This project implements Point Feature Histograms (PFH), a method for capturing local geometric features around a point in a 3D point cloud.
PFHs are designed to be pose-invariant and encapsulate surface model properties through the spatial relationships of a point and its nearest neighbors. The process involves analyzing 3D coordinates and surface normals to compute angular variations between points, offering a robust descriptor for 3D shapes. This technique is particularly effective in representing complex geometries and has been optimized for efficiency.
virtualenv pfh_env
source pfh_env/bin/activate
pip install -r requirements.txt
- For each point
$p$ , all of$p$ 's neighbors enclosed in the sphere with a given radius(self.radius
) are selected. - For every pair of point
$p_i$ and$p_j$ ($i$ $\neq$ $j$ ) in the k-neighborhood of$p$ and their estimated normals$n_i$ and$n_j$ ($p_i$ being the point with a smaller angle between its associated normal and the line connecting the points), we define a Darboux$uvn$ frame and computes the angular variation of$n_i$ and$n_j$ .
- Define Darboux
$uvn$ frame
- Computes the angular variation of
$n_i$ and$n_j$
Horse | Cat |
---|---|
![]() |
![]() |
- Paper: Fast Point Feature Histograms (FPFH) for 3D registration
- Estimating Surface Normals in a PointCloud
- Point Feature Histograms (PFH) descriptors
- Yi-Cheng Liu, Email: liuyiche@umich.edu
- Tien-Li Lin, Email: tienli@umich.edu