Area scan cameras

Industrial machine vision uses cameras, optics, lighting, image processing and industrial interfaces to evaluate visual information automatically. Typical tasks are inspecting, measuring, identifying, positioning, sorting, reading codes and guiding robots. Unlike ordinary photography, the point is not a beautiful image but a reproducible technical decision or measurement.

Computer vision is the broader technical field in which computers extract information from images or image sequences. Machine vision typically refers to the industrial implementation of that - including camera, lighting, optics, triggering, communication with the PLC and production line, and reproducible image capture. The terms overlap but are not entirely identical.

A typical system can consist of the following components: industrial camera lens lighting lighting controller trigger or sensor cables industrial PC or vision controller frame grabber, if required machine vision software PLC or fieldbus communication In smart cameras or vision sensors, several of these components are already integrated into one device. HIKROBOT, for example, integrates lighting, image capture, processing and communication into a single device on certain vision sensors.

Typical applications are: presence inspection completeness check position and orientation detection measurement surface inspection colour detection OCR OCV barcode and 2D code reading code verification robot guidance pick-and-place quality classification completeness and assembly inspection Which task can be solved reliably depends decisively on whether the relevant feature can be represented in the camera image with sufficient contrast and sufficient spatial resolution.

A presence inspection checks whether a particular part or feature is present. Examples: Is a screw fitted? Is a label present? Is a part inside the packaging? Is a connector plugged in? This often does not require highly complex image processing. What matters is stable lighting and an unambiguous representation of the feature.

A completeness check verifies whether all required components or features are present. For example, a camera system can check whether a connector has all its pins or whether a package contains all the intended products.

In position detection, the vision system determines the position and often also the rotation of an object. Typical outputs are: X position Y position angle Z position where applicable This information can then be used by a robot, for example.

In pattern matching, the vision system searches the image for known geometric or visual features. Modern methods can use classic pattern matching algorithms or AI and deep learning techniques.

Here, geometric properties of a workpiece are determined from a camera image, for example: diameter distance width angle position gap height with suitable 3D methods For precise measurements, the optics, calibration, lighting and mechanical stability matter just as much as the camera.

Before selecting products, at least the following points should be clarified: What has to be detected or measured? How large is the overall field of view? How large is the smallest relevant feature? How fast does the object move? What working distance is possible? What surface does the object have? Is color relevant? How accurately does it have to measure? How quickly must the result be available? Which interfaces to the line are needed? Only then should camera, lens, lighting and interface be selected. Basler likewise recommends considering resolution, sensor, speed, shutter and interface together when choosing a camera.

An industrial camera is a camera designed for automated, reproducible image capture. Typical characteristics are: hardware trigger configurable exposure time digital inputs and outputs standardised image interfaces defined pixel and raw data formats continuous industrial operation SDK or GenICam support Unlike consumer cameras, the focus is on controllable image capture and industrial integration.

An area scan camera has a two-dimensional sensor and captures a complete 2D image in a single exposure. It is the typical choice for individual workpieces or areas that fit entirely into the field of view.

A line scan camera captures one or more narrow sensor lines. As the workpiece or camera moves, a two-dimensional image is built up line by line. It is particularly suited to continuous materials and very wide or long objects.

An area scan camera produces a complete image in one capture. A line scan camera captures individual lines continuously. This requires a defined relative movement between camera and object. Line scan cameras are particularly interesting for: film paper textiles metal strip battery foil wood glass print cylindrical objects

A monochrome camera essentially measures light intensity per pixel, without the color separation of a Bayer filter. This makes particularly direct intensity information available for many inspection tasks.

Many industrial color cameras have a color filter pattern in front of the sensor, often a Bayer pattern. The individual sensor pixels capture different color components. A color image is then produced from this raw data by demosaicing or debayering.

Where color itself is not a relevant inspection feature, a monochrome camera is often the better choice. Typical examples: measurement edge detection surface inspection barcode reading position detection Without the Bayer color filter, no color information has to be captured or interpolated.

A color camera makes sense where color differences are part of the inspection. Examples: checking cable cores sorting by color packaging inspection print inspection food coloured markings

CMOS is the sensor technology that dominates industrial cameras today. Modern CMOS sensors offer high resolutions, high frame rates, low power consumption and, depending on the sensor type, global or rolling shutter operation. CMOS has largely replaced CCD in industrial machine vision.

CCD is an older image sensor technology that was long known for high image quality and high sensitivity. Today, new industrial cameras predominantly use CMOS sensors. CCD therefore remains relevant mainly in older systems or special applications.

