Machine Vision: 5 Simple Steps to Choose the Right Camera

The machine vision and industrial camera market is offering thousands of models with different resolutionssizes, speeds, colors, interfaces, prices, etc. So, how do you choose? Let’s go through 5 simple steps which will ensure easy selection of the right camera for your application. 

1.  Defined task: color or monochrome camera  

2.  Amount of information: minimum of pixels per object details 

3.  Sensor resolution: formula for calculating the image sensor 

4.  Shutter technology: moving or static object 

5.  Interfaces and camera selector: lets pick the right model 

STEP 1 – Defined task  

It is always necessary to start with the size of the scanned object (X, Y), or you can determine the smallest possible value (d) that you want to distinguish with the camera.

For easier explanation, you can choose the option of solving the measurement task. However, the basic functionality can be used for any other applications.

In the task, the distance (D) between the centers of both holes is determined with the measurement accuracy (d). Using these values, we then determine the parameter for selecting the right image sensor and camera.

Example:
Distance (D) between 2 points with measuring accuracy (d) of 0.05 mm. Object size X = 48 mm (monochrome sensor, because color is not relevant here)

Note: Monochrome or color?
Color sensors use a Bayer color filter, which allows only one basic color to reach each pixel. The missing colors are determined using interpolation of the neighboring pixels. Monochrome sensors are twice as light sensitive as color sensors and lead to a sharper image by acquiring more details within the same number of pixels. For this reason, monochrome sensors are recommended if no color information is needed.

STEP 2 – Amount of information

Each type of application needs a different size of information to solve. This is differentiated by the minimum number of pixels. Lets again use monochrome options.

Minimum of pixels per object details

  • Object detail measuring / detection       3
  • Barcode line width                                           2
  • Datamatrix code module width                4
  • OCR character height                                    16

Example:
The measuring needs 3 pixels for the necessary accuracy (object detail size d). As necessary accuracy (d) which is 0.05 mm in this example, is imaged on 3 pixels.

Note:
Each characteristic or application type presupposes a minimum number of pixels. It avoids the loss of information through sampling blurs.

STEP 3 – Sensor resolution

We already defined the object size as well as resolution accuracy. As a next step, we are going to define resolution of the camera. It is simple formula to calculate the image sensor.

S = (N x O) / d = (min. number of pixels per object detail x object size) / object detail size

Object size (O) can be describe horizontally as well as vertically. Some of sensors are square and this problem is eliminated 😊

Example:
S = (3 x 48 mm) / 0.05 mm = 2880 pixels

We looked at the available image sensors and the closest is a model with resolution 3092 x 2080 => 6.4Mpixels image sensor.

Note:
Pay attention to the format of the sensor.

For a correct calculation, it is necessary to check the resolution, not only in the horizontal but also in the vertical axis.

 

STEP 4 – Shutter technology

Global shutter versus rolling shutter.

These technologies are standard in machine vision and you are able to find hundreds of cameras with both.

Rolling shutter: exposes the motive line-by-line. This procedure results in a time delay for each acquired line. Thus, moving objects are displayed blurrily in the resulting motive through the generated “object time offset” (compare to the image).

Pros:

    • More light sensitive
    • Less expensive
    • Smaller pixel size provides higher resolution with the same image format.

Cons:

    • Image distortion occurs on moving objects

Global shutter: used to get distortion-free images by exposing all pixels at the same time.

Pros:

    • Great for fast processes
    • Sharp images with no blur on moving objects.

Cons:

    • More expensive
    • Larger image format

Note:
The newest rolling shutter sensors have a feature called global reset mode, which starts the exposure of all rows simultaneously and the reset of each row is released simultaneously, also. However, the readout of the lines is equal to the readout of the rolling shutter: line by line.

This means the bottom lines of the sensor will be exposed to light longer! For this reason, this mode will only make sense, if there is no extraneous light and the flash duration is shorter or equal to the exposure time.

STEP 5 – Interfaces and camera selector

Final step is here:

You must consider the possible speed (bandwidth) as well as cable length of camera technology.

USB2
Small, handy and cost-effective, USB 2.0 industrial cameras have become integral parts in the area of medicine and microscopy. You can get a wide range of different variants, including with or without housings, as board-level or single-board, or with or without digital I/Os.

USB3/GigE Vision
Without standards every manufacturer does their own thing and many advantages customers learned to love with the GigE Vision standard would be lost. Like GigE Vision, USB3 Vision also defines:

    • a transport layer, which controls the detection of a device (Device Detection)
    • the configuration (Register Access)
    • the data streaming (Streaming Data)
    • the handling of events (Event Handling)
    • established interface to GenICam. GenICam abstracts the access to the camera features for the user. The features are standardized (name and behavior) by the standard feature naming convention (SFNC). Additionally, it is possible to create specific features in addition to the SFNC to differentiate from other vendors (quality of implementation). In contrast to GigE Vision, this time the mechanics (e.g. lockable cable connectors) are part of the standard which leads to a more robust interface.

I believe that these five points will help you choose the most suitable camera. Are you still unclear? Do not hesitate to contact us or contact me directly: I will be happy to consult your project, needs or any questions.

 

 

Which 3D Vision Technology is Best for Your Application?

3D machine vision. This is such a magical combination of words. There are dozens of different solutions on the market, but they are typically not universal enough or they are so universal that they are not sufficient for your application. In this blog, I will introduce different approaches for 3D technology and review what principle that will be the best for future usage.

Bonus:  I created a poll asking professionals what 3D vision technology they believe is best and I’ve shared the results.

Triangulation

One of the most used technologies in the 3D camera world is triangulation, which provides simple distance measurement by angular calculation. The reflected light falls incident onto a receiving element at a certain angle depending on the distance. This standard method relies on a combination of the projector and camera. There are two basic variants of the projections — models with single-line structure and 2-dimensional geometric pattern.

A single projected line is used in applications where the object is moving under the camera. If you have a static object, then you can use multiple parallel lines that allow the evaluation of the complete scene/surface. This is done with a laser light shaped into a two-dimensional geometric pattern (“structured light”) typically using a diffractive optical element (DOE). The most common patterns are dot matrices, line grids, multiple parallel lines, and circles.

Structured light

Another common principle of 3D camera technology is the structured light technique. System contains at least one camera (it is most common to use two cameras) and a projector. The projector creates a narrow band of light (patterns of parallel stripes are widely used), which illuminate the captured object. Cameras from different angles observe the various curved lines from the projector.

Projecting also depends on the technology which is used to create the pattern. Currently, the three most widespread digital projection technologies are:

  • transmissive liquid crystal,
  • reflective liquid crystal on silicon (LCOS)
  • digital light processing (DLP)

Reflective and transparent surfaces create challenges.

Time of Flight (ToF)

For this principle, the camera contains a high-power LED which emits light that is reflected from the object and then returns to the image sensor. The distance from the camera to the object is calculated based on the time delay between transmitted and received light.

This is really simple principle which is used for 3D applications. The most common wavelength used is around 850nm. This is called near infrared range, which is invisible for human and eye safety.

This is an especially great use since the camera can standardly provide 2D as well as 3D picture in the same time.

An image sensor and LED emitter are used as an all-in-one product making it simple to integrate and easy to use. However, a negative point is that the maximum resolution is VGA (640 x 480) and  for Z resolution expect +/- 1cm. On the other hand, it is an inexpensive solution with modest dimensions.

Likely applications include:

  • mobile robotics
  • door controls
  • localization of the objects
  • mobile phones
  • gaming consoles (XBOX and Kinect camera) or industrial version Azure Kinect.

Stereo vision

The 3D camera by stereo vision is a quite common method that typically includes two area scan sensors (cameras). As with human vision, 3D information is obtained by comparing images taken from two locations.

The principle, sometimes called stereoscopic vision, captures the same scene from different angles. The depth information is then calculated from the image pixel disparities (difference in lateral position).

The matching process, finding the same information with the right and left cameras, is critical to data accuracy and density.

Likely applications include:

  • Navigation
  • Bin-picking
  • Depalletization
  • Robotic guidance
  • Autonomous Guiding Vehicles
  • Quality control and product classification

I asked my friends, colleagues, professionals, as well as competitors, on LinkedIn what is the best 3D technology and which technology will be used in the future. You can see the result here.

As you see, over 50% of the people believe that there is no one principle which can solve each task in 3D machine vision world. And maybe that’s why machine vision is such a beautiful technology. Many approaches, solutions and smart people can bring solutions from different perspectives and accesses.

Buying a Machine Vision System? Focus on Capabilities, Not Cost

Gone are the days when an industrial camera was used only to take a picture and send it to a control PC. Machine vision systems are a much more sophisticated solution. Projects are increasingly demanding image processing, speed, size, complexity, defect recognition and so much more.

This, of course, adds to the new approach in the field of software, where deep learning and artificial intelligence play a bigger and bigger role. There is often a lot of effort behind improved image processing, however,  some people, if only a few, have realized that part of it can already be processed by that little “dummy” industrial camera.

I will try to briefly explain to you in the next few paragraphs how to achieve this in your application. Thanks to that, you will be able to get some of these benefits:

  • Reduce the amount of data
  • Relieve the entire system
  • Generate the maximum performance potential
  • Simplify the hardware structure
  • Reduce the installation work required
  • Reduce your hardware costs
  • Reduce your software costs
  • Reduce your development expenses

How to achieve it?  

Try to use more intelligent industrial cameras, which have a built-in internal memory sometimes called a buffer. Together with FPGA (field programmable gate array) they will do a lot of work that will appreciate your software for image processing. These functions are often also called pre-processing features.

What if you have a project where the camera must send images much faster than the USB or Ethernet interface allows?

For simple cameras, this would mean using a much faster interface, which of course would make the complete solution more expensive. Instead, you can use the Smart Framer Recall function in standard USB and GigE cameras, which generates small preview images with reduced resolution (thumbnails) with an extremely accelerated number of frames per second, which are transferred to the host PC with IDs. At the same time, the corresponding image in full resolution is archived in the camera’s image memory. If the image is required in full resolution, the application sends a request and the image is transferred in the same data stream as the preview image.

The function is explained in this video.

Is there a simpler option than a line scan camera? Yes!

Many people struggle to use line scan cameras and it is understandable. They are not easy to configurate, are hard to install, difficult to properly set and few people can modify them. You can use an area scan camera in line scan mode. The biggest benefit is standard interface: USB3 Vision and GigE Vision instead of CoaXPress and Cameralink. This enables inspection of round/rotating bodies or long/endless materials at high speed (like line scan cameras). Block scan mode acquires an Area of Interest (AOI) block which consists of several lines. The user defines the number of AOI blocks which are used to create one image. This minimizes the overhead, which you would have instead when transferring AOI blocks as single images using the USB3 Vision and GigE Vision protocols.

The function is explained in this video.

Polarization has never been easier

Sony came with a completely new approach to — a polarized filter . Until this new approach was developed, everyone just used a polarization filter in front of the lens and combined it with polarized lighting. With the polarized filter, above the pixel array is a polarizer array and each pixel square contains 0°, 45°, 90°, and 135° of polarization.

 

What is the best part of it? It doesn’t matter if you need a color or monochrome version. There are at least 5️ applications when you want to use it:

  • Remove reflection – > multi-plane surfaces or bruise/defect detection
  • Visual inspection – > detect fine scratches or dust
  • Contrast improvement -> recognize similar objects or colors
  • 3D/Stress recognition -> quality analysis
  • People/vehicle detection -> using your phone while driving

Liquid lens is very popular in smart sensor technology. When and why do you want to use it with an Industrial camera?  

 

Liquid lens is a single optical element like a traditional lens made from glass. However, it also includes a cable to control the focal length. In addition, it contains a sealed cell with water and oil inside. The technology uses an electrowetting process to achieve superior autofocus capabilities.

Benefits to the traditional lenses are obvious. It doesn’t have any moving mechanical parts. Thanks to that, they are highly resistant to shocks and vibrations. Liquid lens is a perfect fit for applications where you need to observe or inspect objects with different sizes and/or working distances and you need to react very quickly. One  liquid lens can do the work of multiple-image systems.

To connect the liquid lens, it requires the RS232 port in the camera plus a DC power from 5 to 24 Volt. An intelligent industrial camera is able to connect with the camera directly and the lens uses the power supply of the camera.

 

What Machine Vision Tool is Right for Your Application?

Machine vision is an inherent terminology in factory automation but selecting the most efficient and cost-effective vision product for your project or application can be tricky.

We can see machine vision from many angles of view, for example market segment and application or image processing deliver different perspectives. In this article I will focus on the “sensing element” itself, which scan your application.

The sensing element is a product which observes the application, analyzes it and forwards an evaluation. PC is a part of machine vision that can be embedded with the imager or separated like the controller. We could take many different approaches, but let’s look at the project according to the complexity of the application. The basic machine vision hardware comparison is

  1. smart sensors
  2. smart cameras
  3. vision systems

Each of these products are used in a different way and they fit different applications, but what do they all have in common? They must have components like an imager, lens, lighting, SW, processor and output HW. All major manufacturing companies, regardless of their focus or market segment, use these products, but what purpose and under what circumstances are they used?

Smart Sensors

Smart sensors are dedicated to detecting basic machine vision applications. There are hundreds of different types on the market and they must quickly provide standard performance in machine vision. Don’t make me wrong, this is not necessarily a negative. These sensors are used for simple applications. You do not want to wait seconds to detect QR code; you need a response time in milliseconds. Smart sensors typically include basic functions like:

  • data matrix, barcode and 2D code reading
  • presence of the object,
  • shape, color, thickness, distance

They are typically used in single purpose process and you cannot combine all the features.

Smart Cameras

Smart cameras are used in more complex projects. They provide all the function of smart sensors, but with more complex functions like:

  • find and check object
  • blob detection
  • edge detection
  • metrology
  • robot navigation
  • sorting
  • pattern recognition
  • complex optical character recognition

Due to their complexity, you can use them to find products with higher resolution , however it is not a requirement. Smart cameras can combine more programs and can do parallel several functions together. Image processing is more sophisticated, and limits may occur in processing speed, because of embedded PC.

Vision Systems

Typically, machine vision systems are used in applications where a smart camera is not enough.

Vision system consists of industrial cameras, controller, separated lighting and lens system, and it is therefore important to have knowledge of different types of lighting and lenses. Industrial cameras provide resolution from VGA up to 30Mpxl and they are easy connected to controller.

Vision systems are highly flexible systems. They provide all the functions from smart sensors and cameras. They bring complexity as well as flexibility. With a vision system, you are not limited by resolution or speed. Thanks to the controller, you have dedicated and incomparable processing power which provides multi-speed acceleration.

And the most important information at the end. How does it look with pricing?

You can be sure that smart sensor is the most inexpensive solution. Basic pricing is in the range of $500 – $1500. Smart cameras can cost $2000 – $5000, while a vision system cost would start closer to $6000. It may look like an easy calculation, but you need to take into consideration the complexity of your project to determine which is best for you.

Pros Cons Cost
Smart sensor
    • Easy integration
    • Simple configuration
    • Included lightning and lenses
    • Limited functions
    • Closed SW
    • Limited programs/memory
$
Smart camera
    • Combine more programs together
    • Available functions
    • Limited resolution
    • Slower speed due to embedded PC
$$
Vision system
    • Connect more cameras(up to 8)
    • Open SW
    • Different resolution options
    • Requires skilled machine vision specialist
    • Requires knowledge of lightning and lenses
    • Increased integration time
$$$

Capture