I am at the process of making a new laser cut stereo rig for the ps 3 Eye
Although I am very pleased with the overall performance achieved by my libraries and the way things are going , due to the complexity of the project , for the foreseeable future I will “outsource” some image-processing to OpenCV mainly for LK feature tracking ( for motion tracking ) and Haar Cascade Feature Detection ( for face detection ).
From the start of the project I was trying to avoid this [ maybe its not so bad and I am just suffering from NIH syndrome ( http://en.wikipedia.org/wiki/Not_Invented_Here ) ]
But although I have some experimental feature tracking and haar cascade support that is written right in the data path of the library to avoid costly computations , it is just taking too long , in fact literally forever , for 1 person to test and debug depth mapping ,then use its results to test and debug tracking and then use the other test results to test face detection and then test and debug the whole thing.
So for now I will use OpenCV to keep some things stable in order to speed up development.
The master plan is to remove OpenCV again and have a final Vision library that will be ISO C and need no external libraries to be compiled and used ..
My Haar Cascade experimental implementation :
My Feature Tracking implementation :
OpenCV Included on this commit : ->
Well once again no time to write lots of things here , the video speaks for itself..!
Well I have made great progress in the past weeks speeding up depth mapping 10x and creating a new stereo rig with the PS3 eye that has an excelent framerate of 120fps @ 320x240 so the result is much smoother images to work with that produce a much smoother result..
Unfortunately I don`t have a lot of time to document everything , so a couple of pictures for now and the rest can be read at github ( https://github.com/AmmarkoV/RoboVision/commits/master )
I will have to keep snapshots for every feature as it is beeing tracked..