Industrial camera selection starts with the workpiece, the smallest feature or defect, and the speed at which it must be captured. Buying a camera by brand or megapixel count before defining those conditions can leave the lens, field of view, lighting, trigger, interface or mounting incompatible with the line. This guide shows how to turn a Thai factory’s inspection need into a camera specification, compare candidates using the same samples, and accept the system through FAT and SAT. It focuses on the imaging component; AI model training and detailed lighting design are covered separately.
Define the inspection task before selecting an industrial camera
Create one requirements sheet with production engineering, quality and IT before looking at catalogues. Record minimum and maximum workpiece dimensions, color and reflectivity, the smallest feature, variation in position and height, direction and speed of travel, takt time, working distance, available space, temperature, dust and cleaning conditions, image retention and machine I/O. “Find scratches” is too vague. State the scratch width, length, contrast, location and borderline between pass and fail, and supply real good, bad and limit samples. A low contrast surface flaw and a clear printed mark impose different image requirements.
Write down what a successful capture means: the whole target remains in the frame, the limit defect remains visible at full speed, and missing frames can be detected and logged during continuous operation. These are design hypotheses until tested with the workpiece and optics. A camera alone cannot rescue an inspection algorithm from unstable input images. Our inspection equipment selection guide covers the whole machine; this article specifies the camera capture subsystem.
Give the sheet an owner and revision number. If speed increases, package color changes or quality adds a new limit sample, reconsider pixels, exposure, lighting and transfer. The document must connect process conditions with acceptance criteria, rather than becoming a shopping list.
Calculate image resolution from field of view and minimum feature
“Is five megapixels enough?” cannot be answered without the field of view and the feature to be distinguished. Set the required horizontal and vertical field of view, including positioning variation and guides. Then test how many image pixels the smallest feature needs at its real contrast and with the intended decision method. As an illustrative assumption, a 100 mm horizontal field sampled by 2,000 pixels gives a nominal object-side sampling interval of 0.05 mm/pixel. A 0.10 mm feature landing on two pixels does not automatically mean reliable detection: lens resolution, focus, vibration, motion blur, lighting and demosaicing change what is visible. This arithmetic example is for requirements planning, not a performance claim for any camera.
Calculate the vertical direction too. A smaller region of interest may increase frame rate on some cameras, but it is useless if the target leaves that region. Include tolerances, positioning error and foreseeable product changes. Conversely, unnecessary pixels increase transfer, memory, storage and processing load. Estimate a resolution from the smallest feature, then use actual images to establish whether it can be distinguished. Basler’s camera selection guide likewise starts with application requirements, field of view and resolution.
Pixel count is also different from measurement accuracy. For an edge measurement, calibrate image coordinates to the workpiece and test distortion, error across the field and repeatability with a reference target. Presence detection does not need the same measurement accuracy as gauging a dimension. A tight tolerance cannot be justified by pixel pitch alone. Specify the minimum feature, allowable decision error and measurement uncertainty separately.
Match sensor size and lens as one optical system
Once sensor dimensions and pixel count are chosen, the lens must have a suitable image circle and resolving capability. A lens intended for a smaller sensor can produce poor edge quality or vignetting, hiding defects near the corners. Derive a focal length candidate from field of view and working distance, then check front clearance, enclosure, cable bend radius and adjustment hardware on the installation drawing. A matching mount, such as C-mount, does not itself guarantee a matching image circle or resolution. Basler’s lens selection guidance explains the relationship between sensor, field, working distance and lens.
Use real workpieces to check depth of field where height varies. A smaller aperture can extend depth of field but changes light intake and diffraction; compensating with a longer exposure can increase motion blur. Lens, light, shutter and line speed therefore have to be decided together. The layout and incident angle of illumination are covered in our visual inspection lighting article.
Consider a telecentric lens where dimensional measurement requires less change of magnification across object positions. Its physical size, working distance, cost and installation space have tradeoffs. A conventional fixed focal length lens may be entirely adequate; make the decision from an error budget and workpiece trials. Ask the supplier to state edge-of-field image quality at the specified field and distance, focus adjustment and the method of securing the setting, not merely a model number.
