PeTrack

For the understanding of the dynamics inside crowds reliable empirical data are needed enabling an increase of safety and comfort for pedestrians and the design of models reflecting the real dynamics. PeTrack (Pedestrian Tracking) automatically extracts accurate trajectories of marked pedestrians from video recordings. The joint trajectories of all pedestrians provide data like velocity, flow and density at any time and position. With such a tool extensive experimental series with a large number of persons can be analyzed. Individual codes enables personalized trajectories with static information of each participant (e.g. age, gender).

The program deals with wide angle lenses and a high density of pedestrians. Lens distortion and perspective view are taken into account. The procedure includes calibration, recognition, tracking and height detection. Different kinds of markers (e.g. with height information, head direction, individual code) are implemented. With a stereo camera more accurate height measurements and also markerless tracking is possible.

Installation:

We publish installers for windows. To get the latest installer, visit our registration form and download the current installer.

How to use PeTrack

To learn how to use PeTrack, please have a look at our documentation in the GitLab wiki here. We are building up another documentation page right now, that you may want to check out as well at https://petrack.readthedocs.io/. While still WIP, it already contains more detailed information on certain steps in the workflow.

Note that PeTrack can only be used if some steps are considered during the planning and execution of the experiments. This includes, but is not limited to, calibration and selecting hats. See Planning of Experiments

After all of that, enjoy successful recognition and tracking 😃

Some documentation is built into PeTrack itself. For example, you can use the help-menu to find a list of all keybindings. For running in a "batch-mode", PeTrack has a CLI, whose options are listed by calling

petrack -help

Tutorial & demo

A small demo project is provided in the demo folder. It contains all necessary files for setting up a project with PeTrack. You can download these files via this link. A tutorial leading through all steps can be found in the Wiki. The intermediate project files for each step are also included in the demo folder.

How to cite

Please cite the general paper and the corresponding software version in your publications if PeTrack helps your research.

General Paper:

• Boltes, M. and Seyfried, A.: Collecting Pedestrian Trajectories; In: Neurocomputing, Special Issue on Behaviours in Video, vol. 100, pp. 127-133 (2013)

BibTeX

@article{BOLTES2013127,
    title = {Collecting pedestrian trajectories},
    journal = {Neurocomputing},
    volume = {100},
    pages = {127-133},
    year = {2013},
    note = {Special issue: Behaviours in video},
    issn = {0925-2312},
    doi = {https://doi.org/10.1016/j.neucom.2012.01.036},
    author = {Maik Boltes and Armin Seyfried},
    keywords = {Pedestrian detection, Laboratory experiment},
}

Software Version:

You can find the DOI and citing information of you version of PeTrack on https://helmholtz.software/software/petrack. To find your corresponding version, check the about menu or look into the terminal output of PeTrack. To cite PeTrack as software without specifying a version, use the DOI 10.5281/zenodo.5078176.

License

This project is licensed under the terms of the GPLv3 license. For further information see LICENSE.

Changelog:

v0.8 : slanted view for external calibration, coordinate system and grid is possible, introduced multicolor marker with black dot, video support for various codecs
v0.7 : introduce stereo cameras; alternate height
background subtraction especially for stereo cameras; recognition of unmarked pedestrians; integration of 4 disparity algorithms
filter to swap picture; more video codecs; color blobs as marker using HSV triangle
marker with two dots added; export of viewing direction
v0.6 : bug fixes; improve speed; expand visualization possibilities; add variable tracking region adapted to head size; provide program combine to merge trajectories
v0.5 : improve robustness of tracking with interpolation and extrapolation; first analysis functionality
v0.4 : tracking import and export; manual adding and deleting of trackpoints
v0.3 : recognition; tracking; color mapping
v0.2 : add first recognition
v0.1 : only camera calibration is supported; import and export of video and image sequence

Dependencies:

• Qwt (https://qwt.sf.net)
• Qt (https://www.qt.io/)
• OpenCV (https://opencv.org/)