FeatureBasedTracking : Différence entre versions

Version actuelle en date du 20 juillet 2011 à 00:50

The goal of this page is to record the objectives of the development of a vision-based road user tracking tool at Ecole Polytechnique, McGill University and Carleton University.

General Objectives

1. Feature-based tracking, using any "standard" feature tracker, eg KLT
   • look at fast features and the opencv sample video_homography
2. Compensating small camera vibration (could work for small re-calibrations)
   • an idea is to use features on background objects
3. Generic features
4. Semi-automated tracking
5. Other vehicle/road user tracking methods, eg using background subtraction, tracking by detection (HoG detection)...

Summer 2011 project by Ananda Narayanan

Expected list of features for the program (following http://n.saunier.free.fr/saunier/stock/saunier06feature-based.pdf)

• feature-based tracking using the KLT implementation of OpenCV, that can be easily extended to use other types of features (eg the fast features used in the opencv sample video_homography)
  • check that new features are generated at more than the specified minimum feature distance when others are lost
  • it should include at least the following constraints on motion regularity to avoid errors and improve results
    • features have a minimum length, minimum displacement over the last n frames, the angle between two successive displacement vectors and the ratio of their norm may not be too large (if not, the feature track is disrupted: if the feature is too short, it is discarded)
    • features in the same group should have a minimum temporal overlap, there is a minimum number of features in a group of moving objects + the connection and segmentation distances
• clean and modular design that can be easily extended: speed should be reasonable
  • in particular, the steps of feature tracking and grouping can be performed separately or together
• use of a homography to transform coordinates from image to world space, accomodate if unavailable (test if it exists or use a configuration parameter)
• simple memory management using smart pointers and as much as possible const references for function arguments
  • using for example the Boost library http://www.boost.org/doc/libs/1_46_1/libs/smart_ptr/smart_ptr.htm
• clean interface using the command line and a configuration file
  • make all tracking parameters available
• saving all trajectory information for features and moving objects using the project https://bitbucket.org/trajectories/trajectorymanagementandanalysis (if possible, re-use data structure from the same project to avoid time-consuming type conversions)

If time allows:

• cross-platform compilation
• test of various thresholds and other constraints to improve tracking
• motion-compensation
• GUI for the correction of results

Resources

• Open source computer vision projects
  • OpenCV http://opencv.willowgarage.com/wiki/
  • http://code.google.com/p/opencv-feature-tracker/ (Warning: many bugs, not a good basis to build upon)
  • https://bitbucket.org/Nicolas/trafficintelligence
  • https://bitbucket.org/trajectories/trajectorymanagementandanalysis
• Version Control http://mercurial.selenic.com/
• C++
  • http://en.cppreference.com/w/cpp
  • http://www.parashift.com/c++-faq-lite/