Comparison of 3D algorithms to ensure accurate 3D inspection

Photonfocus AG
Marcel Krist, CEO
Comparison of 3D Algorithms to ensure
accurate 3D inspection

Titelmasterformat durch Klicken bearbeiten
Titelmasterformat durch Klicken
What kind of applications are using the 3D laser triangulation approach?
Different applications need different methods and different 3D hardware and software
2
▪ Dark areas vs. surface defects
▪ Measurements in different
planes
▪ Quality control of complex
shapes
▪ Volumetric measurements
Source LAP Laser
Source Marexi Marine Techn.

Many 3D camera manufacturers provide a sub-pixel accuracy
Principle of 3D laser triangulation and accuracy drivers
3
Accuracy aspects/drivers
▪ CMOS Sensor (Pixel size,
resolution, Noise, etc.)
▪ Top quality lens
▪ Bandpass filter
▪ Speckle-reduced laser
▪ Ideal illumination (LED or
Laser in “right” color)
▪ No or very low vibrations on
object to be scanned
(“perfect” conveyor belts)
▪ Calibrated system
▪ Set-up of camera and laser
▪ Working distance of camera
▪ Auto exposure
▪ Perfect 3D algorithm

Pros and Cons of compact or portable 3D scanners using laser triangulation
4
Solutions for laser triangulation – compact scanners
CMOS
Detector
Laser
Diode
Lens Lens
Measurement
Range
Working
distance
(stand-off)
Generic compact 3D laser scanner
Pros
▪ Fast set-up
▪ Calibrated sensor
▪ Dedicated software
▪ Standard feature set
Cons
▪ Fixed angels, reflection and
artefacts (disruptions)
▪ Different compact scanners
needed for different
applications
▪ Space is limited for
multisensory systems
▪ Price increases due to
complexity
▪ Algorithm changes are
almost impossible and
algorithm often is unknown

Typical triangulation setup in a free configruable system
5
Camera with one Line Finder Camera with two Line FindersTwo cameras to cope with occlusions
Solutions for laser triangulation – free 3D construction
Pros of Adaptive configuration
▪ Working distance is variable
▪ Flexible angles of camera/s and laser/s
▪ Variable lens selection
▪ Multiple laser options, fast changes
▪ Algorithm can be adapted to application needs
Cons of Adaptive configuration
▪ Deep 3D knowledge required
▪ Components research is needed
▪ Calibration as ongoing task, if set-up change

Characteristics of surfaces determine the inspection quality
Characteristics of specular, Lambertian and Translucid-Lambertian surfaces
6
Specular surface (Mirror, …)
FOV
Cone of light as viewed
by the camera
Lamb. surface (Opaque plastic, …)
Arrow lengths
indicate the
power of the
reflected light
rays in every
direction.
Lamb. – Spec. surface (Metal, …)
These light rays are
also visible by the
camera!!!
Translucid-Lambertian surface

Laser stripe recognition (reflection) on different surfaces
Laser stripe on Lambertian and Translucid-Lambertian surfaces
7
Intensity profile of the stripe
Image
95% of all analysed objects!
The maximum
intensity is NOT in the
middle of the stripe
image.
Translucid-Lambertian surface
Intensity profile of the
stripe
image
The maximum
intensity is in the
middle of the stripe
image
Lambertian Surface

Line Finder algorithm versus Centre of Gravity (COG)
What kind of accuracy can be achieved with the COG method?
8
▪ COG = Centre of gravity
▪ Grey values of pixels above threshold
▪ Calculate weighted COG out of grey values of every pixel
Threshold
Pi
Le
COG = Le + ∑(Pi∙xi)/ ∑(Pi)
Left Edge
First moment of grey
values above threshold
Sum of grey values
above threshold
“Real” resolutions around 1/10
th of a pixel

Line Finder algorithm versus Centre of Gravity (COG)
What accuracy can be achieved with the Line Finder?
9
▪ Digital filtering of noise
▪ Non linear interpolation
▪ Maximum intensity value detector for non-gaussian curves
Resolutions of up to 1/64th of a pixel!!!

