I've been going through a set of slides about a modified version of the Procrustes problem. The whole problem is described by trying to find a transformation that satisfies $$A_i = sRB_i + T$$ where $R\in SO(3), T\in R^3,s\in R^+$ Meaning that R is a rotation matrix and T stands for translation. The scaling parameter $s$ is omitted. $A_i$ and $B_i$ and known point correspondences. This is formulated as the following minimization problem.

$$\min_{R,T}\sum_{i}^N \lVert A_i -RB_i + T\rVert^2$$

The slides says that after differentiating for $T$ we obtain

$$T = \frac{1}{N}\sum_{i}^NA_i-R\frac{1}{N}\sum_{i}^NB_i = \overline A-R \overline B$$ Where $ \overline A$ and $ \overline B$ are so called centroids. Two centroids are then subtracted from the problem as follows $$\min_{R}\lVert A -RB\rVert_F^2$$ where $A=(A_1-\overline A, ...,A_n-\overline A)$ and $B=(B_1-\overline B, ...,B_n-\overline B)$. I have two questions related to this:

  1. How did they obtain the derivative for T?
  2. Why is the $T$ in the original $\min_{R,T}$ with a plus sign? I would expect something like $\min_{R,T}\sum_{i}^N \lVert A_i -(RB_i + T)\rVert^2$ as in trying to minimize the differences between the projected point and its correspondence.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.