Machine Vision Machine Vision

gurgleplayΤεχνίτη Νοημοσύνη και Ρομποτική

18 Οκτ 2013 (πριν από 3 χρόνια και 9 μήνες)

76 εμφανίσεις

UCLab, Kyung Hee University
Andrey Gavilov
1
Machine Vision
Machine Vision
Lecture 15
Machine Vision in manufacturing
UCLab, Kyung Hee University
Andrey Gavilov
2
Outlines
Outlines

Tasks

Why Machine Vision?

Examples

Computational requirements
and using of DSP

Inspection of surfaces
UCLab, Kyung Hee University
Andrey Gavilov
3
Automation & Machine Vision
Automation & Machine Vision
UCLab, Kyung Hee University
Andrey Gavilov
4
Robot and Vision
Robot and Vision
UCLab, Kyung Hee University
Andrey Gavilov
5
Kinds of tasks in manufacturing
Kinds of tasks in manufacturing
for computer vision
for computer vision

Inspection of product or material

Inspection of quality of surface

Inspection of structure of material (detail) (e.g., by x-
ray)

Inspection of quality of color

Non-contact measurement

Assistance in assembling

Checking of directions of details

Checking of positions of details

Checking of relations between details in assembly

In security systems

In robotics
UCLab, Kyung Hee University
Andrey Gavilov
6
An Industrial Computer Vision
An Industrial Computer Vision
System
System
UCLab, Kyung Hee University
Andrey Gavilov
7
Basic Components in
a Typical Configuration
UCLab, Kyung Hee University
Andrey Gavilov
8
Laboratory Setup
Laboratory Setup
UCLab, Kyung Hee University
Andrey Gavilov
9
Factory Setup
Factory Setup
UCLab, Kyung Hee University
Andrey Gavilov
10
Why Machine Vision?
Why Machine Vision?

Many production defects are visibly
identifiable

Manual Inspection used to be the only
option

Many limitations and so automation was
attempted before technology was ready

Over the years it has improved to become
an essential part of innumerable
manufacturing operations
UCLab, Kyung Hee University
Andrey Gavilov
11
Manual Inspection
Manual Inspection
UCLab, Kyung Hee University
Andrey Gavilov
12
Comparison with Manual Inspection
Comparison with Manual Inspection
UCLab, Kyung Hee University
Andrey Gavilov
13
Examples
Examples
(developed by Texas
(developed by Texas
Instruments)
Instruments)
UCLab, Kyung Hee University
Andrey Gavilov
14
Chip
Chip
Inspection
Inspection
UCLab, Kyung Hee University
Andrey Gavilov
15
PCB
PCB
Inspection
Inspection
UCLab, Kyung Hee University
Andrey Gavilov
16
Bottle filling
Bottle filling
Inspection
Inspection
UCLab, Kyung Hee University
Andrey Gavilov
17
Label
Label
Inspection
Inspection
UCLab, Kyung Hee University
Andrey Gavilov
18
Brake Assembly Inspection
Brake Assembly Inspection
UCLab, Kyung Hee University
Andrey Gavilov
19
Brake Assembly Inspection (2)
Brake Assembly Inspection (2)
UCLab, Kyung Hee University
Andrey Gavilov
20
Rice Sorting Machine
Rice Sorting Machine
UCLab, Kyung Hee University
Andrey Gavilov
21
Pencil Sorting Machine
Pencil Sorting Machine
UCLab, Kyung Hee University
Andrey Gavilov
22
Computational requirements
Computational requirements
and using of DSP
and using of DSP
UCLab, Kyung Hee University
Andrey Gavilov
23
Image processing
Image processing

