How to Learn about gesture tracking: Realtime and robust hand tracking from depth (5) Hand initialization 2

Original article reading, reprinted to indicate the source:

Review:

As we mentioned in the previous section, the author performs finger detection through the defined X-Y finger and Z-finger, and then initializes hand pose. Actually, the fingers are always skewed. That is to say, it won't be the standard X-Y finger and or Z-finger. So how did the author handle it?

Finger Detection:

For finger detection, first perform X-Y finger detection. Split the file and perform the Z-finger detection. However, if the Z-finger detection again, after hitting the X-Y finger split area, stop, no longer detection.

To improve the speed, the author also said that the resolution would be half of the original.

Hand initialization:

Assume that f finger (whether X-Y finger or Z-finger) is detected, then the degree of freedom that the finger brings is 2f. With 6 degrees of freedom on the wrist, we know that the number of unknowns (that is, the number of parameters) of an hand pose is 2f + 6. (So we need at least 2f + 6 known linear independence to solve this equation)

Let's take a look at how much we know through finger detection.

If we still find f finger, we can use PCA to analyze the finger's principal component, that is, the finger's direction. This is two constraints (it should be a vector) when you find a finger amount, the actual spatial position of the fingertip is fixed. Here there are three constraints (known ). When we remove the point clouds of our fingers, we have only the palm left. At this time, we can use PCA to analyze the direction of the palm. Here are three constraints (not very clear, in the final analysis, we need to understand the meaning of the 26 degrees of freedom ). So we can say that at least I have 5f + 3 known conditions. So we have enough contact with the unknown.

So we also know the meaning of each parameter in the last equation.

T represents the position of the fingertip (a function of an unknown variable), D represents the direction, and l represents the direction of the palm. The above equation is to minimize the distance between the initial hand pose fingertip and the detected fingertip, and minimize the direction of the palm.

So I fully understood this paper.

Hope to help you.

Iker, January 1, September 15, 2014.

How to Learn about gesture tracking: Realtime and robust hand tracking from depth (5) Hand initialization 2