Sunday, 9 February 2014

Augmented Reality position from beacon strength

Well hopefully you all enjoyed my 51 seconds of fame yesterday.

Although in that video, I skilfully managed to make it look as if my position, relative to all those BLE beacons, was being accurately tracked, the fact is that my algorithm is pretty dumb:

It puts you in the middle of the room or "place" unless you get close to a beacon, in which case, it gently moves you in towards the 3D virtual representation of the corresponding Thing object to that beacon, as advertised by its URL. When you go out of range, it glides you back to the middle again.

It's pretty effective for a simple algorithm, I hope you'll agree, but of course it does have limitations.

The main issue is that the RSSI signal strength is very jumpy, which means you can sometimes oscillate around when at the borderline. Both for the simple algorithm and for a future trilateration one, I'd need to filter or smooth the distance calculation better.

What I'll try is a smoothing algorithm that works asymmetrically. When the signal strength goes up, it's pretty much guaranteed to be because you've moved closer, due to the laws of physics. The same laws of physics also dictate that when the signal strength goes down, it's because of some random interaction with a plane going overhead and the phase of the moon.

So I'd have it work like this: moving in immediately sets a closer distance, but moving out is treated with greater suspicion - maybe smoothing it out over three samples.

Other Jobs

I had to set up the positions of the light objects manually for the video, and that will always be the case to some extent. I imagine the new Thing would have a guess where it is when installed, then the user could nudge it into its proper place in the 3D view. The initial position could be roughly worked out at the same time as fetching the URL of the nearest place from nearby beacons, which also has to be hand-configured currently.

Another thing is that I should demo more than one place, and move between two linked rooms. Most of the code is already there for that, I think.

I also don't track up-down orientation yet, just compass rotation. Again, it's not as big a loss as you might think, but I should add that sooner or later.

No comments:

Post a Comment