Machine Vision Basics 05 - Image Processing

Summary

This script delves into the fundamentals of machine vision, focusing on feature detection and edge detection. It explains the importance of contrast for accurate object differentiation and the minimum 20% contrast requirement for effective feature detection. The video illustrates how machine vision cameras inspect objects on a conveyor belt and how algorithms analyze pixel contrast to identify features and edges. It also touches on sub-pixel accuracy and the practical considerations for setting up machine vision systems for reliable measurements.

Takeaways

• 🔍 Feature Detection: The script discusses feature detection in machine vision, which is based on contrast and is used for finding features on products for various inspections like presence/absence, size, flaw, strain, and color uniformity.
• 📸 Camera Setup: It describes a camera setup with backlighting, where a red square object is used to illustrate the importance of contrast in machine vision systems for accurate differentiation.
• 🔆 Contrast Importance: The necessity of good contrast for better image quality and more accurate results is highlighted, with a rule of thumb being a minimum of 20% contrast for good differentiation.
• ⚫️ Black and White Contrast: The script points out that black and white offer the most contrast, and as shades of grey are introduced, the contrast decreases.
• 📊 Contrast in Pixels: It explains how contrast sensing algorithms work at the pixel level, and how a single pixel's movement can affect the contrast spread over multiple pixels.
• 🔬 Threshold Setting: The process of setting thresholds in vision software to detect features by finding pixels within certain gray level ranges is described.
• 📐 Edge Detection: The script covers edge detection, which is used for measuring applications and is not based on gray levels, with examples of how it works on a robot inspecting wheels.
• 📈 Gray Level Profile: It explains how the gray level profile is used to identify points of sharp contrast change in edge detection and how it helps in setting thresholds for pass/fail decisions.
• 🔎 Sub-Pixel Accuracy: The discussion includes the possibility of achieving sub-pixel accuracy in machine vision through specific optics and backlighting, and the conservative value of ±0.5 pixel.
• 🛠️ Testing for Repeatability: The script emphasizes that to ensure repeatability and feasibility of a machine vision application, parts must be tested with cameras, lighting, and software to obtain consistent results.
• 📈 Edge Window Algorithm: It concludes with an explanation of the edge window algorithm that collects multiple sub-pixel samples for precise calculation of sub-pixel edges.

Q & A

What is feature detection in machine vision?

Feature detection in machine vision is a method based on contrast detection used for finding features on a product, which can be applied to tasks such as presence/absence checks, size inspection, flaw detection, strain inspection, and color uniformity checks.

How is the camera setup typically used in machine vision for inspecting objects on a conveyor belt?

The camera is often set up on one side, looking at objects as they move down a conveyor belt, with a backlight to enhance contrast and make features on the objects more distinguishable.

What is the importance of contrast in machine vision applications?

Contrast is crucial because it allows the vision system to differentiate between features on an object. Better contrast results in a clearer image and more accurate inspection outcomes.

What is the minimum contrast typically needed for good differentiation between features in a machine vision system?

A minimum of 20% contrast is generally needed for good differentiation between features in a machine vision system.

How does the contrast affect the grayscale values in an image?

High contrast results in more distinct grayscale values, with complete black being 0 and complete white being 255. As contrast decreases, the grayscale values become more similar, indicating less differentiation between features.

What is edge detection in the context of machine vision?

Edge detection is a method used for finding edges in images, which is mainly used for measuring applications and feature detection that is not gray-level based, such as dimension measurement, positioning, and orientation.

How does the edge detection algorithm work in machine vision?

The edge detection algorithm identifies points where the image contrast changes sharply, comparing the grayscale value of each pixel along the region of interest to its neighbors to determine the rate of change and the steepness of the gradient.

What is the purpose of using sub-pixel algorithms in machine vision?

Sub-pixel algorithms are used to enhance the accuracy of edge detection by examining the gray levels of adjacent pixels around the measurement endpoint and interpolating the location of the actual edge to a fraction of a pixel.

