opencv - Measure size of box from camera image(s) -
i'm working on grasping cuboid objects (aka small cardboard boxes) placed on table robot has camera attached gripper. need exact size of box (error ~ 1mm) , position on box. have ideas hear other approaches before start implementing.
some additional info:
the boxes textured, although edges commonly white don't appear in edge image.
i change surface, objects placed on, use kind of pattern simplify segmentation process or use electro-luminous foils have perfect edges (at boundary of object).
i can actively move arm (which has (<1 mm) odometry) take 2 images different positions
my strategy far: take 2 images known positions @ least can compute height of box. find edges of box (segmentation against background) , use intersections of edges find corners of box in image. known height can compute size , position of box on table. in cases corners sufficient there 1 hard case: if i'm looking straight on box (no sides visible), can detect 4 corners in edge image. , in case can't decide if have on short high box or longer flat box.
what think approach?
or have different idea? gripper still work in progress attach new sensors.
Comments
Post a Comment