Resolution describes the number of available sensor pixels. A camera with 2,448 × 2,048 pixels has, for example: 2,448 × 2,048 = 5,013,504 pixels so roughly 5 megapixels. More pixels do not automatically mean a better image, however. The lens, pixel size, light, sensor quality and inspection task have to fit as well.

One megapixel is roughly one million pixels. The megapixel count alone says nothing about how well small features can actually be detected.

That depends on the field of view and the smallest feature to be detected. An important first calculation is: object resolution [mm/pixel] = field of view [mm] / number of pixels Example: field of view = 200 mm camera = 2,000 pixels horizontally 200 / 2,000 = 0.1 mm/pixel That means one pixel corresponds to 0.1 mm on the object in that direction.

If a 0.5 mm feature is to be represented by at least 5 pixels, for example: required object resolution = 0.5 mm / 5 = 0.1 mm/pixel With a field of view of 200 mm: 200 mm / 0.1 mm/pixel = 2,000 pixels So at least roughly 2,000 pixels are needed horizontally. Important: the number of pixels required per feature depends on the application. There is no universally valid figure such as “2 pixels are always enough”.

No. 0.05 mm/pixel does not automatically mean the system measures to an accuracy of 0.05 mm. The real measurement uncertainty also depends on: lens resolution and MTF distortion lighting and contrast calibration algorithm sub-pixel methods focus mechanical stability temperature perspective High measurement accuracy therefore requires the complete optical system to be considered.

That cannot be defined universally. A high-contrast defect can be detected reliably with fewer pixels than a very faint defect on a textured surface. For the design you should therefore define: smallest defect + required detection reliability + real contrast and then, where possible, carry out a practical test.

Pixel size describes the physical size of a sensor pixel, for example: 2.74 µm 3.45 µm 5.0 µm It influences light collection, sensor size and optical requirements, among other things. A camera with smaller pixels typically places higher demands on the resolution of the lens.

No. Under comparable conditions, larger pixels can collect more photons and therefore offer advantages in low light. Smaller pixels, on the other hand, allow a higher pixel count on the same sensor area. Which is better depends on resolution, light, sensor quality and optics.

Sensor size describes the physical size of the active image sensor. It influences: the image circle required from the lens the achievable field of view lens size the mount required The lens must provide a large enough image circle to illuminate the sensor fully.

These are historical optical format designations and do not correspond directly to the actual sensor diagonal in inches. A “1-inch sensor” therefore does not have a sensor diagonal of 25.4 mm. Fujinon, for example, quotes roughly 12.8 × 9.6 mm, or a 16 mm diagonal, for the classic 1" format. With modern sensors the specific dimensions can differ from the historical format, so the actual sensor width, height and diagonal should always be checked when selecting a lens.

Quantum efficiency (QE) describes what proportion of the photons hitting a pixel is converted into usable electrons. It is wavelength-dependent. High QE can be an advantage particularly in low light. Basler therefore names quantum efficiency as an important factor for image quality.

Full well capacity describes approximately how many photoelectrically generated charge carriers a pixel can hold before it saturates. Together with the noise behavior, it determines the maximum usable signal range.

Dark noise arises from electrical and thermal processes inside the sensor even without usable light. It typically becomes more significant with longer exposure times and higher sensor temperatures.

SNR stands for signal-to-noise ratio. The higher the useful signal is in relation to the noise, the more cleanly small differences in intensity can be detected. Electronic gain amplifies not only the useful signal but typically also the noise present, and therefore does not automatically improve the SNR.

Dynamic range describes a sensor's ability to distinguish very weak and very strong signals within one capture. A high dynamic range is helpful, for example, in scenes with very dark and very bright areas at the same time.

EMVA 1288 is a standard for the objective characterization of cameras and image sensors. It allows comparable figures to be quoted for, among other things: quantum efficiency sensitivity noise saturation dynamic range signal-to-noise ratio For a serious camera comparison, such standardized measurements are more meaningful than marketing terms.

In a global shutter sensor, the pixels relevant to an image are exposed largely simultaneously. This makes such sensors particularly suitable for fast-moving objects and precise geometric inspections.

In rolling shutter sensors, sensor lines are exposed and read out one after another. With moving objects this can cause geometric distortion. For static or slow-moving scenes, rolling shutter can be perfectly adequate.

