Practical selection guide

Selecting a machine vision camera: from field of view to interface

A camera is suitable only when the smallest relevant detail remains visible at the required speed, working distance and product variation. This guide shows how to determine field of view, resolution, shutter type, lighting, lens, frame rate and data interface step by step.

Connecting and configuring an industrial machine vision camera

First architecture decision

Choose between area scan and line scan first

An area-scan camera captures a complete rectangular image for each trigger and usually suits discrete products or a defined inspection area. A line-scan camera builds the image one line at a time as material moves and suits webs, sheets, film and very wide or long objects. This choice determines synchronization, lighting and how required resolution is calculated.

Compare area scan and line scan

Compare product motion, image geometry, encoder use, lighting and mechanical installation before selecting a sensor format.

Selection sequence

Six decisions before comparing camera products

Translate the production requirement into measurable inputs first. This prevents resolution, interface or camera type from being selected in isolation while the complete imaging chain determines the actual limit.

Step 1

Define the decision

Describe what the software must detect, measure, read or classify. Also define which errors matter and which borderline cases require human review.

Step 2

Measure field of view and working distance

Record the required width and height in millimetres, the available distance to the product and variation in product height or position.

Step 3

Determine the smallest feature

Use the smallest feature that must support the decision. Then establish through testing how many pixels are required across that feature.

Step 4

Define motion and cycle time

Record product speed, trigger frequency, available stop time and permitted motion blur. These inputs determine shutter type, exposure time and frame rate.

Step 5

Test lens and lighting

Check contrast, distortion, depth of field and reflections with representative products. Resolution without sufficient optical contrast does not produce a dependable feature.

Step 6

Check integration and ownership

Determine interface, cable length, bandwidth, power, trigger I/O, ingress protection, software, recipe management and maintenance responsibility.

Camera type and processing

Determine which camera family performs the task

Use product motion and required depth information as the first boundary. Then decide where image processing should run and how much integration the machine requires. The area scan versus line scan page explains the 2D geometry decision in more detail.

Area-scan camera

For a complete 2D image per trigger. Suitable for presence, position, dimensional checks and defect inspection within a defined field of view.

Line-scan camera

For continuously moving material or objects whose length and width require a very large 2D image. Motion synchronization is part of image formation.

Smart camera

For acquisition and processing in one housing. Suits defined inspection tasks where local configuration, I/O and limited machine integration are sufficient.

3D camera

For height, depth, shape or volume when contrast in a 2D image does not provide enough information. Measurement range, surface and required uncertainty determine the 3D method.

Calculation method for area-scan cameras

Calculate the required image scale and sensor resolution first

Use the same units throughout each formula. The result is a lower bound for the initial shortlist. Optical contrast, algorithm, positioning and product variation must then be tested with images.

1. Image scale

Divide the size of the smallest relevant feature by the number of pixels required across it. That count is not a fixed industry standard: establish it with the selected algorithm and representative samples.

mm/pixel ≤ feature size (mm) ÷ required pixels

2. Sensor resolution

Divide the width and height of the field of view separately by the permitted millimetres per pixel. Round up and allow space for positional variation outside the nominal inspection area.

pixels ≥ field of view (mm) ÷ mm/pixel

3. Motion blur

The distance travelled by the product during exposure, divided by image scale, gives blur in pixels. Strobe lighting can shorten the effective exposure time when the installation supports it.

blur (px) = speed (mm/s) × time (s) ÷ mm/pixel

Worked example

A 0.5 mm feature in a 240 × 160 mm field of view

Inputs for this example

Field of view: 240 × 160 mm.
Smallest relevant feature: 0.5 mm.
Test criterion: at least 4 pixels across this feature.
Product speed: 500 mm/s; no more than 1 pixel of motion blur.

Calculation

Image scale: 0.5 ÷ 4 = no more than 0.125 mm/pixel.
Horizontal: 240 ÷ 0.125 = at least 1920 pixels.
Vertical: 160 ÷ 0.125 = at least 1280 pixels.
Exposure time: 1 × 0.125 ÷ 500 = no more than 0.00025 s, or 250 µs.

The calculation does not yet select a camera model. Also check the sensor, lens, working distance, available light and detail reproduction at the image corners.

Check frame rate and data throughput

The camera must follow the highest trigger frequency and read each image in time. Protocol overhead, image metadata, multiple cameras and storage require capacity beyond the raw data stream.

MB/s ≈ width × height × bit depth × frames/s ÷ 8,000,000

Complete imaging chain

Camera, lens and lighting must make the detail visible together

Calculated resolution is useful only when the lens transfers sufficient detail and contrast and the lighting separates the relevant feature from its background. Assess sensor format, lens mount, working distance, depth of field, distortion, lighting direction and required exposure time as one design.

