stm-tensor-convolution 💥

This repo is in work-in-progress status.

Info

This code will be the result of the research related to training the Convolutional Neural Network (CNN) to classify the traffic states of the large traffic networks as "anomalous" or "normal" (binary classification). To represent the traffic state, a novel traffic data model called Speed Transition Matrix (STM) is used. More on STMs can be found here.

Here are the main steps of the method:

1. Preprocess GPS data (save in database as routes)
2. Create transitions from routes (needed for STM computation)
3. Compute STMs
4. Segment map using grid-based segmentation
5. For every cell in the grid extract all STMs and construct a tensor (in other words, every cell will be represented with one tensor)
6. Extract characteristic matrices for every cell using the tensor decomposition method
7. Label characteristic matrices using the center of mass for every STM (based on this article)
8. Train CNN with labeled examples
9. TODO: validation

More info

If you are interested in this topic please contact me: Leo Tisljaric

How to Cite

Plain text

L. Tišljarić, T. Carić, B. Abramović, and T. Fratrović, “Traffic State Estimation and Classification on Citywide Scale Using Speed Transition Matrices,” Sustainability, vol. 12, no. 18, p. 7278, 2020.

Bibtex

@article{tivsljaric2020traffic, title={Traffic State Estimation and Classification on Citywide Scale Using Speed Transition Matrices}, author={Ti{\v{s}}ljari{'c}, Leo and Cari{'c}, Ton{\v{c}}i and Abramovi{'c}, Borna and Fratrovi{'c}, Tomislav}, journal={Sustainability}, volume={12}, number={18}, pages={7278}, year={2020}, publisher={Multidisciplinary Digital Publishing Institute} }

Requirements

1. Install Python (3.8 recommended) Download link.
2. Install required packages from requirenments.txt using virtual environment.