Global shutter is recommended in particular for: fast conveyor belts robotics highly dynamic objects precise geometric measurement in motion Rolling shutter can make sense for: stationary objects slow-moving objects cost-sensitive applications applications where the particular rolling shutter sensor offers other relevant advantages

The exposure time determines how long a pixel collects light for one capture. Longer exposure: more light greater risk of motion blur Shorter exposure: less motion blur less light

The distance traveled during an exposure is: movement = speed × exposure time Example: speed = 1,000 mm/s exposure time = 100 µs = 0.0001 s 1,000 × 0.0001 = 0.1 mm At 0.05 mm/pixel that corresponds to: 0.1 / 0.05 = 2 pixels of motion blur

If a maximum of 0.05 mm of movement is permissible and the object moves at 1,000 mm/s: t = distance / speed t = 0.05 / 1,000 = 0.00005 s = 50 µs The permissible motion blur has to be defined for the specific application.

Gain is the amplification of the camera signal. With analogue gain the signal is amplified before analogue-to-digital conversion; digital gain works on data that has already been digitized. High gain can make a dark image appear brighter, but it does not create additional photons and also amplifies any noise present. Better lighting is therefore often preferable.

If the object is stationary and there is enough time, a longer exposure time is often more advantageous than high gain. With fast objects, however, the exposure time has to be limited because of motion blur. More light or strobe lighting should then be the first choice.

HDR stands for high dynamic range. Here the camera or image processing attempts to represent a wider brightness range than a normal capture would. The specific implementation differs by sensor and camera - multiple exposures or special sensor readout methods, for example.

With multi-exposure, several captures or signals with different exposure conditions are used. This allows bright and dark areas of an object to be captured together more effectively, for example. How exactly it works depends on the camera.

In burst mode several images are captured in quick succession, often after a trigger. The maximum number and speed possible depend on the camera.

ROI means region of interest; AOI often means area of interest. Only part of the sensor is read out or processed. The advantages can be: less data shorter processing a higher frame rate on many sensors Whether and how much the frame rate increases depends on the sensor architecture and the ROI geometry.

In binning, neighbouring sensor elements or their values are combined. This reduces the spatial resolution. Depending on the sensor implementation, signal behavior, data rate or sensitivity can improve in return. As binning is implemented in analogue, digital or other ways depending on the sensor type, the specific effects should be checked in the camera manual.

In decimation only some of the pixels are used or output - every second pixel, for example. The image resolution drops, without the method necessarily being identical to true sensor binning.

Gamma is a non-linear mapping between input brightness and the pixel value output. It allows dark or mid-range brightness levels in particular to be represented differently. For quantitative measurements, bear in mind that a gamma correction changes the original linear intensity relationship.

LUT means look-up table. An input pixel intensity is mapped to a new output value using a table. This allows brightness and contrast to be transformed deliberately, for example.

Flat-field correction compensates for position-dependent brightness differences caused by, for example: differences between sensor pixels lens vignetting uneven lighting It should not, however, be seen as a substitute for lighting that is fundamentally badly designed.

Shading correction describes the correction of spatial brightness differences in the camera image. Depending on the manufacturer it can be part of a flat-field or vignetting correction.

A Bayer filter is a color filter pattern in front of the sensor pixels. Individual pixels typically capture red, green or blue components. An RGB image is then reconstructed from the different pixel information.

“Bayer RG” describes the arrangement of the Bayer color pattern. The number denotes the pixel bit depth or data format - 8, 10 or 12 bit, for example. The exact packing or transfer format has to be considered in addition.

Debayering, or demosaicing, is the reconstruction of a complete color image from the individual Bayer sensor values. Missing color components are calculated from neighbouring pixels.

White balance adjusts the gain of the color channels so that neutral surfaces are rendered as neutral as possible. For reproducible color evaluation, the lighting should also be stable and spectrally defined.

The important industrial interfaces are: GigE Vision 2.5/5/10GigE-based vision systems USB3 Vision CoaXPress Camera Link Camera Link HS The choice depends above all on bandwidth, cable length, latency, number of cameras and system architecture.

GigE Vision is a machine vision standard based on Ethernet. Its advantages include: standard Ethernet infrastructure long cable runs good scalability multi-camera systems optional PoE With classic Gigabit Ethernet, up to roughly 100 m of copper cable per Ethernet segment is common.

These designations stand for Ethernet connections with higher nominal transfer rates than classic 1GigE. They allow correspondingly larger volumes of image data. Whether a particular camera supports these standards has to be checked in its datasheet.

PoE means power over Ethernet. A compatible camera can be supplied with both data and power over the same Ethernet cable. This reduces the cabling effort.

Jumbo frames are Ethernet frames with a larger payload than the usual standard MTU. They can increase the efficiency of transferring camera data. All the components involved - camera, network adapter and any switch - have to be configured accordingly.

