Traclus_dl is a program inspired by Traclus (Trajectory Clustering: A partition and group framework, published by Lee et al in 2007 in ACM SIGMOD) designed for desire lines consisting of origin-destination pairs in two dimensions.
traclus_dl/
├── data/ # Input data files
├── src/ # Source code
├── test/ # Test scripts
├── requirements.txt # Dependencies
├── README.md # This file
└── run.py # Main entry point
The input is a file with each line consisting of one desire line, space or tab separated. Each line should have the following format:
line_id weight start_x start_y end_x end_y
Where:
line_id: Identifier for the desire line.weight: Numeric value indicating the weight of the desire line.start_x, start_y: Starting coordinates of the desire line.end_x, end_y: Ending coordinates of the desire line.
Example input file (data/simple_test_data.txt):
east1 10.0 0.0 0.0 100.0 0.0
east2 10.0 0.0 1.0 100.0 1.0
east3 10.0 0.0 2.0 100.0 2.0
The algorithm produces two output files in the same directory as the input file:
<filename>.<max_dist>.<min_density>.<max_angle>.<segment_size>.segmentlist.txt: Contains information about the trajectory segments and their corridor assignments.<filename>.<max_dist>.<min_density>.<max_angle>.<segment_size>.corridorlist.txt: Contains the identified corridors.
These files can be readily visualized using the included visualization tool or imported into QGIS.
The software is written in Python and can be run from the command line where Python is installed.
# Create a virtual environment
python -m venv venv
# Activate the virtual environment (Windows/PowerShell)
.\venv\Scripts\Activate.ps1
# Install requirements
pip install -r requirements.txt
Use the following command:
python run.py --infile <filename> --max_dist <max_dist> --min_density <min_density> --max_angle <max_angle> --segment_size <segment_size>
Where:
filename: Path to the input filemax_dist: Maximum distance parameter for the DBScan algorithmmin_density: Minimum density parameter for the DBScan algorithmmax_angle: Maximum angle parameter for the DBScan algorithmsegment_size: Segment size parameter for the DBScan algorithm
Example usage:
python run.py --infile data/simple_test_data.txt --max_dist 5 --min_density 3 --max_angle 5 --segment_size 20
To run a basic test with the simple test data:
python test/test_algorithm.py
To visualize the results after running the algorithm:
# First run the algorithm to generate output files
python run.py --infile data/simple_test_data.txt --max_dist 5 --min_density 3 --max_angle 5 --segment_size 20
# Then visualize the results
python test/visualize.py --input data/simple_test_data.txt
The visualization tool will automatically search for the output files in the current directory, the data directory, and the directory containing the input file. The visualization will be saved as clustering_result.png by default.
Traclus_dl implements a modified version of the Traclus algorithm specifically designed for origin-destination pairs. The algorithm:
- Segments each trajectory into smaller pieces
- Clusters similar segments based on angle, distance, and density constraints
- Uses a priority queue to identify corridors in order of their weights
- Outputs corridors as weighted averaged line segments for visualization
- Python 3.6+
- NumPy
- Matplotlib
See requirements.txt for specific versions.
If you use this implementation in your research, please cite:
Bahbouh, K., Wagner, J. R., Morency, C., & Berdier, C. (2015). TraClus-DL: A Desire Line Clustering Framework to Identify Demand Corridors. Transportation Research Board 94th Annual Meeting. Washington DC, United States. Paper No. 15-3508.