How can the accuracy of a machine vision system be influenced by the choice of optics and lighting?

The use of telecentric optics and calibrated backlight can make higher accuracy measuring possible. However, using other optics or lighting techniques can affect the system's accuracy.

What is the best way to determine the repeatability and feasibility of a machine vision application?

The best way to determine repeatability and feasibility is to set up the application with the actual parts, optics, lighting, and software, then record and analyze the results over several hundred triggers in the same static park to see if consistent results are achieved that meet the application needs.

What is the role of the region of interest (ROI) in machine vision inspection?

The region of interest (ROI) is the area within the field of view that the vision software evaluates to determine the amount of dark or bright pixels against a threshold, which can then be used to pass or fail an inspection based on predefined limits.

Outlines

00:00

🔍 Fundamentals of Machine Vision: Feature Detection and Contrast

The first paragraph introduces the basics of machine vision applications, focusing on feature detection methods. It emphasizes the importance of contrast in identifying features such as edges, flaws, and color uniformity in products. The setup of a machine vision system using a camera and backlight is described, illustrating how contrast helps in distinguishing features with a minimum of 20% difference for effective differentiation. The paragraph also explains how the camera detects features by looking for pixels with significant contrast to their background and how this affects the accuracy of the vision system. A visual representation of contrast levels and their impact on pixel values is provided, along with an example of how feature detection can be affected by the movement of a feature across pixels.

05:00

📏 Thresholding and Edge Detection in Machine Vision

The second paragraph delves into the technical aspects of thresholding and edge detection in machine vision. It explains how vision software uses threshold ranges to differentiate between dark and bright pixels, with the region of interest being evaluated for the number of dark and bright pixels. The concept of edge detection is explored, describing how the algorithm identifies points of sharp contrast change and measures the gray level of each pixel along the region of interest. The paragraph also discusses the accuracy of edge detection, mentioning that sub-pixel accuracy can be achieved under certain conditions, such as the use of telecentric optics and collimated backlight. The importance of testing and recording results for repeatability in machine vision applications is highlighted, with a brief mention of sub-pixel algorithms and their role in precise measurement.

10:01

🔬 Advanced Edge Detection Techniques and Sub-Pixel Accuracy

The third paragraph provides a deeper look into advanced edge detection techniques, specifically the sub-pixel edge detection algorithm. It describes how multiple sub-pixel samples are collected along the detected edge and how these samples are used to calculate the precise location of the edge. The paragraph explains the process of generating sub-pixel samples by examining individual lines of pixels across a detected edge and how the algorithm uses gray level information to interpolate the actual edge location. The importance of understanding the edge pixel's gray level in relation to adjacent pixels is emphasized, and a simplified example is provided to illustrate how the algorithm determines the sub-pixel location of an edge. The paragraph concludes with a general overview of the factors that influence the accuracy of edge detection, including the setup of the vision system and the software used.

Keywords

💡Feature Detection

Feature Detection is a method in machine vision that involves identifying distinct characteristics or patterns in an image. It is crucial for tasks such as presence/absence checks, size inspection, flaw detection, and color uniformity. In the video, feature detection is used to find features on a product, highlighting its importance in quality control and inspection processes.

💡Contrast Detection

Contrast Detection refers to the process of identifying areas in an image where there is a significant difference in brightness. This is essential for feature detection as it helps in distinguishing objects from their background. The video emphasizes the importance of a minimum 20% contrast for good differentiation, illustrating how contrast affects the accuracy of machine vision systems.

💡Edge Detection

Edge Detection is a technique used in machine vision to identify the boundaries of objects within an image. It is primarily used for measuring applications and is not based on grey level. The video script mentions edge detection in the context of a robot moving a camera to inspect different areas of wheels, demonstrating its application in positioning and orientation.