Packet loss means that network packets are lost. With GigE cameras this can lead to incomplete or lost images. Possible causes are: overloaded NIC or switch buffers incorrect network settings unsuitable cables a data rate that is too high a host system that is not powerful enough Basler names overfilled network buffers during simultaneous data bursts from several cameras as a possible cause in particular.

USB3 Vision is a standardized machine vision protocol for USB 3.x based cameras. It is particularly suited to compact direct camera-to-PC connections and high data rates. The practical cable length is considerably shorter than with GigE and depends heavily on the cable type and system; active or optical solutions are used for longer runs.

CoaXPress is a machine vision interface developed specifically for high-performance image transfer. Over suitable coaxial connections it carries: image data control data trigger information with PoCXP, power as well CXP-12 supports 12.5 Gbit/s per link.

The number denotes the respective CoaXPress link rate. CXP-12 stands for up to 12.5 Gbit/s per channel, for example. Several channels can be combined for one camera.

PoCXP means power over CoaXPress. A compatible camera can be powered over the CoaXPress cable.

In a classic PC system, a corresponding acquisition or interface card is needed for CoaXPress, because normal PCs have no native CoaXPress interface. This card often also handles trigger, encoder and synchronization functions.

Camera Link is an established digital interface for industrial cameras. It normally requires a matching frame grabber. In new designs with very high data rates, more modern interfaces such as CoaXPress are often used, but Camera Link remains relevant in many existing installations.

GenICam stands for Generic Interface for Cameras. The aim is a largely uniform programming interface for cameras, independent of the underlying transfer technology. GenICam is used with GigE Vision, USB3 Vision, CoaXPress and Camera Link HS, among others.

GenTL is part of the GenICam ecosystem and standardizes the communication between the application and the transport layer in particular. This allows image acquisition devices and transport mechanisms to be addressed through a standardized interface.

A trigger is a defined event that starts a capture. Trigger sources can be, for example: a photoelectric sensor the PLC an encoder digital camera inputs a software command

With a hardware trigger, an electrical signal is fed directly to a camera input. This normally allows more precise timing than a command issued through the operating system and network in software.

A software trigger is issued through the camera software or API. It is practical for applications that do not need exactly deterministic timing tied to a mechanical process.

In free-run operation the camera captures images continuously at its configured frame rate. No external frame trigger is required.

That depends on the application. Triggered operation: when a particular workpiece has to be captured at exactly the right moment. Free run: when continuous capture makes sense and the evaluation then selects the relevant image.

Trigger delay is a defined delay between the trigger event and the start of the selected camera operation. It allows the capture to be timed to a workpiece position or to the lighting, for example.

Trigger latency describes the internal delay between a trigger event and the camera's actual response. It depends on the camera, the sensor mode and the function used.

Trigger jitter describes the variation of that delay from capture to capture. For high-precision synchronization, low and reproducible jitter is important.

On cameras that provide it, exposure active is an output signal indicating when the camera is actually exposing. This signal can be used to synchronize lighting, for example.

Ideally the lighting is only switched on during the relevant exposure window. A typical scheme is: trigger → camera → exposure/strobe signal → lighting controller With very short exposures, the signal delays of the camera and controller have to be taken into account.

Possible methods are: a shared hardware trigger a central trigger controller frame grabber synchronization IEEE 1588/PTP on suitable Ethernet cameras Which method is suitable depends on the timing accuracy required.

PTP stands for Precision Time Protocol. It synchronizes the clocks of several network devices with high accuracy and can therefore enable coordinated capture across several GigE cameras. The synchronization accuracy actually achievable depends on the hardware and the network.

FPS means frames per second. A camera rated at 60 fps can produce up to 60 complete images per second under the specified conditions. The frame rate actually achievable can be limited by: exposure time ROI pixel format interface bandwidth sensor readout

With individually triggered workpieces, the camera has to keep up with the production rate at least. At 100 parts per second, the entire system - capture, transfer and processing - has to be designed accordingly. Some technical headroom should also be allowed for.

Depending on the camera: data cable power supply I/O or trigger cable With PoE or PoCXP, the separate power supply can be omitted.

An I/O cable typically carries signals such as: trigger exposure active strobe digital inputs and outputs encoder The pin assignment depends on the camera.

Industrial environments place particular demands: EMC vibration bend radius drag chain tensile load locking temperature An office cable that works is therefore not automatically suitable for a permanently moving industrial installation.

A drag-chain-suitable cable is designed for repeated bending movements. The bend radius, travel and permissible number and type of movement cycles specified by the manufacturer have to be observed.

EMC stands for electromagnetic compatibility. A system should on the one hand not cause impermissible interference and on the other be sufficiently immune to external electromagnetic interference. Motors, frequency converters and power cables can interfere with vision systems, which is why shielding, earthing and cable routing matter.

