Pyramidal Lucas-Kanade Feature Tracker

tool_image


Description

This tool lets you compute optical flow based on the Lucas-Kanade feature tracker in real time. The optical flow computation is implemented in pyramidal fashion, from coarse to fine resolution. Consequently, the algorithm can handle large pixel flows, while keeping a relatively small window of integration, and therefore achieve high local accuracy (of the order of 0.1 pixel). The tracking is done by
cvCalcOpticalFlowPyrLK function that implements algorithm, described in detail in the report algo_tracking.pdf.

LKDemo.exe

A simple demo of Lucas-Kanade in pyramid (LKpyr) was created as shown below: This demo assumes you have USB and a USB camera installed with directShow. If you are running on Microsoft* Windows* NT 4.0, this demo will only work if you have written your own video capture source filter. Microsoft Windows 2000 is best for this and all the rest of the demos.

Controls

[1] Start/Stop video capture toggle.
[2] Find a set of trackable points (labeled by green dots) and track them.
[3] Turn off the video image and just show the tracked green points from [2].
[4] Change the size of the video display window.
[5] Adjust USB camera parameters.

Limitations

This is just a simple, "hard coded" demo. We don't allow you to change the pyramid depth nor allow you to adjust the tracking parameters nor the thresholds for acceptable points (minimum eigen value size). We do not monitor the quality of the tracked points in this demo, thus when a point is occluded, it often sticks near or on the occluding boundary. We will add user parameter adjustments and point quality monitoring next rev. -- ran out of time for this demo, so sorry.