3D Object Reconstruction from Imperfect Depth Data Using Extended YOLOv3 Network

. 2020 Apr 3;20(7):2025. doi: 10.3390/s20072025

Algorithm 1 Cascading flood-fill
1: procedure GET_SEED( $b o x, m a s k, ϵ$ )	▹ Seeds initial values.
2: $c x, c y \leftarrow b o x$	▹ Get center for $b o x$ .
3: $s e e d \leftarrow$ ∅
4: $s e e d \leftarrow f i n d_c l o s e s t_m a x (b o x, m a s k)$	▹ Find closest max pixel within bounds.
5: if $s e e d \neq \emptyset \land s e e d_{v a l u e} \geq ϵ$ then
6: $s e e d_{i d} \leftarrow b o x_{i d}$	▹ Set seed $i d$ to box $i d$
7: return $s e e d$	▹ Return seed if value greater than $ϵ$
8: end if
9: return ∅	▹ No valid seed was found.
10: end procedure
11: procedure FILL_NEIGHBOURS( $s e e d, θ$ )	▹ Recursively fill free neightbours with same or lower values.
12: for each $n \in s e e d_{n e i g h t b o u r s}$ do	▹ For every neighboring mask pixel.
13: if $n_{i d} = \emptyset \land n_{v a l u e} \leq s e e d_{v a l u e} \land n_{v a l u e} > θ$ then
14: $n_{i d} \leftarrow s e e d_{i d}$	▹ Set neighbor to same $i d$ as $s e e d$ .
15: $F I L L_N E I G H B O U R S (n)$	▹ Call recursively.
16: end if
17: end for
18: end procedure
19: $b o u n d i n g_b o x e s \leftarrow s o r t_c o n f i d e n c e (b o u n d i n g_b o x e s)$	▹ Sort bounding boxes by confidence.
20: for each $b o x \in b o u n d i n g_b o x e s$ do	▹ For each bounding box b
21: $s e e d \leftarrow G E T_S E E D (b o x, ϵ)$
22: if $s e e d \neq$ ∅ then
23: $F I L L_N E I G H B O U R S (s e e d, θ)$
24: end if
25: end for