M8 and M12 denote families of round industrial connectors. The designation alone does not say unambiguously which signals are carried. The number of pins, the coding and the pin assignment have to be checked.

Hirose is a connector manufacturer. With industrial cameras, “Hirose” is often used colloquially for certain multi-pin camera I/O connectors. The term alone is not a pinout specification.

RJ45 is the colloquial name for the typical modular Ethernet connector on many GigE cameras. In industrial environments, screw-locked or more robust variants are sometimes used.

USB 3 Micro-B is a USB connector used on many compact industrial cameras. Industrial cables may have additional screw locks.

Small changes in: focus camera alignment light angle working distance can change the image considerably. Mechanical stability is therefore part of measurement and process reliability.

Check in this order: Is the camera powered? Is the camera detected by the system? Correct driver or transport layer? Has acquisition been started? Trigger enabled but no trigger present? Lens cap removed? Exposure time sufficient? Is lighting present? For GigE: is the IP configuration correct?

Common causes: camera and NIC not on the same subnet firewall or network problem wrong network adapter cable power supply driver IP configuration The network setup should be checked systematically.

Check: light output exposure time aperture gain focus trigger strobe timing lens cap filter In strobe operation, check in particular whether the light pulse and camera exposure actually overlap in time.

Possible measures: reduce light output shorten the exposure time stop down the aperture reduce gain Saturated pixels lose image information and should be avoided where possible in measurement and inspection tasks.

Distinguish between: defocus → check focus, working distance and aperture motion blur → reduce the exposure time optical blur → check the lens resolution vibration → improve the mechanics

Common causes: too little light high gain very short exposure high temperature unsuitable sensor The best first measure is often to get more usable photons onto the sensor.

Possible causes: network bandwidth exhausted NIC or switch buffers overflowing host too slow unfavourable network parameters cable problem several cameras sending very large data bursts at the same time Basler explicitly documents buffer overflows as a typical cause of lost packets and frames.

Check: exposure time ROI pixel format sensor readout interface bandwidth trigger frequency camera settings host performance The maximum frame rate on the datasheet usually applies under defined operating conditions.

Check: correct trigger source trigger mode active correct edge electrical levels wiring trigger frequency Is the camera ready for the next trigger? debounce/delay

Possible causes: the lighting spectrum changes ambient light white balance auto gain auto exposure saturation lighting temperature For reproducible color measurement, the lighting and camera settings should be as stable and calibrated as possible.

mm/pixel = FOV [mm] / number of pixels

number of pixels = FOV × desired pixels per feature / feature size The desired number of pixels per feature depends on the application.

blur [mm] = speed [mm/s] × exposure time [s]

blur [pixels] = movement [mm] / object resolution [mm/pixel]

exposure time = permissible movement / speed

Approximation for uncompressed data: data rate = width × height × fps × bits/pixel / 8 Example: 2,448 × 2,048 50 fps 8 bit ≈ 250.7 MB/s of raw data Protocol overhead and the data format have to be taken into account in addition.

memory ≈ width × height × bits/pixel / 8 With packed formats, headers, padding or compression, the real figure can differ.

The selection should follow this order: area scan or line scan mono or color FOV smallest feature resolution required speed global or rolling shutter sensor size interface mount trigger/IO ambient conditions Only then should a specific camera model be chosen.

Without application data no serious model recommendation is possible. At minimum we need: field of view smallest feature object speed colour/mono working distance desired interface available space The current HIKROBOT portfolio can then be filtered accordingly.

Mono if color does not have to be evaluated. Colour if the color information is itself an inspection feature. For purely geometric, code and many surface inspections, mono is often the preferred starting point.

Fast movement or precise geometry in motion: global shutter Static or slow applications: rolling shutter can be sufficient.

GigE Good for: long cable runs several cameras distributed architecture PoE USB3 Good for: short direct connection to the PC compact installation high data rate without a network architecture Basler names cable lengths up to 100 m and multi-camera systems in particular as advantages of GigE/5GigE.

GigE simple PC connection long cables scalable moderate to high data rates CoaXPress very high data rates low latency precise triggering frame grabber high-end line scan and high-speed applications

What counts is the quality of the whole system of sensor, optics and light.

The real accuracy has to be determined for the complete system.

It matters above all with movement; rolling shutter can be the sensible choice for static applications.

Always use the real sensor dimensions for optical calculations.

Cable length, flash duration, overdrive, PPM, frame rate, reading distance and maximum data rate always depend on the specific product and operating state.