Why Camera Selection Matters More Than You Think

In any machine vision system, the camera is the front end of the data pipeline. Every downstream algorithm, every measurement, every pass/fail decision depends on the quality and suitability of the image that the camera delivers. Select the wrong camera and no amount of software tuning will rescue the application. Select the right one and the rest of the system design becomes significantly easier.

Yet camera selection is routinely treated as a secondary decision. Engineers often start with a familiar model or whatever the integrator has on the shelf. That approach works until it does not — and the failure modes tend to show up late in commissioning, when changing the camera means re-engineering the optics, the mounting, the lighting geometry, and possibly the communication architecture.

This guide walks through the key parameters and trade-offs involved in choosing an industrial camera for a manufacturing vision application.

Sensor Technology: CCD vs. CMOS

For years, CCD sensors dominated industrial vision because of their superior image uniformity and low noise characteristics. That landscape has shifted dramatically. Modern CMOS sensors have largely closed the gap in image quality while offering higher frame rates, lower power consumption, and significantly lower cost. The global shutter CMOS sensors now available from Sony (Pregius and Pregius S families) and onsemi deliver the distortion-free imaging that manufacturing applications demand.

Unless you are maintaining a legacy system that already uses CCD cameras, CMOS is the practical choice for new applications. The key specification to watch is whether the sensor uses a global shutter or a rolling shutter. For any application where parts are moving — which covers most production-line scenarios — a global shutter is essential to avoid image distortion.

Resolution: How Much Is Enough

Resolution selection should be driven by the smallest feature you need to reliably detect or measure, not by a desire for the sharpest possible image. The rule of thumb in industrial vision is that the smallest critical feature should span at least 3 to 5 pixels in the image. For measurement applications requiring sub-pixel accuracy, 7 to 10 pixels across the feature is a safer target.

Work backward from there. If your field of view is 200 mm wide and the smallest defect you need to catch is 0.1 mm, you need at minimum 200 / 0.1 × 5 = 10,000 pixels across the sensor width. That puts you in the range of a 10-megapixel or higher sensor. If that resolution drives up cost, processing time, or data bandwidth beyond your budget, consider whether you can reduce the field of view and use multiple cameras, or whether a line-scan architecture makes more sense.

Higher resolution always comes with trade-offs: larger file sizes, longer transfer times, more processing overhead, and typically higher camera cost. There is no benefit to over-specifying resolution. A 2-megapixel camera running at 100 frames per second will outperform a 20-megapixel camera struggling to hit 15 frames per second in most production environments.

Area Scan vs. Line Scan

Area scan cameras capture a complete two-dimensional image in a single exposure, much like a digital photograph. They are the default choice for most inspection stations where parts are presented in a defined position, either stationary or moving through a trigger zone.

Line scan cameras capture one row of pixels at a time and build up an image as the part moves past the camera. They excel in applications involving continuous web materials (paper, film, textiles, steel strip), cylindrical part inspection, and any scenario where the object is much longer than it is wide. Line scan cameras also provide extremely high resolution across the scan direction without the file-size penalty of a correspondingly large area scan sensor.

The decision between the two usually comes down to the nature of the part and the production flow. If parts arrive on a conveyor with consistent spacing and orientation, an area scan camera with a triggered acquisition is simple and effective. If you are inspecting a continuous surface or need to image the full circumference of a round part, line scan is the right architecture.

Frame Rate and Exposure

Frame rate determines how many images per second the camera can deliver. For production-line inspection, the required frame rate is driven by line speed and part spacing. If parts arrive every 500 milliseconds and you need one image per part, a 2 fps camera would technically suffice — but you want significant margin. A camera capable of 30 fps or more gives you headroom for double-triggers, retries, and future line-speed increases.

Exposure time is linked to frame rate but is a separate parameter. Shorter exposures freeze motion but require more light. This is where camera selection intersects directly with lighting design. A camera with high quantum efficiency (the percentage of photons converted to electrons) will produce usable images at shorter exposures, reducing motion blur without requiring extremely high-intensity lighting. This matters in applications with fast-moving parts or where heat-sensitive products cannot tolerate high-intensity illumination.

Interfaces: Getting the Data Out

The interface between the camera and the processing system is a practical constraint that shapes the entire system architecture.

GigE Vision is the most widely deployed interface in industrial vision. It uses standard Ethernet cabling, supports cable runs up to 100 meters, and allows multiple cameras on a single network. Bandwidth tops out around 1 Gbps for standard GigE and 10 Gbps for 10GigE, which is sufficient for most applications up to about 5 megapixels at 30 fps.

USB3 Vision offers higher bandwidth (up to 5 Gbps) with simpler cabling, but cable lengths are limited to about 5 meters without active extenders. It is a good fit for compact inspection stations where the processing PC is close to the camera.

Camera Link and CoaXPress are high-bandwidth interfaces designed for demanding applications. Camera Link supports very high data rates but requires dedicated frame grabber cards and specialized cabling. CoaXPress delivers up to 12.5 Gbps per cable over coaxial connections up to 40 meters, making it the current choice for high-resolution, high-speed applications.

For most new installations, GigE Vision strikes the best balance of bandwidth, cable flexibility, and ecosystem support. Reserve Camera Link and CoaXPress for applications where data rates genuinely demand them.

Environmental and Mechanical Considerations

Industrial cameras operate in environments that are hostile compared to a lab bench. Temperature extremes, vibration, dust, moisture, and electromagnetic interference are all factors that can degrade image quality or shorten camera life.

Look for cameras with IP-rated enclosures if they will be exposed to washdown, coolant mist, or airborne particulates. Verify the operating temperature range — many industrial cameras are rated to 50°C, but installations near ovens, welding stations, or in non-climate-controlled facilities can exceed that. Mounting provisions matter too: a camera with an integrated C-mount or CS-mount lens adapter and standard mounting holes simplifies mechanical integration.

Vibration is an underestimated issue. A camera mounted on a frame that vibrates at the frequency of a nearby press or conveyor drive will produce blurred images even with a global shutter and short exposure. Mechanical isolation or a remote camera head connected by fiber may be necessary in severe cases.

Matching Camera to Application

Rather than starting with camera specifications, start with the application requirements and work backward:

  1. Define the inspection task — what features, defects, or measurements must the system deliver
  2. Determine the field of view — the area the camera must image in a single acquisition
  3. Calculate the required resolution — based on the smallest critical feature and the 3-to-5-pixel rule
  4. Establish the speed requirement — parts per minute, line speed, or cycle time
  5. Assess the environment — temperature, contamination, vibration, available space
  6. Select the interface — based on bandwidth needs, cable length, and processing architecture

This structured approach prevents the common mistake of selecting a camera and then trying to make the application fit around its limitations.

Integration With the Broader Vision System

A camera does not operate in isolation. Its performance is inseparable from the lens, lighting, mounting, and processing system surrounding it. Integrating deep learning algorithms for defect classification, for example, may shift the emphasis from raw resolution toward consistent image quality and frame-to-frame repeatability that lets the neural network train effectively.

The best camera selection decisions happen when the camera is evaluated as part of a complete vision system, not as a standalone component. That means conducting feasibility testing with representative parts, realistic lighting conditions, and the actual processing pipeline before committing to a production camera specification.

Partner With AMD Machines

AMD Machines has designed and deployed vision systems across automotive, medical device, electronics, and consumer products manufacturing. Our engineers select and specify cameras as part of a complete system design, ensuring that every component — from sensor to software — works together to meet your production requirements. Contact us to discuss your vision application.