I have to detect my elbow. I'm able to produce an image in which I'm facing the camera, standing before it and my arms are in my body plane(body plane is plane containing my head,chest,and legs)(i.e. to say that my arms don't stick out of this plane though they may be placed on my chest parallel to the body plane). Then using skin detection I'm getting my both arm, hand and palm.(I'm wearing half shirt so little portion of my arm is detected between elbow and shoulder). Then using convex hull defect method the point giving the largest defect should be my elbow.
This thing is fine theoretically but does not work practically. Before using the convex hull and defects method my hand and arm contour is a bit wavy (and if the waves become a bit more high then that point is detected as my elbow than the real one). Or if i use polynomial approximation method then also it is not detecting.
Skeletonization of image takes considerable time and I'm doing this real time.(My image has two contours of my both hands, that's it).
I'm doing this in opencv c++. Any Help is appreciated.