Development of Robust LineFinders
Robustness Analysis of LineFinders in Presence of Reflections with Uncalibrated 3D Profiles
10

Generation of Calibrated 3D Profiles with Structured-light 3D Scanners and Tactile Probes
11

Comparison of Robustness and Sub-Pixel Resolution with Calibrated 3D Profiles
12

Simulations of Sub-Pixel Resolution vs SNR – Gauss Profile with Speckle Noise
13

Simulations of Sub-Pixel Resolution vs SNR – Weibull Profile with Speckle Noise
14

Mix cameras to complete the 3D
15
This area is not
visible by camera A
Rane Map Camera B
Rane Map Camera A Rane Map Camera A and B
Solutions for laser triangulation
Camera A
Camera B

Line Finder – Surfaces of 3D objects
COG delivers 3D scans with rough surfaces – Line Finder has better accuracy
16
8x to 10x better
detection accuracy
COG Line Finder

Modes of Photonfocus 3D cameras
Photonfocus 3D camera operating modes
17
2D only
Laser line detection is turned off and the camera behaves as a normal area scan camera. This mode
serves as a preview mode in the setup and debugging phase of the vision system.
2D & 3D
Laser line detection is turned on. The sensor image (2D image) is transmitted together with the 3D
data. In the PF 3D Suite, the detected laser line is shown as a coloured line in the 2D image. This
mode serves as a preview mode in the setup and debugging phase of the triangulation system.
3D only
Laser line detection is turned on and only 3D data plus an additional image row for line scan
applications is transmitted. The scan rate of this mode is considerably faster than the 2D&3D mode.
This is the real working mode of the standard cameras.
Moving ROI
This feature enables at small ROI’s high frame rates due to the automatic tracking of the laser line.

Moving ROI – Camera based line tracking
Automatic Tracking of Laser Lines in Small ROI‘s
18

Summary
Benefits of creating your “own” 3D algorithm
19
Build your own 3D algorithm and….
▪ .....you get a flexible, robust and accuract 3D inspection tool
▪ …..you get better detection results and smaller details will be visible
▪ .....you are able to take into consideration your customer feedback and expert knowledge
▪ .....you can improve your line stripe saturation and sub-pixel resolution
▪ .....you know exactly the behaviour of your 3D algorithm and you can control it!

20
Thank you very much
for your attention
Photonfocus AG
Bahnhofplatz 10
CH-8853 Lachen SZ
Switzerland
Phone: +41 55 451 00 00
www.photonfocus.com
info@photonfocus.com
sales@photonfocus.com
Date: March 2018
Produced by: Marcel Krist, CEO

3D laser triangulation can uncover the non visible (non-descript)
Dark areas vs. surface defects
21
Dark areas and holes can be
mixed or holes can be not well
seen
In a depth map, holes can be
easily distinguished from the
base material
Pseudo-color depth image of the wood block
Grey image of a wood block

3D laser triangulation used for volume measurement and sorting
Volume measurement
22
▪ Determination of volumes
▪ Determination of
dimensions
▪ Sorting into quality classes
Source: Baixcat Vision S.L.
Source Marexi Marine Technology.

Inspection system for quality control in railway
23
Inspection system for quality control
of rails, using 2 laser lines (3 and 4)
and 2 cameras (1 and 2).
[A]: Light-section image from the top
camera [1], used to inspect the
running surface (triangulation angle of
60° to the laser line generator [3]).
[B]: Light-section image from the side-
mounted camera [2], used to inspect
the rail flank.
3D laser triangulation used for quality inspection

Line Finder vs. COG
Comparison of COG and Line Finder without noise
24
▪ Lambertian surface
COG
Line Finder
▪ Translucid-Lambertian surface
COG & Line Finder

Line Finder vs. COG
Comparison of COG and Line Finder in the presence of noise
25
▪ Lambertian surface
▪ Translucid-Lambertian surface
Line Finder
COG
COG
Line Finder

Comparison of 3D algorithms to ensure accurate 3D inspection

More Related Content

What's hot (6)

Similar to Comparison of 3D algorithms to ensure accurate 3D inspection (20)

Recently uploaded (20)

Comparison of 3D algorithms to ensure accurate 3D inspection