Related Experiment Videos
Learning Traffic as Images: A Deep Convolutional Neural Network for Large-Scale Transportation Network Speed
Xiaolei Ma1, Zhuang Dai2, Zhengbing He3
1School of Transportation Science and Engineering, Beijing Key Laboratory for Cooperative Vehicle Infrastructure System and Safety Control, Beihang University, Beijing 100191, China. xiaolei@buaa.edu.cn.
Sensors (Basel, Switzerland)
|April 11, 2017
Summary
This study introduces a convolutional neural network (CNN) method for accurate, large-scale traffic speed prediction by treating traffic data as images. The CNN approach significantly improves prediction accuracy compared to existing algorithms.
Area of Science:
- Computer Science
- Artificial Intelligence
- Transportation Engineering
Background:
- Accurate prediction of large-scale traffic speed is crucial for intelligent transportation systems.
- Existing methods struggle with the complex spatiotemporal dynamics of traffic flow.
- Developing efficient and accurate traffic prediction models remains a significant challenge.
Purpose of the Study:
- To propose a novel convolutional neural network (CNN)-based method for predicting network-wide traffic speed.
- To convert complex spatiotemporal traffic dynamics into image representations for CNN analysis.
- To evaluate the proposed method's performance against established algorithms and deep learning architectures.
Main Methods:
- Traffic data was transformed into a two-dimensional time-space matrix, representing traffic flow dynamics as images.
- A CNN was employed for feature extraction and subsequent network-wide traffic speed prediction.
- The method was validated using real-world traffic data from two major transportation networks in Beijing.
Main Results:
- The proposed CNN-based method achieved an average accuracy improvement of 42.91% over comparative algorithms.
- The method demonstrated superior performance compared to ordinary least squares, k-nearest neighbors, artificial neural network, random forest, stacked autoencoder, recurrent neural network, and long-short-term memory network.
- The CNN model training was completed within an acceptable execution time.
Conclusions:
- The CNN-based approach effectively captures spatiotemporal traffic features for accurate speed prediction.
- This method offers a significant advancement in large-scale traffic speed forecasting.
- The CNN model is suitable for practical implementation in large-scale transportation networks due to its efficiency and accuracy.