Global shutter versus rolling shutter: compare results at line speed
A global shutter exposes the image at the same time; a rolling shutter offsets exposure timing by sensor row. Depending on travel direction, readout direction, speed and light modulation, rolling capture may make a shape appear skewed, change an apparent dimension or create row-wise brightness differences. It should not be rejected automatically for a stationary, well controlled target. Basler’s Sensor Shutter Mode documentation is a starting point for checking how a specific model implements each mode. Verify the selected model rather than relying on a general label.
In a trial, image the same moving workpiece at actual line speed and compare with the stopped image, including both orientations of travel where relevant. Check skew in print, edge displacement, shape distortion and bands from modulated light. Long exposure can cause motion blur with either shutter type, so record exposure, light pulse, trigger time and vibration when investigating a problem. “Fast line means expensive shutter” is not a sufficient rule.
In the specification, state allowable geometric distortion and measurement error under maximum speed, orientation and lighting. If a supplier offers a different shutter mode, real images meeting those conditions can still be evaluated. The shutter is a means; the required captured image is the outcome.
Set frame rate from takt time and exposure from allowable blur
Frame rate describes how many images can be acquired; exposure is how long each image receives light. Confusing them may leave a high-fps camera producing blurred images. For one capture per passing part, draw a timing line through spacing, trigger delay, transfer, processing and reject actuation. Five parts per second does not imply that exactly five frames per second are sufficient; account for timing variation and whether more than one image is required. For continuous acquisition, define image overlap and retention.
A first approximation to motion blur is object-side speed multiplied by exposure time. If, purely as an example, travel is 500 mm/s and allowable blur 0.10 mm, simple division suggests 0.0002 s as an upper exposure limit. Actual testing must include speed variation, rotation, camera angle, light pulse and decision logic. Shorter exposure reduces received light. Do not simply raise gain without assessing light output and arrangement, aperture, sensor response, heat and strobe life.
Check what resolution, pixel format, ROI and interface apply to a catalogue’s maximum fps. Full-resolution color data and a narrow monochrome ROI differ greatly in transfer load. Estimate data volume from actual pixels, bit depth, fps and camera count, then measure effective host transfer and processing headroom at FAT. A table of fps numbers measured under different conditions is misleading.
Monochrome or color: follow the information required for the decision
Color is justified where the decision actually depends on color: a label, printing or an alien material of a different color. Then specify light spectrum and aging, white balance, chart-based calibration and stored image color space. Monochrome may suit edges, holes, shapes or presence of a mark where luminance suffices. Basler’s selection guidance also frames this as an application decision. Adding color to a luminance-only task can increase data and processing without helping the decision.
Do not assume every monochrome sensor is more sensitive or that any color camera can separate all color differences. Sensor generation, pixel size, filters and illumination spectrum change results. Image difficult real samples such as dark plastic, reflective metal and transparent packaging under the proposed light. Check whether the same threshold works across shifts and after replacing the light. If more product colors are likely, define how new references and settings will be approved rather than testing only today’s samples.
Compare GigE Vision and USB3 Vision as installed systems
The interface carries image data from camera to host. GigE Vision and USB3 Vision differ in practical choices around cable run, host layout, shared bandwidth, power and maintenance. Use Basler’s interface comparison as a starting point, but do not assign a guaranteed transfer rate or cable length from the interface name alone. Check the specific camera, cable, switch, NIC or USB controller, hub, power supply and environment against their manufacturer specifications.
For GigE Vision, examine simultaneous network traffic, switch and NIC settings, packet-loss monitoring and PoE power where used. For USB3 Vision, examine which host ports share a controller, cable retention and the components and power required for any extension. Run all cameras together and test dropped frames, latency, recovery and cable replacement. On moving axes, verify bend ratings and protection of the chosen cable. A successful single-camera bench demonstration is not enough.
If a future camera substitution matters, assess GenICam. The EMVA GenICam introduction describes its aim of offering a common programming interface across device types and transport interfaces. Compliance does not mean that every camera supports identical features, SDK behavior or trigger semantics. Connect each candidate to the actual application and exercise required functions. Treat GenICam as a basis for comparison and migration, not a guarantee of plug-and-play replacement.