Computationally very intensive

Typical processes are filtering and
correlations

Both use multiple and accumulate (MAC)
as the basic operations

Highly suited for DSPs

Consider an example to understand the
computational demands
UCLab, Kyung Hee University
Andrey Gavilov
24
Bolt Inspection
Bolt Inspection
UCLab, Kyung Hee University
Andrey Gavilov
25
Bolt Inspection 2
Bolt Inspection 2
UCLab, Kyung Hee University
Andrey Gavilov
26
Bolt Inspection 3
Bolt Inspection 3
UCLab, Kyung Hee University
Andrey Gavilov
27
Bolt Inspection 4
Bolt Inspection 4
UCLab, Kyung Hee University
Andrey Gavilov
28
Bolt Inspection
Bolt Inspection
UCLab, Kyung Hee University
Andrey Gavilov
29
Computation Speeds
and Production Rates
UCLab, Kyung Hee University
Andrey Gavilov
30
DSP Innovations
Leveraged for Machine Vision

Direct Memory Access

Machine Vision involves very high data

rates

Pipelining and on chip L1 and L2 Cache

SIMD / VLIW

Multicore

DAVINCI (C64x+ DSP and ARM)
UCLab, Kyung Hee University
Andrey Gavilov
31
Trends
Trends
UCLab, Kyung Hee University
Andrey Gavilov
32
Inspection of surfaces
Inspection of surfaces
UCLab, Kyung Hee University
Andrey Gavilov
33
Inspection of Complex Surfaces
Inspection of Complex Surfaces
Surface reflectance
Surface reflectance
(Albedo)
(Albedo)
Surface profile/topography
Surface profile/topography
(Bump map)
(Bump map)
Typical complex surface
Typical complex surface
(
(
Centre for Innovative Manufacturing &
Centre for Innovative Manufacturing &
Machine Vision Systems (CIMMS) UWE, Bristol
Machine Vision Systems (CIMMS) UWE, Bristol
)
)
UCLab, Kyung Hee University
Andrey Gavilov
34
Photometric Stereo
Photometric Stereo

Illuminating an object from three different
directions produces three different images...
UCLab, Kyung Hee University
Andrey Gavilov
35
Photometric Stereo
Photometric Stereo

The observed brightness at a given point is a
function of both the reflectivity and the
orientation of the surface
at that point

Three images thus give us three equations with
three unknowns

We can solve to isolate both the shape and the
reflectance
of the surface
UCLab, Kyung Hee University
Andrey Gavilov
36
Application
Application
to surface
to surface
analysis
analysis
The magnitude of
the vectors gives the
surface colouring -
‘albedo’
The directions of
the normalsform
the ‘bump map’-
topography
Image of scene as
viewed by camera -
apparent as a
matrix of grey
levels
Surface normals
recovered as a
series of vectors
with magnitude and
direction
UCLab, Kyung Hee University
Andrey Gavilov
37
Prototype tile inspection system
UCLab, Kyung Hee University
Andrey Gavilov
38
Example Results
Example Results
Albedo (2D)
Rendered
Bump Map (3D)
Raw image of surface
Possessing concomitant
2D and 3D features
UCLab, Kyung Hee University
Andrey Gavilov
39
Isolation of two
Isolation of two
-
-
and
and
three
three
-
-
dimensional data
dimensional data
UCLab, Kyung Hee University
Andrey Gavilov
40
Example of virtual rendering of
Example of virtual rendering of
the captured 3D surface data
the captured 3D surface data
Note altered surface appearance as user moves virtual
light source
UCLab, Kyung Hee University
Andrey Gavilov
41
Visual Inspection of Polished Stone
Visual Inspection of Polished Stone
(VIPS)
(VIPS)
UCLab, Kyung Hee University
Andrey Gavilov
42

Aim to develop imaging techniques
suitable for
the on-line surface inspection of polished stone
in the field.

Classify and quantify surface defects
in the
presence of acceptable natural stone features.

Allow potential for closed-loop control of the
polishing process.

Partners in Italy and Portugal.
UCLab, Kyung Hee University
Andrey Gavilov
43
Problems: -Very large images (64 Mega-pixels)
-Very large indistinct scratches (dia10 x image)
-Noisy images