Elliptical Head Tracking Using Intensity Gradients

The old version of the elliptical head tracker uses only one vision module. When a new image becomes available, it finds the best position and size for the ellipse by simply maximizing the sum of the gradient magnitude around the perimeter of the ellipse. Despite the local search, velocity prediction removes any restriction on maximum lateral image velocity. In real time, the tracker is able to reliably and automatically control the camera's pan and tilt (but not zoom) in order to keep the subject centered in the field of view at a desired size. Its main difficulty is in handling the ambiguity of textured scenes (although it handles the moderately textured scene shown below).

Below are two image sequences showing the head tracker's performance in two different environments. Click on any of the images to download its corresponding MPEG file.

Sequence 1
(untextured room)
Sequence 2
(textured room)
Intensity
0.9 MB

1.2 MB
Gradient magnitude
1.9 MB

2.8 MB

There is also a short MPEG clip (166 KB) demonstrating the tracker's speed.

Source code for SGI Indy computer (gzipped tar file)