Use EMVA 1288 data to compare image sensors fairly
Marketing terms such as “high sensitivity” and “low noise” are not a comparison method. EMVA 1288 defines common procedures to measure, calculate and present parameters of machine vision cameras and image sensors. The EMVA description explicitly explains its role in making manufacturer data sheets more comparable. Ask each bidder for available EMVA 1288 data and measurement conditions.
Those laboratory parameters are not an inspection pass rate. Whether enough light reaches the camera, whether the defect separates from its background, whether the lens uses the sensor’s capability, and how temperature affects an installed system all require actual images. Compare read noise, saturation capacity, dynamic range and other parameters relevant to the scene’s light and dark areas under like conditions rather than ranking one “sensitivity” figure. Mark absent data as unknown and ask the supplier; do not turn estimates into facts.
Give the comparison matrix columns for requirement, candidate A, candidate B, evidence method and decision. Separate facts verifiable from data sheets, such as resolution and shutter mode, from facts that require workpiece images, continuous running or on-site installation. A documentation pass is not automatically a production-line pass.
Mechanical, environmental and maintenance conditions in Thai factories
A climate-controlled electronics room and a hot, dusty machining line in Thailand have different needs. Measure the local minimum and maximum temperature, humidity, condensation, dust, oil mist, washdown liquids, vibration and stray light. Confirm the camera’s rated temperature and ingress protection, together with any housing, against the actual environment. There is no universal rule that every Thai installation requires IP67. A lens, connector, power supply, light or bracket can be the weak point even if the camera body is rated appropriately.
Review bracket stiffness, locking after adjustment, lens protection and places touched during cleaning. If a small relative movement changes field of view or measurement, prepare a reference target and calibration routine. Include replacement lead time, product life cycle, spares and configuration backup in the proposal. Document a fixed mounting position and recovery procedure so maintenance can replace a component without silently changing focus.
For equipment deployed at multiple sites, check differences in power, networks and ambient conditions. The same camera model may produce a different image under different wiring and lighting. At SAT, a discrepancy should first be tested against the supporting conditions written in the requirement sheet, before attributing it solely to the camera.
What an industrial camera RFP should contain
An RFP gives all suppliers the same process assumptions. Provide workpieces, inspection purpose, good/bad/limit samples, field of view, minimum feature, line speed and takt, camera count, installation drawing and environment. State image format and retention, trigger and result I/O to the PLC, host and network, security requirements and maintenance support. If sample handling requires confidentiality, state that at the outset.
Ask for a complete bill of materials: camera and sensor/shutter, lens, light, interface, cables, power, software and brackets. Request substitute parts, supply period, lead time, warranty, local support and backup method. Compare body-only price separately from the total cost of a working imaging subsystem. For each requirement, ask bidders to mark compliant, conditional or non-compliant, cite evidence and describe additional tests or design changes for conditional items.
Specify an image submission protocol: the same workpieces and position range, maximum speed, site-like lighting, raw images and full settings. One attractive demonstration photo is not evidence. Include normal, typical defective, borderline, highly reflective and dirty samples, and retain failed images. Joint review with the quality team moves the conversation from catalogue impressions to observable suitability.
FAT: compare candidates under agreed conditions
Factory acceptance testing should reproduce the RFP conditions at the supplier’s site. In addition to static parts, simulate maximum travel speed where feasible. Check visibility of the minimum feature, image quality at field edges, adjustment margin for exposure and gain, association between triggers and images, lost-frame detection and recovery during continuous running. For measurements, test calibrated references at the center and corners and at varied heights. For color decisions, test specified lighting variation.
Record sample revision and identifiers, camera and lens model, firmware, SDK, image format, exposure, gain, aperture, light output, working distance and travel speed. Keep raw images and logs as well as pass/fail summaries, exceptions and corrections. Agree the number of trials and allowable limits for the project’s risk before testing. Do not copy an arbitrary number from an article and call it an acceptance threshold.
Distinguish a comparison under identical settings from a comparison of each candidate optimized as a complete imaging system. The first helps isolate sensor differences; the second tells you what the proposed installation can actually achieve. Keep both results separate. Resolve non-conformities, replacement parts and conditions for retesting before shipment.
