opencv - Real time tracking of hand -

i trying detect , track hand in real time using opencv. thought haar cascade classifiers yield fair result. after training 10k , 20k positive , negative images respectively, obtained classifier xml file. unfortunately, detects hand in positions, proving works best rigid objects. thinking of adopting algorithm can track hand, once detected through haar classifier.

my question is,if make sure haar classifier detects hand in frame, position, method yield robust tracking of hand further?

i searched web bit, , have understood can go optical flow of detected hand , or kalman filter or particle filter, have come across own disadvantages.

also, if incorporate stereo vision, me, can possibly reconstruct hand in 3d.

you concluded rightly haar features - aren't useful when comes non-rigid objects.

take @ following papers use skin colour detect hands.

1. interaction between hands , wearable cameras
2. markerless inspection of augmented reality objects

and paper uses klt features track hand after first detection:

fast 2d hand tracking flocks of features , multi-cue integration

i stereo camera not cause much, 3d reconstruction of non-rigid objects isn't straightforward , require whole lot of innovation , development. however, can take @ papers in hand pose estimation section of this page if wish pursue 3d tracking.

edit: take @ this recent paper, seems results.