Originally posted by getnamo
View Post
I tried with small pathes initially (1.5cm squares) on my mobile phone but if I went over 30-40cm away from the leap I got very inconsistent results. The resolution is very low on the leap, well, actually the resolution coupled with the huge angle gives a too greater solid angle for each pixel so you need BIG things to track.
Here is how I used it at the time:
As I said in the note, I find the CPU use exaggerated, so I'm trying to export 3 vectors for each blob instead of one, the other 2 being coordinates on the image, in order to be able to use some condition so that when I satisfy a list of conditions of relative distances from each other. (like a triange, or a sqare with diagonals) I "lock" only on the blobs and only search in the vicinity of where they were the last frame. It's looking great, it's a lot better CPU wise, but now I'm trying to change the recursive flood with just a regular pass on a square surrounding the last position.
Also, if anyone knows some algorithms to eliminate a high-frequency shake from my measurements (but not averaging as not to introduce latency) I'm very interested. I remember reading something of the sort on an Oculus blob post, quite a while back.
edit: I initially worked on a version in LeapImageList so as to do directly the 2 images and return directly the 3d blobs, to avoid the 2 separate branches in blueprint, but while it worked I got often crashes (sometimes around the 60sec mark, sometimes later) relating to setLeapImage. Maybe I was doing something wrong but I basically kept the 2 pointers that you used and re-defined them at each call. If you want I can send the related files, but I eventually thought that maybe, someone would like the separate results by image...
Leave a comment: