The Art of Machine Vision Lighting “ People often say that machine vision lighting is an art. Being an avid student of fine arts for most of my life, the notion of becoming a machine vision lighting artist intrigues me. I can imagine offering a class at a local university, ‘ The Art of Machine Vision Lighting - Design Aesthetics for the Artsy Engineer. ’ I find, however, my endeavor to study the discipline with its various mediums comes to an abrupt halt when I attempt to find an art textbook covering the topic of machine vision lighting. Machine vision lighting is a science. All the information to help you be successful is found within science and engineering resources. By creatively (perhaps the creative effort confuses it as an art form) and intelligently applying knowledge of the properties of light to the technology of machine vision, the practitioner will create successful images for machine vision analysis.” -Allen Burns, Northeast Robotics, Inc.
Pollinating insects, particularly bees, see well into the UV range, which assists them when searching for flowers. As light wavelength increases, energy levels per photon, decrease; this is primarily why UV light is more dangerous; black lights are UV-A (315 nm to 400 nm) UV light in the 280nm to 315nm range (UV-B) is the most damaging to eye tissue, whereas UV-C (100nm – 280nm), although more energetic, is readily absorbed in the atmosphere within a few meters, thus from a distance, relatively harmless to humans. Near IR (720nm – 1100nm) is commonly used in active vision and surveillance applications; far IR (>1100nm) is referred to as thermal or heat signature IR, and is passively recorded by an IR-sensitive CCD camera. IR light is more difficult to focus and diffuse, and because of its longer wavelength, it penetrates deeper into materials than visible light; UV light, on the other hand can interact with or be completely absorbed by some lens materials, so special optics may be needed. In contrast, a CCD sensor has a more linear response to light, as opposed to both the human visual system and photographic film. In other words, the CCD sensor more accurately reproduces the true light intensities between black and white. Photographic film has much less dynamic range than a CCD sensor, and about the same quantum efficiency as normal daytime human vision. B&W photographic film does not collect Near-IR light above 700 nm because the light may pass through the film emulsion, including the 5 um thick photo-sensitive area; similarly, it does not collect UV because the 1.5 um thick protective top coating absorbs the UV before it can pass into the photo-sensitive area.
Photopic: Light adapted human vision – peak 555 nm (yellow-green); Actual visible range varies from 420 nm to over 700 nm. Scotopic: Dark adapted human vision – peak 507 nm, and approx 2X more sensitive than photopic vision. The quantum efficiency of the photopic human visual system is < 5% at 555 nm, meaning it can detect 100 photons for a 100 usec exposure time. However, the human visual system requires approx 10X this much light, aided by the mind’s interpretive powers, to recognize a scene in an image. Whereas human visual acuity is good, particularly to color, humans generally can see fewer than 20 gray levels between pure black and pure white.
Camera sensitivity is not necessarily related to the spatial resolution of the sensor, e.g. – 640 x 480 vs. 1280 x 1024, but more by individual pixel size. A pixel 2X larger in X and Y dimensions is 4X more sensitive, given other parameters are similar. So, a camera of pixel resolution 640 x 480 may have less sensitivity than one w/ 1280 x 1024 if its pixels are smaller. To accurately determine sensitivity and resolution of a final image, one must know the number of pixels (in X&Y), the individual pixel size, and the sensor actual dimensions in X&Y. B&W CCD cameras are more sensitive than their color counterparts because of the process used to pass the color information onto the sensor. Standard color CCD cameras employ a color filter array mask, which makes pixels sensitive to certain colors, such as R,G,B in the case of a Bayer Pattern, for example. This mask attenuates some of the light intensity, resulting in differences in peak sensitivity measurements. Differences in wavelength-specific sensitivity among cameras varies according to many parameters – including, but not limited to – sensor design, materials, camera electronics, and coatings or treatments. For example, many large format sensor cameras, such as 1K x1K or larger megapixel types, have enhanced UV light sensitivity because of a coating that fluoresces when UV light is incident on the sensor, thus increasing the apparent UV photon content, which is then processed in similar fashion to the other wavelengths. Optimizing the LED or other light wavelength to the CCD camera sensor responsivity can greatly increase the efficiency and effectiveness of a vision system. For example, xenon light, while very bright, has wavelengths, both in the near UV and near IR, that most CCD sensors cannot detect, and thus collect. Whereas, LED light, particularly red, is virtually all collected at or near the sensor’s peak sensitivity. Whereas, the xenon lamp may appear brighter, the LED light may be more efficient, and not offer all the other negative issues associated with xenon lighting in general.
Near-IR light (from 720 – 1100 nm wavelength) is used in many surveillance and vision applications. It is very effective in negating color differences, differentiating objects based on textural and/or materials composition differences, and is easier to use in strobing applications, particularly if there are ergonomic issues to consider.
Note that even though the red 660 nm light reveals the blue dot matrix print, it does not penetrate as well as the IR light through the bottle paper label. The IR transmitted so much better through the glass that the lens was stepped down 5 stops to match the intensity measured for the red light; remember that each step down in f-stop represents a decrease in light intensity of 2X. Conversely, the red light interacted with the blue ink of the date and lot code to darken it, whereas the IR simply shot right through the ink, rendering it virtually invisible and certainly undetectable.
Note the “holes” in the PCB near top center in both images – in the case of the red light, it is so intense that the hole appears larger because of blooming. Even though the red light is much more intense, i.e. – the camera is more sensitive to the red light, the IR light penetrates the board more to distinguish the internal traces more clearly and with better definition. The primary reason the IR back lit board has better internal trace definition, seemingly contradictory, is that the red light, being a shorter wavelength, simply diffuses and scatters more readily in the green plastic board material.
UV fluorescent lights are commonly used to read cancelled check stamps, or other codes, such as in the lower set of motor oil bottle images above. Many polymers, particularly nylon fluoresce readily under UV-A, and UV-B light, including structural fibers and threads as well as seat belt stitching. Specially “doped” oils are used for leak tests and for detecting cracks and/or defects in critical engine parts, such as cranks or blocks. Often, the dye manufacturer can custom the dye or ink to fluoresce in response to different excitation wavelengths.
Several conditions are necessary for fluorescence vision to be effective. The sample must be excitable under UV light, the UV light wavelength, as well as the wavelength necessary to excite the sample must be known, and there needs to be a certain threshold level of intensity from the light. UV LED lights are just now available in 390 and 370 nm wavelengths; however, the 370 nm versions are not very bright.
The use of a monochrome light, such as red 660 nm, and a matched band pass filter, attached to the camera lens, is a very cost-effective substitute for using a work cell enclosure to block ambient light. The red 660 nm band pass filter will allow 95% of the red light into the camera, while diminishing excluded wavelengths by as much as 35X the input contributed by fluorescent lighting, for example. It is also possible to enhance the apparent responsivity of the camera by using wavelength-matched filters to block any ambient light contribution, which might “drown” out the light of interest. For example, one could use a short pass filter of 410 nm on the light, which passes only the 390 nm UV light to the sample, then use a 520 nm yellow-green band pass filter on the lens to capture only the excitation light from the fluorescing sample. Most color CCD cameras have a 700 nm short pass filter, meaning that they pass all light to 700 nm, because IR light affects the color consistency and calculations, often making greens appear brown.
A pair of polarizers are very effective light reduction devices, similar to neutral density filters.
Vision Lighting Seminar DVT Advanced Training Minneapolis, MN July 2004 Daryl Martin Midwest Sales & Support Manager Ann Arbor, MI A Creators of Evenlite®
Today’s Objectives <ul><li>Introduction to Advanced illumination (Ai), including Evenlite ® and LED Lighting </li></ul><ul><li>Machine Vision Illumination Principles & Techniques </li></ul><ul><li>Sample Applications </li></ul><ul><li>Imaging Beyond the “Visible” – Near IR & UV </li></ul><ul><li>“ Pass” Filters and Polarization </li></ul>
<ul><li>Privately owned LED light design and manufacturing company, located in Central Vermont </li></ul><ul><li>Creators of Evenlite® Technology </li></ul><ul><li>Legitimized use of LEDs in Machine Vision </li></ul><ul><li>Grew by applying a customer-service based approach to solving real customer illumination problems </li></ul>
Source Comparisons Very Thin; Low Heat 2000 to 5000 Dim Green Electro- Luminescent Expensive Stable 3000 to 7000 Very Bright White w/blue Xenon Inexpensive High Heat 200 to 3000 Very Bright White w/yellow Halogen Inexpensive; Need High Freq 5000 to 7000 Bright White w/blue-green Fluorescent Long life Stable output Up to 100,000 Bright to Very Bright Various LED Comments Life (hrs) Intensity Spectrum Type
Optical Characteristics of an LED Mechanical Axis Optical Axis Evenlite ® Technology
Broad Area Linear Arrays AL4424-660 (BALA) Evenlite ® Technology
RGB Lights <ul><li>Red, Green, Blue light heads offer flexibility </li></ul><ul><li>Equal parts RGB creates white light </li></ul><ul><li>Color mixing capabilities </li></ul>Evenlite ® Technology
Machine Vision Imaging <ul><li>Is it Art or Science? </li></ul>lkasdjflkj lkasdjflkj lkasdjflkj Or both ?
Methods for Solving Lighting Problems <ul><li>Wave and Look (most common) </li></ul><ul><ul><li>Image the part while trying different sources at different positions </li></ul></ul><ul><li>Scientific Analysis (most effective) </li></ul><ul><ul><li>Analyze the imaging environment and recommend the best solution </li></ul></ul>
<ul><li>Lighting is one of the least expensive, most flexible components in any machine vision inspection system. </li></ul><ul><li>Sample-appropriate Lighting is critical to a successful inspection. </li></ul>The right light helps the vision system do its job.
Brief Review of Light and Optics for Vision Illumination
400 nm 500 nm 600 nm 700 nm 390 455 470 505 520 595 625 660 695 735 The Visible Light Spectrum <ul><li>Light is Seen Differently by film, humans and CCDs </li></ul>UV IR Human Visible Range
Spectral Response - CCD vs. Human Vision <ul><li>Photopic: Normal light-adapted vision </li></ul><ul><li>Scotopic: Dark-adapted vision </li></ul><ul><li>Don’t always rely on your eyes! </li></ul>
Spectral Response of Std Resolution CCDs 450-700 nm *Averaged responses from major CCD manufacturers <ul><li>B&W CCDs are ~ 10X more sensitive than color CCDs </li></ul><ul><li>CMOS cameras are more IR-sensitive than Std CCDs </li></ul>
Where Does the Light Go? <ul><li>Total Light In = </li></ul><ul><li>Reflected Light + </li></ul><ul><li>Absorbed Light (may be re-emitted) + </li></ul><ul><li>Transmitted Light </li></ul>Illumination Reflect Emit Absorb Transmit
Reflection on Specular Surfaces <ul><li>Light reflects at the angle of incidence </li></ul><ul><li>Just like a pool ball off the bumper </li></ul><ul><li> </li></ul> <ul><li>Surface Angle determines where light comes from in order to illuminate the surface </li></ul>
Divergence and Intensity <ul><li>Intensity falls with the inverse square of the divergence radius </li></ul><ul><li>I = 1/r 2 </li></ul><ul><li>Use collimation and short working distances when possible </li></ul>
Lighting Environment and the Part <ul><li>Ring Light </li></ul><ul><ul><li>Small Solid Angle </li></ul></ul><ul><li>Continuous Dome </li></ul><ul><ul><li>Large Solid Angle </li></ul></ul>
Issues to Consider <ul><li>How an object responds to light </li></ul><ul><li>What kind of surface an object has </li></ul><ul><li>Object material composition/texture </li></ul><ul><li>Object color </li></ul><ul><li>Light modeling and structure </li></ul><ul><li>Camera sensor response </li></ul><ul><li>Uncontrolled aspects of the inspection environment, such as ambient light, etc. </li></ul>
The Optimized Image <ul><li>It’s All About Contrast! </li></ul><ul><li>Feature Separation or Segmentation </li></ul><ul><ul><li>Maximum contrast </li></ul></ul><ul><ul><ul><li>features of interest </li></ul></ul></ul><ul><ul><li>Minimum contrast </li></ul></ul><ul><ul><ul><li>features of no interest (noise) </li></ul></ul></ul><ul><ul><li>Minimum sensitivity to normal variations </li></ul></ul><ul><ul><ul><li>minor part differences </li></ul></ul></ul><ul><ul><ul><li>presence of, or change in ambient lighting </li></ul></ul></ul><ul><ul><ul><li>sample handling / presentation differences </li></ul></ul></ul>
Creating Contrast – Lighting Cornerstones <ul><li>Change Light Direction w/ Respect to Sample and Camera (Geometry) </li></ul><ul><li> - 3-D spatial relationship - sample, light & camera </li></ul><ul><li>Change Light Pattern (Structure) </li></ul><ul><li> - Light Head Type: Spot, Line, Dome, Sheet </li></ul><ul><ul><li>- Illumination Type: B.F. - D.F. - Diffuse - B.L. </li></ul></ul><ul><li>Change Spectrum (Color / Wavelength) </li></ul><ul><li> - Monochrome, white vs. sample / camera response </li></ul><ul><ul><li>- Warm vs. cool color families – object vs. background </li></ul></ul><ul><ul><li>Need to understand the impact of incident light on both the part of interest and its immediate background! </li></ul></ul>
Using Color <ul><li>Use Colored Light to Create Contrast </li></ul><ul><ul><li>Use Like Colors or Families to Lighten (yellow light makes yellow features brighter) </li></ul></ul><ul><ul><li>Use Opposite Colors or Families to Darken (red light makes green features darker) </li></ul></ul>Warm Cool R V O B Y G
Using RGB View with color camera under white light View with B&W camera under white light
Red LED Green LED Blue LED Using Monochrome LED Illumination White Light
Dark Field <ul><li>Angled light </li></ul><ul><li>Used on highly reflective surfaces </li></ul><ul><li>OCR or surface defect applications </li></ul>
Result of Dark-Field Light <ul><li>Emphasize Height Changes </li></ul><ul><li>Diffuse Surfaces are Bright </li></ul><ul><li>Flat Polished Surfaces are Dark </li></ul><ul><li>Shape and Contour are Enhanced </li></ul>
Axial Diffuse <ul><li>Light directed at beam splitter </li></ul><ul><li>Used on reflective objects </li></ul>
Result of Axial Diffuse Illumination <ul><li>Surface Texture Is Emphasized </li></ul><ul><li>Angled Elevation Changes Are Darkened </li></ul>
Diffuse Dome <ul><li>Similar to the light on an overcast day. </li></ul><ul><li>Creates minimal glare. </li></ul>
Stamped Date Code <ul><li>Recessed metal part </li></ul><ul><li>Reflective, textured, flat or curved surface </li></ul>Dark Field ring light Line light Bright field ring light Bright field spot light
Data Matrix <ul><li>Peened data matrix </li></ul><ul><li>Flat, shiny surface </li></ul>Broad Area Linear Array Standard Dome Bright Field Ring Light Dark Field Ring Light
UPC Bar Code <ul><li>Printing beneath cellophane wrapped package </li></ul>Broad Area Linear Array Bright Field Ring Light Dark Field Ring Light Axial Diffuse Illuminator
Bar Code under Clear Wrap Broad Area Linear Array
Ink Jet OCR <ul><li>Purple Ink </li></ul><ul><li>Concave, reflective surface </li></ul>Diffuse Dome Bright Field Spot Light Axial Diffuse Illuminator Dark Field Ring Light Bright Field Ring Light
Imaging Beyond “Visible” – Near IR <ul><li>Infra-red (IR) light interacts with sample material properties, often negating color differences. </li></ul>Red White Yellow White light – B&W Camera IR light – B&W Camera Black
Imaging Beyond “Visible” – Near IR <ul><li>Red 660 nm light reveals the blue dot matrix printed bottle date & lot codes. </li></ul>Red 660nm Back Light IR 880nm Back Light
Imaging Beyond “Visible” – Near IR <ul><li>Near IR light can penetrate materials more easily because of the longer wavelength. </li></ul>Red 660 nm Back Light IR 880 nm Back Light
Imaging Beyond “Visible” – Near UV <ul><li>Near UV light when used w/ a matched UV excitation dye, illuminates codes and structural fibers. </li></ul><ul><li>Top Image Set: Diaper </li></ul><ul><li>Lower Image Set: Motor Oil Bottle </li></ul>
Imaging Beyond “Visible” – Near UV <ul><li>Near UV light fluoresces many polymers, including nylon. </li></ul><ul><li>Top Image: UV Light, B&W CCD </li></ul><ul><li>Lower Image: UV Light, Color CCD </li></ul>
Designing Vision Lighting <ul><li>When designing a vision system and parts handling / presentation, optimize the lighting solution early in the process, if possible. </li></ul><ul><li>Remember like colors (or color family) lighten and opposite colors darken. </li></ul><ul><li>If using monochrome LED light, let a band pass filter control ambient light. </li></ul><ul><li>Think Geometry – the 3-D spatial relationship among sample, light, and camera. </li></ul><ul><li>Consider how the light pattern and color will potentially interact with the sample and background surfaces. </li></ul>
Light Application Tips -Light close to part -Large footprint -Camera close to light -Spot size is ½ light inner diameter -Light close to part -Large footprint -Ambient light minor -Beam splitter lowers light to camera -Light must be very close to part -Large footprint -Limited spot size -Ambient light may interfere -No WD limit (limited only to intensity need on part) Requirements -Specular / Non -Curved surfaces -If ambient light issues -Specular / Non -Flat / Textured -Angled surfaces -Specular / Non -Surface / Topo -Edges -Look thru trans- parent parts -Non specular -Area lighting -May be used as a dark field light When To Use Use Specular Use Specular Negate Specular No Specular Dome Diffuse Box Angled Ring, Bar Ring, Spot Type Diffuse Dome Full Bright Field Diffuse Axial Full Bright Field Dark Field Partial Bright Field
<ul><li>Pass filters exclude light based on wavelength. </li></ul><ul><li>Sunlight and mercury vapor light are reduced by 4X </li></ul><ul><li>Fluorescent light is reduced by 35X </li></ul>510 nm Short Pass 715 nm Long Pass 660 nm Band Pass Pass Filters in Machine Vision
Pass Filters <ul><li>Top Image: UV light w/ strong Red 660 nm “ambient” light. </li></ul><ul><li>Bottom Image: Same UV and Red 660 nm “ambient” light, with 510 nm Short Pass filter applied. </li></ul>
Polarizing Filters in Machine Vision <ul><li>Top image: It’s difficult to determine whether there is nylon in the nut. </li></ul><ul><li>Bottom image: Application of crossed polarizers allows the nylon to be detected in the nut on the right side. </li></ul>Back Light - No Polarizer Back Light - Crossed Polarizers
Light Specification: Rules-of-Thumb <ul><li>Maximize the contrast of features of interest; minimize contrast on all others; minimize external influences </li></ul><ul><li>Differentiating colors? – Think B&W camera & color lights </li></ul><ul><li>Ambient light issues? – Try monochrome light and a matched band pass filter </li></ul><ul><li>Shiny, curved surfaces? – Try a diffuse dome light </li></ul><ul><li>Shiny, flat, but textured surfaces? – Try axial diffuse </li></ul><ul><li>See surface topography? – Think dark-field (low angle) </li></ul><ul><li>When inspecting plastics – Try UV or IR light </li></ul><ul><li>Need to see features through a reflective cover? – Try low angle linear lights (dark-field) </li></ul><ul><li>Light combinations can solve problems too </li></ul><ul><li>Strobing can generate up to 20x as much light </li></ul>
In Summary: <ul><li>It’s all about contrast! </li></ul><ul><li>Maximize contrast on important features </li></ul><ul><li>Minimize contrast on unimportant features </li></ul><ul><li>Minimize sensitivity to noise and ambient </li></ul><ul><li>Remember structure/geometry and warm </li></ul><ul><li>vs cool colors </li></ul><ul><li>Don’t hesitate to call a qualified lighting </li></ul><ul><li>professional if you get stuck! </li></ul>
<ul><li>Super Bright LumiLed Rings & Spots </li></ul><ul><li>LumiLed Line Lights for High-speed Linescan CCD Applications </li></ul><ul><li>New Family of High Intensity Strobe and DC Continuous Power Sources </li></ul><ul><li>More Variations on Diffuse Lights </li></ul><ul><li>More IP65 Compliant Lights for Wash Down </li></ul><ul><li>Standardized Industrial Cabling Options </li></ul><ul><li>Long-lived UV LED Lights (390 nm – 40,000 hr life) </li></ul>Available In the Future