💡Contrast

Contrast in the context of machine vision is the difference in brightness between an object and its background. It is a critical factor in image clarity and the accuracy of machine vision systems. The video script explains that better contrast leads to better image quality and more accurate results, with examples showing how contrast affects pixel detection.

💡Threshold

Threshold in machine vision is a value set to differentiate between different levels of pixel intensity. It is used to define what constitutes a 'dark' or 'bright' pixel in an image. The video discusses setting thresholds to determine the acceptable range of pixel intensity, which is crucial for identifying defects or features in an object.

💡Region of Interest (ROI)

Region of Interest (ROI) is a specific area within an image that is selected for analysis. In the video, the ROI is highlighted as a blue square in the field of view, where the machine vision software evaluates the amount of dark and bright pixels against a set limit to determine if an inspection passes or fails.

💡Gray Level

Gray Level refers to the intensity of a pixel in an image, measured on a scale from 0 (black) to 255 (white). It is used in machine vision to analyze the contrast and edges in an image. The video script provides examples of how gray level thresholds are used to detect features and determine the success of an inspection.

💡Sub-Pixel

Sub-Pixel refers to the accuracy of measurement beyond the resolution limit of a single pixel. The video script discusses sub-pixel algorithms that can achieve accuracies of plus or minus 0.5 pixels, which is significant in high-precision applications. It explains how these algorithms examine the gray levels of adjacent pixels to interpolate the location of the actual edge.

💡Backlight

Backlight in the context of machine vision is a lighting technique that illuminates an object from behind, making it easier to detect edges and features against a dark background. The video script mentions a red square object being illuminated by backlight, demonstrating its use in enhancing contrast for feature detection.

💡Machine Vision Software

Machine Vision Software is the set of programs used to process and analyze images captured by machine vision cameras. It plays a crucial role in interpreting the data and making decisions based on the visual information. The video script describes how this software uses algorithms to detect features, set thresholds, and evaluate the results of inspections.

💡Repeatability

Repeatability in machine vision refers to the consistency of results when the same object is measured multiple times under the same conditions. The video script emphasizes the importance of testing and recording results over several hundred triggers to ensure that the application is reliable and meets the required standards.

Highlights

Feature detection in machine vision is based on contrast detection for finding features on a product, which is used in applications such as presence/absence, size inspection, flaw detection, and color uniformity inspection.

Camera setup involves backlighting to enhance contrast, which is crucial for accurate feature detection in machine vision applications.

A minimum contrast of 20% is typically needed for good differentiation between features in machine vision systems.

Contrast is measured by the difference between dark and bright areas, with higher contrast leading to better vision system performance.

Edge detection is used for measuring applications and is not based on grey level, which is essential for dimension measurement, positioning, and orientation.

Robotic integration with machine vision cameras allows for dynamic inspection of objects, such as wheels, by moving the camera around to inspect different areas.

The importance of contrast is emphasized for accurate image capture and inspection of objects in machine vision systems.

Machine vision software uses contrast sensing algorithms to detect features, even when they move slightly, by spreading the original contrast over multiple pixels.

A two by two pixel feature is guaranteed to cover at least one full pixel, ensuring repeatable detection in machine vision applications.

Thresholds in machine vision applications are set based on feedback from the software, determining the acceptable range of grey levels for features.

Edge detection algorithms identify points of sharp image contrast changes and measure the grey level of each pixel along the region of interest.

Sub-pixel algorithms provide higher accuracy in edge detection by examining the grey levels of adjacent pixels and interpolating the location of the actual edge.

The use of telecentric optics and calibrated backlight enhances the accuracy of higher precision measuring in machine vision systems.

Accuracy in machine vision is influenced by the choice of optics and lighting techniques, which can affect the system's performance.

True repeatability in machine vision applications is determined by setting up the application, testing parts, and recording consistent results over multiple trials.

Sub-pixel edge detection involves collecting multiple samples along the detected edge and using these to calculate the precise location of the edge.

Edge detection algorithms compare the value of each pixel along the edge to its neighbor to determine the gradient, indicating the steepness of the edge.

