[feature/spdf] Add SpectralDecompositionFilter (SpDF)

[MathLabu]

Add a new DataPointsFilter, namely the Spectral Decomposition Filter based on the following reference:

Labussière, M., Laconte, J., & Pomerleau, F. (2020). Geometry Preserving Sampling Method Based on Spectral Decomposition for Large-Scale Environments. Frontiers in Robotics and AI, 7(September), 1–15. https://doi.org/10.3389/frobt.2020.572054

---

Spectral Decomposition Filter (SpDF) is a sampling algorithm based on spectral decomposition analysis to derive local density measures for each geometric primitive.

• First, we identify the geometric primitives along with their saliencies using the tensor voting framework.
• Then, we derive density measures from saliencies: if the density for each geometric primitive is less than the desired density, we stop; else we sub-sample each over-represented geometric primitive, and re-iterate.
• As output, we have a uniform sampled point cloud enhanced with geometric information.

I added the implementation of the Tensor Voting framework based on the closed-form solution of

Wu, T.-P., Yeung, S.-K., Jia, J., Tang, C.-K., & Medioni, G. (2012). A Closed-Form Solution to Tensor Voting: Theory and Applications. IEEE Transactions on Pattern Analysis and Machine Intelligence, 34(8), 1482–1495. https://doi.org/10.1109/TPAMI.2011.250

---

if the PR is accepted, the branches feature/tensorvoting, feature/spdf and feature/spdf-experimentalare no longer required and can be deleted.

[pomerlef]

I only passed through it rapidly. Good job for the comments and reference. Do you want to add a quick entry in DataFilters.md to make sure people finds it?

[MathLabu]

Thanks! DataPointsFilters documentation have been updated by adding entries for the Spectral Decomposition Filter (SpDF) and the Saliency Filter.