[ros-users] Tracking of 2D Features in Images (Dejan Pangercic)

[Marc Killpack]

Hi Dejan,
The pyramidal Lucas Kanade tracker in OpenCV (
http://opencv.willowgarage.com/documentation/cpp/video_motion_analysis_and_object_tracking.html?highlight=lk#calcOpticalFlowPyrLK)
tracks key features from frame to frame using optical flow.  It can run at
15-30 Hz depending on the size of the image.  It tracks each feature
independently but is not extremely robust to large motions or changes in
view angles between frames.  If you have a good estimate of how some of the
points are moving this can be fed into the algorithm.  Also playing around
with the pyramid terms (i.e. number of levels, size of feature window, etc)
can help but your mileage may vary.  I believe there is an lkdemo in the
opencv2 ros package.

In general this is good if you need frame rate feature tracking.  Otherwise,
from my limited experience, I think there are better methods for feature
matching but this is worth a try.

Marc Killpack
Healthcare Robotics Lab - Georgia Tech




Date: Mon, 11 Apr 2011 18:14:02 +0200
> From: Dejan Pangercic <dejan.pangercic at gmail.com>
> Subject: [ros-users] Tracking of 2D Features in Images
> To: User discussions <ros-users at code.ros.org>
> Message-ID: <BANLkTi=YrNZBYsbNA+KPeCgSLDESmiuY7A at mail.gmail.com>
> Content-Type: text/plain; charset=ISO-8859-1
>
> Dear ROS-istas,
>
> is anyone aware of an efficient package or library implementation that
> performs the tracking of 2D features in images?
> My aim is to find out which features are moving rigidly and which not
> with respect to each other.
>
> Thx for your info and best, D.
>
> --
> MSc. Dejan Pangercic
> PhD Student/Researcher
> Intelligent Autonomous Systems Group
> Technische Universit?t M?nchen
> Telephone: +49 (89) 289-26908
> E-Mail: dejan.pangercic at cs.tum.edu
> WWW: http://ias.cs.tum.edu/people/pangercic
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osuosl.org/pipermail/ros-users/attachments/20110411/7b5bf31f/attachment-0002.html>