Industrial machine vision lenses for different sensor formats

Lens and working distance

Relate field of view, sensor format, focal length, working distance and permitted distortion.

Lighting and contrast

Choose geometry, colour, polarization and, where required, strobe operation based on the surface and inspection feature.

Field of view and resolution

Calculate millimetres per pixel from the smallest feature and the required number of pixels across that feature.

Trigger, shutter and timing

Fit trigger delay, exposure, readout and processing within the available cycle time.

Technical boundaries

Compare camera characteristics only after the calculation

These choices affect acquisition quality, cabling, PC load and maintenance. Assess them under machine conditions, not from a data sheet alone.

Global shutter or rolling shutter

Shutter type determines whether all image rows record the same instant. This becomes visible as soon as the product or camera moves during acquisition.

Global shutter: all pixels start and stop exposure at nearly the same time. This is usually the safer choice for moving parts and accurate geometry.
Rolling shutter: image rows are exposed sequentially. This may suit stationary objects, but motion can distort shape and position.

Monochrome or colour

Choose colour when colour information is required for the decision or helps separate the feature with controlled lighting.

Monochrome often suits dimensional checks, edges, codes and contrast analysis.
Colour classification requires control of illumination colour, white balance and colour variation in the product and environment.

GigE, USB3 or higher bandwidth

Select the interface based on throughput, cable length, number of cameras, PC connections and service method.

GigE often suits longer cable runs; check bandwidth, power and network configuration.
USB3 offers high bandwidth over shorter runs; secure the connector, cable quality and controller capacity.
10GigE or CoaXPress may be needed for high resolution and frame rate; also assess the frame grabber, PC slots and heat dissipation.

Sensor format, pixel size and lens mount

More pixels may mean a larger sensor or smaller pixels. Both change the requirements for the lens, lighting and mechanical installation.

The lens must cover the complete sensor format without unacceptable dark corners or loss of detail.
Smaller pixels require sufficient optical resolution and light; additional pixels do not compensate for blurred optics.
Check lens mount, available space, focus adjustment and the range of aperture or autofocus.

Processing architecture

Decide where images are processed and managed

A smart camera processes images locally. An industrial camera with a PC provides more scope for multiple cameras, extensive logic, data storage and custom interfaces. Classical vision suits features that can be described with fixed rules; AI becomes relevant when variation cannot practically be captured in those rules.

Review the vision architecture decisions

From calculation to a tested camera setup

Use the calculation for an initial shortlist. Then test representative good, poor and borderline samples with the camera, lens and lighting. Sedeco can compare multiple combinations and provide a component recommendation with technical considerations for implementation.

Compare camera families Discuss a feasibility test

Acceptance test

Define when the imaging chain is suitable

Test the complete setup with representative good, poor and borderline samples. Use the actual product distance, speed, shielding, ambient light and trigger source.

Agree a measurable limit and test scope for each criterion. A general camera resolution or demonstration image is not an acceptance criterion for the production process.

Detail and contrast

The smallest relevant feature remains discernible throughout the permitted field of view and across all approved product variants.

Focus and distortion

Height and positional variation remain within the agreed depth of field; lens distortion stays within the measurement or localization tolerance.

Motion and timing

Blur stays below the agreed pixel limit and the camera handles the highest trigger frequency without missed or mismatched images.

Lighting

Contrast remains usable across the agreed variation in surface, contamination, ambient light and light-source ageing.

Communication

Trigger, result, error status and product association work with the PLC or PC, including agreed behaviour on timeout or loss of connection.

Recovery and ownership

After restart or recipe change, the correct settings are loaded. Operators and maintenance staff can recognize and handle documented faults.

Business-case inputs

Calculate process value using production data

The camera is only part of the investment. Include optics, lighting, PC or controller, mounting, shielding, cabling, software, engineering, validation, training and maintenance.

Required production-line data

Number of products or inspections per shift and number of production days.
Current time spent on manual inspection, setup, recording and reinspection.
Cost of missed defects, false rejects, scrap and rework.
Downtime and operator interventions caused by camera, lighting, trigger or communication faults.
Time and materials for changeovers, periodic cleaning, calibration and replacement.

Simple calculation model

Annual process cost = manual hours + cost of defects and rejects + downtime + maintenance. Compare the current process and the tested camera setup using the same volumes, rates and acceptance limits.

After defining the camera

Develop lens and lighting with the same product samples

A different lens, lighting direction or exposure time can change the usable image more than additional megapixels. Finalize the camera only after detail, contrast, depth of field and motion have been tested together.

Go to the optics and lighting guide

Concepts and camera families

Explore the selected camera technology

Use these articles when you need to understand how a specific camera family works and where its limits lie. Then return to the calculation and test method on this page to compare models.