Comparison between linear blend skinning and dual quaternion skinning
There is a certain relationship between skeleton transformation and skin transformation when we aim to do skeleton animations, especially when using the skeleton to drive more complex objects that with skin. This process called skinning.
Basically, there are several ways to achieve that. Firstly, simple skinning: iterate each vertex on the mesh and bound each vertex to a certain bone. Secondly, linear blend skinning(LBS): each vertex is influenced by multiple bones, and each vertex has a certain influenced weight with each influencing bone. Compared to simple skinning, LBS is a classic way to do skinning, it can create natural animation. Thirdly, dual quaternion skinning(DQS): based on the same strategy as LBS, DQS converts rotation matrices to quaternions and compute normalized quaternions, using the related joints’ weight internally, rather than using the rotation matrixes directly.
Here comes a comparison between linear blend skinning and dual quaternion skinning.
- Dural quaternion skinning can avoid the loss of volume problem.
As picture1–1 shows: the left side is an implementation of LBQ, the blend shape is unnatural, the elbow has a loss of volume. The right sides shows the output of using DQS method, which can reduce the artifacts within minimal cost. Suppose we know two vertices positions: P1 and P2. If we use linear interpolation to get the new position, the new position will be lying on the segment between P1 and P2. If we use DQS, the new position will be lying on the arc circle contains P1 and P2, which will avoid mesh shrinkage.
- Dural quaternion skinning may cause unnatural protrusions on the knees and elbows when they bend.
picture1–2 shows a joint bulging problem with dural quaternion method.
Reference: http://rodolphe-vaillant.fr/?e=29