The process of finding edges in machine vision involves examining individual lines of pixels across a detected edge to generate sub-pixel samples.

Transcripts

these are kind of getting into

application-based

what uh different types of basic methods

of machine vision so one is feature

detection this methods based on contrast

detection and use for finding features

on a product the example

presence/absence in size inspection flaw

strain inspection color uniformity etc

and you can tell to the right how the

camera is set up on the left hand side

looking at objects going down a conveyor

and you can see that there is a actually

a backlight on this as well the red red

object rick sweat red square object is

actually is it up supposed to be a

backlight also edge detection this is

mainly used for finding edge edges in

measuring applications and feature

detection that are usually not grey

level based so dimension measurement on

positioning and orientation you can time

to tell that a robot there's supposed to

there's a machine vision camera on the

min of a robot and the robot is moving

the camera around to inspect looks like

different areas of wheels so when we

detect pixels of a feature these pixels

must have contrast to its background so

when you're looking at an object getting

contrast is one of the most important

things because this way if you have

better contrast you'll have better an

image and you'll get more accurate

results and looking at your objects and

inspecting your objects the difference

between the dark and bright the better

contrast and the better vision system is

able to differentiate them so a rule of

thumb is you want to have a minimum of

20% contrast and that is typically

needed for good different-different

ation between the features providing

good contrast to your object and below

this is just a little chart of kind of

you can tell black and white you have

you're gonna have the most contraire

as soon as you get on down the line when

the greys look similar you're gonna have

less and less contrast so I showed this

slide earlier but this is just kind of

reiterate a few things we're getting

ready to go over on the bottom-left the

image is a sample measurement program we

were doing so when you zoom in also on

the middle image if you zoom in that

would be what the image would look like

and on the big image to the right that

is if you zoom in to maximum as far as

you can go and you can tell on this

image you have zero 1540s and sixty-five

so like we went over earlier complete

black is 0 in the 255 is supposed to be

supposed to be a solid white but I'm not

really sure on my screen it's kind of

looking great but I think that's due to

the background of the template but it's

supposed to be white so you can tell by

in this image to the right it's kind of

showing you on when you go from

completely black and you go out

integrate how your numbers will go up

and you can take a look at this in the

machine vision software so contrast

sensing algorithm it's able to detect

one pick one single pixel with a good

contrast but let's say if this pixel

moves half a pixel diagonal you get the

same features now spread over four

pixels each having a little part of the

original contrast so you can see you

have on the left-hand image you have

completely black it would be a zero in

all 100's around it but if that moved

diagonal you would have that black image

over four pixels now and then see how

your numbers kind of went from zero

completely black and it's now more gray

mukesh white it would go up to a 75 so

what's the smallest size of features

that can be repeatedly detective we say

that a two by two pixel feature is

always kind of guaranteed to be

completely covering at least one full

pixel and

this is not something that we really get

a lot into but when we're kind of

setting our thresholds on our

application and on our arm perfectly

good image when we start a machine

vision application we can kind of look

at these numbers and kind of make a

start from what our threshold should be

depending on what kind of feedback we

get from the machine vision software so

the most common way for vision software

detective feature is finding like strain

defect whatever we're looking for is

finding pixels with certain grade level

ranges like your threshold so threshold

range is set between a lower and opal

threshold of the gray level bright

pixels are defined as pixels from a gray

level threshold to gray level 100% dark

pixels like we went over earlier or zero

so in the image you can tell on the left

that the lower threshold the dark is in

the yellow and on the right the bride is

in the yellow that's just kind of

showing you the difference between dark

and bright and on the image below when

we're checking for the pixels that are

dark yes your vision software is gonna

be able to find him and then we can get

to know their gray level in position in

the image so you can see that there's a

blue square inside the field of view and

that's our region of interest in the

software so we can evaluate the amount

of dark pixels against the bright pits

pixels in the region of interest and

then put a limit to it so if the amount

