FreeTrack Forum

Hello dear developers!

Ever considered using Mean Shift Algorithm to track the head motion like in this video http://www.youtube.com/watch?v=36xMdVdEXDE?

For example you could wear a hat or cap and colour the areas with a distinct pattern attach small objects like small balls, where the LEDs would normally be, and the mean shift algorithm would automatically track these objects.

It is a very robust method and invariant concerning rotation and translation, and ideal for (Free)tracking :).

http://www.cs.toronto.edu/~dross/ivt/

This page and its links explain it better than I can do in this post.

Take care! You have done a tremendous amount of work for the simming community. Thank you for that (I still have to buy and assemble my 3point LED thingie ^^)

You don't have to wear LEDs, as mentioned in the FAQ you can use retroreflective material to represent the tracking points, but they need to be illuminated by LEDs. You need to have some form of active light for webcam headtracking to work at 30fps because at that frame rate the exposure time is short and since most cheap webcams have tiny aperatures, the amount of light that hits the sensor in normal room lighting is not normally enough to create a picture good enough for image recognition. If the webcam were modded (IR filter removed) the increased light sensitivity may allow for some form of image recognition.

However image recognition is typically less robust, less accurate (as seen in the video you linked) and consumes considerable cpu. The limitations of current image recognition based motion tracking are partly why TrackIR, the Wii Remote, Sony PS3 SIXAXIS and many other consumer level motion tracking systems use LEDs/reflectors.

Ok thanks for clearing that up!

Totally forgot about the CPU usage :).

Well, maybe it is doable in the future...

Despite the shortcomings of image recognition based motion tracking, the advantage of simpler hardware is significant... it's definitely worthy of consideration.