Marker-Less 3d Object Recognition and 6d Pose Estimation for Homogeneous Textureless Objects: An RGB-D Approach

. 2020 Sep 7;20(18):5098. doi: 10.3390/s20185098

Algorithm 1 Object Localisation

1:
From top view camera
2:
$[H_{1}, E_{1}, D_{1}] \leftarrow$ transfer the image into HED space
3:
$[B B_{1}] \leftarrow$ segmentation( $H_{1}$ ) ∪ segmentation( $E_{1}$ ) ∪ segmentation( $D_{1}$ )
4:
From side view camera
5:
$[H_{2}, E_{2}, D_{2}] \leftarrow$ transfer the image into HED space
6:
$[B B_{2}] \leftarrow$ segmentation( $H_{2}$ ) ∪ segmentation( $E_{2}$ ) ∪ segmentation( $D_{2}$ )
7:
if $| B B_{1} | = | B B_{2} |$ then
8:
Use side view images / / there is no occlusion in the side view
9:
else
10:
for all $B B$ in $B B_{1}$ do
11:
$C \leftarrow c o r n e r_{B B}$
12:
$m a t c h \leftarrow f i n d_c o r r e s p o n d e n c e (C, B B_{2})$
13:
end for
14:
end if
15:
if $m a t c h =$ empty then
16:
Do nothing / / complete occlusion from the side view
17:
else
18:
Divide $B B_{2}$
19:
end if