« VideoAnnotation » : différence entre les versions

De Transport
Aller à la navigationAller à la recherche
Page créée avec « Video annotation and semi-automated tracking for performance evaluation http://stackoverflow.com/questions/10609455/image-sequence-annotation-tool »
 
 
(2 versions intermédiaires par le même utilisateur non affichées)
Ligne 1 : Ligne 1 :
Video annotation and semi-automated tracking for performance evaluation
The goal of this project is to develop a graphical tool to annotate video data, in particular road user trajectories and their characteristics (user type). It will help correct and improve the output of the Traffic Intelligence feature-based tracker.


http://stackoverflow.com/questions/10609455/image-sequence-annotation-tool
merging, splitting, moving positions and adding missing positions (extend trajectories)
 
Bitbucket repository: https://bitbucket.org/Wendlasida/trafficintelligenceannotationtool/
 
==Tests==
* a test of manual tracking on at least 5 min or 50 vehicles (without starting from the database produced by the automated tracker). The test should be done once with a homography, and a shorter one (on a few tracks (~5) without a homography.
* a test of correcting the automated tracker on a longer period (eg 30 min video, 100s of road users)
In addition to testing the software, this will give us estimates of the time required to clean the data.
* If the previous tests contain few pedestrians (<10), you should extend both so that you track at least 20 pedestrians.
 
These tests must include saving the data to SQLite, then reloading it and replaying it with the display-trajectories script and plotting them in ipython.
 
==Resources==
 
* Traffic Intelligence project
* Existing tools
** https://github.com/openvinotoolkit/cvat
** http://www.cs.columbia.edu/~vondrick/vatic/
** tool from OpenCV (not free): https://superannotate.com/download
** example of static image annotation http://www.robots.ox.ac.uk/~vgg/software/via/

Dernière version du 30 juin 2021 à 20:08

The goal of this project is to develop a graphical tool to annotate video data, in particular road user trajectories and their characteristics (user type). It will help correct and improve the output of the Traffic Intelligence feature-based tracker.

merging, splitting, moving positions and adding missing positions (extend trajectories)

Bitbucket repository: https://bitbucket.org/Wendlasida/trafficintelligenceannotationtool/

Tests

  • a test of manual tracking on at least 5 min or 50 vehicles (without starting from the database produced by the automated tracker). The test should be done once with a homography, and a shorter one (on a few tracks (~5) without a homography.
  • a test of correcting the automated tracker on a longer period (eg 30 min video, 100s of road users)

In addition to testing the software, this will give us estimates of the time required to clean the data.

  • If the previous tests contain few pedestrians (<10), you should extend both so that you track at least 20 pedestrians.

These tests must include saving the data to SQLite, then reloading it and replaying it with the display-trajectories script and plotting them in ipython.

Resources