SAT: reproduce image quality on the actual production line
Site acceptance testing checks whether FAT conditions survive in the real installation. Include vibration, adjacent machinery’s light, cable routing, network traffic, temperature across shifts, cleaning and changeover. Production engineering, quality and IT should jointly test trigger-to-workpiece identity, reject timing, image storage and alarm behavior. FAT approval does not replace SAT.
Record any on-site adjustment and why it was needed. Stronger illumination, longer exposure, smaller ROI or more gain may change detection capability or comparability with FAT images. Compare the FAT reference set with SAT images and document the cause of differences. Continually compensating for mechanical changes by camera settings alone makes later fault analysis harder.
Hand over final configuration files, architecture drawing, sample images, reference target, calibration, cleaning and inspection intervals, recovery after replacement and log location. The objective is for the next maintenance team to diagnose falling image quality step by step. Periodic checks should use the same target and light conditions to separate focus, position, illumination and communication faults.
Common selection mistakes and how to avoid them
First, deciding from megapixels alone ignores lens quality and exposure. Second, judging only a static demo ignores motion, trigger timing, reflections and temperature. Third, comparing body price alone omits lens, lighting, wiring, host, mounts, testing, commissioning and support. Fourth, designing data transfer from theoretical interface maxima ignores host sharing and simultaneous operation. Fifth, testing only good samples misses the quality team’s limit defect. Sixth, postponing spare parts and recalibration instructions until after delivery makes recovery unpredictable.
These are failures in translating a process requirement into a capture specification, not proof that a particular brand is good or bad. Put “not measured” and “to verify at site” in the comparison sheet so uncertainty remains visible. The aim is a configuration that repeatedly acquires the required image and lets people investigate faults, rather than the most elaborate camera available.
Practical industrial camera selection checklist
Before quotations, assemble inspection purpose, minimum feature, field of view, speed, samples, environment, installation drawing and image retention. During comparison, verify resolution against field, lens image circle and corner quality, working distance, depth of field, shutter, monochrome or color, real exposure, required fps, interface capacity, trigger I/O and required GenICam functions. Before FAT, agree how raw images, settings and logs are saved, which limit samples are used and who owns correction and retesting. Before SAT, check cable path, protection, temperature, light, machine synchronization, network and maintenance handover.
This sequence preserves the decision criteria even if a model number changes: requirements, optical and timing calculations, actual workpiece tests, and reproduction at the site. For the larger system context, revisit our inspection equipment selection guide. For the effect of light on images used by AI inspection, see our lighting and image quality guide.
FAQ: industrial camera selection
How many megapixels does an industrial camera need?
No fixed number works without field of view and minimum feature. Estimate horizontal and vertical pixel needs, then image borderline samples. Measurement applications additionally need calibration and testing of error across the field.
Is global shutter always better than rolling shutter?
Global shutter is a strong candidate when movement-induced distortion matters. Rolling shutter can still be assessed for stationary, controlled targets. Compare images at maximum speed under actual exposure and lighting against allowable distortion.
Should I choose GigE Vision or USB3 Vision?
Decide from actual data load, cable route and distance, camera count, host, power and maintenance. Run the selected components and all cameras together, and test dropped frames and recovery rather than relying on an interface’s nominal maximum.
Does a better EMVA 1288 figure guarantee inspection success?
No. EMVA 1288 standardizes measurement and presentation of camera characteristics; it does not certify the image of your workpiece. Use FAT and SAT raw images under the proposed lens, light and line speed.
What is the difference between RFP, FAT and SAT?
RFP aligns requirements and bid assumptions; FAT verifies real workpiece images in the supplier environment; SAT verifies reproducibility and maintenance handover on site. Track the same criteria through settings, images, logs and changes.
Conclusion
Industrial camera selection is not a vote on megapixels, shutter names or interfaces. Define the smallest feature and field of view, design lens, light, exposure, movement and transfer as one capture system, and compare candidates with the same borderline samples. Standards such as EMVA 1288 and GenICam help comparison and connection but do not guarantee a production image. A clear RFP, raw-image FAT record and on-site SAT leave a decision basis that lasts longer than one camera model.
If you are beginning a camera shortlist or RFP for a Thai inspection line, contact TOMAS TECH with whatever workpiece images, minimum defect, field of view and line speed are already known. We can help structure what to test on the actual parts and how to define FAT and SAT acceptance.