of dark pixels exceeded this limit on

the inspection on whatever we're looking

for then this will cause the inspection

or your object this will cause it to

fail this is kind of how we go about

getting a good a good pass or a bad fail

so on edge detection this algorithm this

kind of identifies the points where the

image contrast changes sharply this

algorithm measures great level of each

pixel along the region of interest you

can see that on this edge detection on

the left-hand side it starts out gray in

when we get into the black part the gray

level line will drop down to zero and

then when we get back into the gray the

gray level line profile comes back up

back into black goes back down to zero

and then back into gray it goes back up

it looks like it's going back up to a

little over 60 so this is kind of how we

go about setting our thresholds on

objects to set in Pat's we want to pass

or fail and this is kind of based on

this edge detection right here here's

that the grey line profile enhanced at a

better view so this obviously it

indicates the rate of change this

algorithm compares the value of each

pixel along the edge to its neighbor the

greater value indicates a positive

gradient dark to bright lesser value

indicates a negative right to dark and

you can tell that differences in these

two lines kind of indicates the

steepness of the gradient and as you

could tell in the previous slide how the

gradient line profile changed and it

will change depending on whatever object

you're looking at so there's an

algorithm for our edge detection it

usually works with plus or minus 1 pixel

and sub pixel and algorithm the

accuracies plus or minus one twentieth

or a point zero five pixel can be

obtained but you know plus or minus 0.5

can be achieved as more as a

conservative value in a real application

with the camera and the optics in the

light situation so the use of teller

centric optics and Calamity backlight

makes higher accuracy measuring possible

using other optics or light lighting

techniques will influence the accuracy

of the system when you're doing

sub-pixel een honestly it should really

just be considered as insurance only

true true way to know how repeatable a

measurement application will be is to

actually set up the application do the

optics lighting handling of the software

and record the results over set several

hundred triggers in the same static park

then you can compare the users

of what you got on the information and

go from there so a lot of times people

want to do me ask about machine vision

applications and you can't you can run

the numbers so we're kind of decide if

the application is doable or not but

like I just said in the to get um

repeatability and to know if the

application is actually plausible and we

can do it we have to get some parts we

have to test them out we have to set up

the cameras and Lighting's and go

through the software and actually run

the test up several several times and to

see if we can get consistent results

that meet your application needs

so sub-pixel and here's just a little

image a quick image of sub-pixel an

algorithm i just kind of examines how

the gray levels are of the adjacent

pixels around the measurement endpoint

this just uses the information to

interpolate the location of the actual

eggs to the fraction of a pixel here's

well maybe a little better view of it

the edge window algorithm collects

multiple sub pixel samples along the

detective edge on both sides of the

initial dictation point this areas along

the edge of this window determine by the

subpixel settings in the tool and these

settings allow for precise calculation

of sub pixel edge so this algorithm

generates each sub pixel sample by

examining an individual line of pixels

across a detected edge so when we're

doing edge detection this is kind of

just getting into detail pretty much

about what's going on in the software

where you're getting these these numbers

from where you're getting a detective

edges your edge windows in your region

of interests this is also just called

best finding edge it's just another

example of edge detection and what goes

into it sub pixel location of an edge

within a single pixel so you can compare

the gray level to that

edge pixel to the gray levels and the

pixels of either side of it in the

simplified example shown below

on the edges pixels or gray level is a

mix of the left-hand pixels gray level

and the right-hand pixels grade level so

the proportions of the mix depends on

the location in the edge within the edge

pixel so pretty much it's just saying

that all your grade level when your

edges are going to just kind of

determine on where your lines are where

your edges are or your region of

interests are what your pixel counts are

and here's just kind of a algorithm of

how you go about looking for your edges

on your gray level and your right level

this is really nothing that can we

really get into or need to know the

numbers of but these are just kind of

some of the numbers and some of the

feedback in the machine vision software

that we get this is just kind of where

they come from and how we get those

numbers