Point Cloud

A 3D Represenation . Stores a set of 3D points, potentially carrying other attributes, too. Point clouds are often the result of raw data captures.

Properties #

• Unordered set of points
• More memory efficient than dense Voxel Grid
• No spacial structure
• Cant query neighbors easily

Estimating Normals #

For any given point in the point cloud, the normal can be estimated by collecting a few (depends how many you need) point around it and use Principal Component Analysis to find out which plane they can be mapped to best, and use the planes normal as the point’s normal.

Conversions #

Can be converted to a Implicit Surface by fitting a function to be 0 at the points and positive outside and negative on the inside.

If normals are known, can be converted to Implicit Surface with Poisson Surface Reconstruction .