirkabir University of Technology, Tehran, Iran

Paper ID:1625

Presenter:Maryam Esmaeili

Maryam Esmaeili^*1

, Ehsan Nazerfard¹

¹ Department of Computer Engineering, Amirkabir University of Technology, Tehran, Iran,

Routes analysis and dependency detection based on traffic volume: a deep learning approach

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Introduction

• the cost of traffic congestion (in terms of wasted time and fuel) in the United States was $121 billion in 2011. Route prediction can be useful in a variety of situations, including traffic control, expected traffic hazards, and advertising near highways [6].

​

​

• The GPS data are represented by three-dimensional (3D) points (x, y, and t), with two spatial dimensions and one temporal dimension.

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Introduction

• route prediction can be obtained based on traffic volume for a more accurate and complete prediction.

​

• the data are registered and stored by GPS recorders to represent the current state of the urban network.

​

• Then, the dataset is converted into an accurate dataset based on the main information and standard form of geographical coordinates.

​

• the maps' matching can overlap the points outside the main route based on the noise of GPS devices and turn them into the main points of the desired route.
• .

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Introduction

• This method creates a new data source according to the standard roadmap.

​

• The distance criterion demonstrates the extent to which routes influence one another. The distance matrix can effectively specify the distance between the routes that can be connected to others

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Introduction

• The distance matrices were extensively examined in biology and anthropology, as mentioned in [14-16]. Also, there are many additional and more recent references regarding the use of distance

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Introduction

• Computer-aided techniques like Machine Learning (ML) and DL models contribute to predicting the route.

​

• .

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Related work

• Choi, Yeo, and Kim [37] suggested a deep learning approach for learning and predicting network-wide vehicle movement patterns in metropolitan networks that can replicate real-world traffic patterns.

​

• Kamble and Kounte [25] presented a method for route prediction using a clustering algorithm and their GPS sensor data.

​

• Marmasse and Schmandt [26] experimented with a Bayes classifier, histogram matching, and a hidden Markov model to match a partial route with stored routes.

​

• Krumm [27] utilized a Markov model to anticipate a driver's direction quickly.

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Related work

• Endo, Nishida, Toda, and Sawada [11] proposed a method for destination prediction by querying historical data based on RNNs. The authors avoided the data sparsity, and their proposed method could model long-term dependencies.

​

• Toqué, Côme, El Mahrsi, and Oukhellou [12] to anticipate dynamic origin-destination (OD) matrices in a subway network. As reported in the related studies, many different data exploration methods have been presented for determining movement patterns so far.

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Innovation

• This research employs DL algorithms based on traffic volume for route prediction.

​

• the traffic volume information is determined at a specific time for obtaining the two distance matrices in the present and the future.

​

• In order to compare the current and future trends, the correlation coefficient is considered.

​

• The proposed method can be on par with many state-of-art prediction models and even outperforms them in terms of accuracy and authenticity .

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Proposed Model

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Proposed Model

• GPS raw data------> GPS data after

map matching

​

Map matching

• indicating points on the main and standard routes.
• It can be used as the correct input data for analysis

section

​

​

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Proposed Model

obtaining the traffic volume as a percentage between the routes

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Proposed Model

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Proposed Model

Forecasting the next routes using deep learning

• LSTM
• CNN

​

An example of a time series of routes considered as an input and output to deep learning is:

​

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Proposed Model

• Analyzing distance matrices and determining the impact rate
• If the correlation coefficient of the two paraPmeters is zero, it means they are independent of each other. With the available information, it is impossible to comment on the increase or decrease of one parameter compared to the other.

​

• If the correlation coefficient of the two parameters is positive, increasing one parameter increases the other parameter, or decreasing one reduces the other one. In our proposed method, if the correlation coefficient of two routes is positive, it means that the increase of traffic in one route has a positive impact on the other route and can increase the traffic on that route, too.

​

​

Proposed Model

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Proposed Model

• Analyzing distance matrices and determining the impact rate
• If the correlation coefficient of two parameters is negative, increasing one parameter decreases the other and vice versa. In our proposed method, if the correlation coefficient of the two routes is negative, increasing the traffic in one route reduces the traffic in the other route and vice versa

​

Proposed Model

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Experiments and Results

Selected area for the case study of the proposed method

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

EXPERIMENTS AND RESULTS

a) Minkowski distance

b) Manhattan distance

c) Euclidean distance

Distance matrices calculated for the current state of the roadmap

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

EXPERIMENTS AND RESULTS

a) Minkowski distance

b) Manhattan distance

c) Euclidean distance

​

Distance matrices calculated for the future state of the roadmap

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Schematic view of CNN layers

Schematic view of LSTM layers

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

EXPERIMENTS AND RESULTS

a) Minkowski distance

b) Manhattan distance

c) Euclidean distance

​

Result of calculating the correlation coefficient based on distance matrices

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

LSTM model loss

LSTM model accuracy = 0.9300860583782196

Comparing CNN and LSTM models

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

C) CNN model loss

) CNN model accuracy = 0.9113683295249939

​

Comparing CNN and LSTM models

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Conclusion

an innovative method was presented in which the GPS data were used to prepare a data source to display routes, nodes, their relationship, and the volume of route traffic.

​

The data source obtained for traffic volume was converted into a distance matrix, indicating the difference between the routes and the traffic volume.

​

Using DL algorithms, the distance matrix was achieved for the future state. LSTM and CNN predicted the motion states of moving objects in the future.

​

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Both algorithms' loss and accuracy metrics were obtained with the time series along the routes.

​

Finally, the correlation coefficient test determined the distance matrix of the current and future states.

​

​

. It was found that the number of changes in the volume of one route affected the other routes.

​

Conclusion

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

Notably, increasing or decreasing the volume of moving objects in one route caused alterations in the other routes.

​

In other words, changes on one route of the roadmap were responsible for the traffic variation on other routes.

​

The vital information to reduce urban traffic congestion, time, cost, energy, etc., was achieved by analyzing the impact on present and future routes based on the traffic volume.

Conclusion

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com

References

1. Z. Yi, X. C. Liu, N. Markovic and J. Phillips, "Inferencing hourly traffic volume using data-driven machine learning and graph theory," Comput. Environ. Urban. Syst, vol. 85, pp. 101548, 2021.

2. R. Saha, M. T. Tariq and M. Hadi, "Deep Learning Approach for Predictive Analytics to Support Diversion during Freeway Incidents," Transp. Res. Rec, vol. 2674, no. 6, pp. 480-492, 2020.

3. H. Georgiou, S. Karagiorgou, Y. Kontoulis, N. Pelekis, P. Petrou, D. Scarlatti and Y. Theodoridis, "Moving Objects Analytics: Survey on Future Location & Trajectory Prediction Methods [Unpublished manuscript]," arXiv preprint arXiv:1807.04639, pp. University of Piraeus, 2018.

4. J. Miller, "Dynamically computing fastest paths for intelligent transportation systems," IEEE. Intell. Transp. Syst. Mag, vol. 1, no. 1, pp. 20-26, 2009.

This presentation uses a free template provided by FPPT.com

www.free-power-point-templates.com