Upgrading a web interface to a motion interface in 5 minutes

[roboleary]

HI all,

I just posted a short video showing how to use LeapTrainer to upgrade a standard mousey-clicky flavoured web interface to a gesture-driven motion interface in ohhhh - about 5 minutes

(*5 minutes and 44 seconds to be precise - but I was talking at the same time, eh *:)

You can find it on YouTube right here.

Thanks!
R

[inblueswithu]

You are awesome

I have not yet tried Leap Trainer yet. I'll will do that shortly.

[tylerz]

Brilliant. Love it. Look forward to playing with that.

[bPatrik0521]

This looks really good! How can i implement this to a wordpress site?

[tylerz]

Hey Rob,

Per capture how many data points are you using to train against?

A while back I attempted to use Nick Gillian's Gesture Recognition Toolkit http://www.nickgillian.com/software/grt to try something similar in Objective C with my resulting project at:
https://github.com/voxq/OLKGRTHandDemo

However, I have to follow up with him as to why the results are not as expected. But I was wondering if maybe I was supplying it with way more data than it needs to catch enough pattern. I was using pretty much all the data for every finger and hand (normals, position, lengths, direction, etc.).

I suspect I may have a bug and using his library incorrectly in some regard, but I am also simply interested in what the recommended number of data points would be.

Cheers,

Tyler

[roboleary]

Hi!

Well, Wordpress will allow you to add custom Javascript to your site - so they only thing you need to do is to train some gestures and add the Javascript seen in the video to connect them to whatever functions on your site you'd like to trigger with motion.

Are there specific features you're thinking of?

Thanks!
R

[roboleary]

Hi again Tyler!

I tried various different approaches to recording - including capturing the differences between frames, just velocities, everything possible, and so on - but eventually settling on recording just the stabilized geometric positions of all visible hands and finger tips.

Of course, this data is then scaled, transformed, and resampled - which is really the step that makes all the difference.

I was using a sample rate of 100 per stroke (so each stroke was resampled to an even 100 frames - though of course, the library still only handles single stroke motions at the moment) - but after much experimentation I recently changed this to 25, because it's just giving me the best recognition rate (so far), while still minimizing false positives.

Thanks!
R

[tylerz]

Hey Rob, thanks for the great feedback and great article on http://labs.leapmotion.com.