HoloNav - Algorithmic Patient Alignment

Algorithmic approaches for performing patient alignment when using the HoloLens as a surgical navigation device.

Part of the 2021-2022 research project at TU Delft.

---

Algorithms

The repository contains different algorithms for performing rought point cloud alignment:

• Fast point feature histograms
• Principal component analysis
• Manual point selection

After the rough point cloud registration is performed using one of these algorithms, the registration is refined using the Iterative Closest Point algorithm.

Repository structure

Each algorithm is contained in its own package. Every package contains a method which performs rough and precise point cloud registration on the provided point clouds. This method provides an easy to use endpoint for point cloud registration.

• The pca_icp_alignment method performs PCA rough registration followed by ICP precise registration
• The fpfh_icp_alignment method performs FPFH rough registration followed by ICP precise registration
• The mps_icp_alignment method performs manual point selection rough registration followed by ICP registration

To use a given algorithm, just import the algorithm from the package. For example:

from pca.pca_icp_alignment import pca_icp_alignment

Input

Every algorithm needs at least the following input parameters:

• source: The source point cloud (the point cloud that should be transformed)
• target_depth_sensor: The depth-sensor version of the target point cloud (the point cloud that the source should be transformed to)
• target_pointer: The pointer version of the target point cloud (the point cloud that the source should be transformed to)

Any additional supported parameters are present in the documentation.