This method voxelizes point clouds in .ply, passes them to a 3D Convolutional Neural Network, and determines building geometries (hopefully)
Code
(1) Import Pointcloud
(2) Voxelize
a) compute Pointcloud boundary
b) divide into cells N x N x N
- cube radius R =
c) occupancy or density
d) provide unique index label for Tsraining Data
- use voting for points with multiple classes
(3) 3D Convolutional Neural Network
a) first convolutional layer
b) first max-pooling layer
c) second convolutional layer
d) second max-pooling layer
e) dense layer
f) softmax function