Algorithm 1 Progressively detected consistent neighborhoods at multiscales.

1:Input: LiDAR point cloud P, octree $O_t$, depth of octree $d_o$, and distance threshold $\delta$ 2:Output: Consistent neighborhoods $\mathcal{N}=\{N_{\mathrm{c},1},N_{\mathrm{c},2},\cdots ,N_{\mathrm{c},\mathrm{k}}\}$ 3:Initialize: $i\leftarrow 1$, $\mathcal{N}\leftarrow ⌀$ 4: while $i\le d_o$ do 5: for each voxel $v_j$ at the ith level of $O_t$ do 6:  Select all available points in $v_j$ and calculate their centroid point $p_c$ 7:  Update current radius r with size $s_v$ of $v_j$ as $r=\sqrt{2}\times (s_v/2)$ 8:  Search r-neighborhood $N_{r,c}$ of $p_c$ from P 9:  Detect a plane model $\widehat{M}$ from $N_{r,c}$ using Equation (1) with $\delta$ 10:  Get inner point set $N_{r,c}^{\prime }$ and outer point set $N_{r,i}^{\mathrm{o}}$ of $\widehat{M}$ 11:  if $N_{r,i}^{\prime }>N_{r,i}^{\mathrm{o}}$ then 12:   Take $N_{r,i}^{\prime }$ as a consistent neighborhood and add it into $\mathcal{N}$ 13:   Let all points in $N_{r,i}^{\prime }$ be unavailable in the following detection 14:  end if 15: end for 16: $i\leftarrow i+1$ 17: end while