Machine Vision Components: Where to Invest and Where to Save

Not every component in a machine vision system deserves the same share of your budget. Some components have a direct, measurable impact on system performance. Others are important but offer diminishing returns beyond a certain quality threshold. Knowing the difference is what separates a well specified system from an expensive one.

This is not about buying cheap. It is about buying smart. The goal is to invest where it makes the biggest difference to image quality and system reliability, and to save where the specification allows it without compromising performance.

Invest more: lighting

If there is one message in this article, it is this. Lighting is the most underestimated and under-budgeted component in machine vision. Engineers routinely spend weeks choosing a camera and minutes choosing the lighting, then wonder why the software cannot reliably detect the features they need.

A good lighting setup creates contrast between the feature you want to detect and the background. A poor lighting setup creates an image where the software has to work harder, run more complex algorithms, and tolerate more noise to find the same feature. The downstream cost of poor lighting is not just a worse image. It is more development time, more fragile algorithms, more false rejects, and more production stoppages.

The price difference between adequate and excellent lighting is often a few hundred pounds. The difference in system performance can be enormous. If you are going to overspend anywhere, overspend here.

Invest more: lenses (when measurement matters)

For inspection applications where you are making dimensional measurements or where edge detection accuracy is critical, the lens matters more than most people think. A good lens with a budget camera will outperform a good camera with a budget lens, because the lens determines the optical quality of the image before the sensor ever sees it.

Telecentric lenses are essential for measurement applications where perspective distortion is unacceptable. They are more expensive than standard fixed focal length lenses, but for measurement work, there is no substitute.

Where you can save on lenses is in applications where the tolerance is generous and the field of view is straightforward. A standard C-mount lens from Computar or Kowa will perform well for the majority of general inspection tasks. You do not need a telecentric lens to check whether a label is present.

Save smart: cameras

This might be surprising, but cameras are one of the areas where thoughtful specification can save significant money without sacrificing performance. Modern CMOS sensors are genuinely excellent across the range, and the differences between cameras at different price points are often about features you may not need, rather than fundamental image quality.

Here is a real example. Suppose you need a 5 megapixel camera with a global shutter for inspecting a moving object. Even within a single manufacturer's range, the options and costs vary widely:

 

Camera	Sensor	Pixel Size	Sensor Size	Interface	FPS	Price
BFS-PGE-50S4M-C	Sony IMX547	2.74µm	1/1.8"	GigE	30fps	~£460
BFS-PGE-50S5M-C	Sony IMX264	3.45µm	2/3"	GigE	33fps	~£702
BFS-U3-51S5M-C	Sony IMX250	3.45µm	2/3"	USB3	75fps	~£1,000
FG-P5G-51S4M-C	Sony IMX547	2.74µm	1/1.8"	5GigE	122fps*	~£1,200

*207fps in burst mode with advanced firmware features

 

All four cameras are 5 megapixel with global shutters. All four will produce a good image. The difference is speed, interface, and features. If your application runs at 20 parts per minute, you do not need 122fps. The £460 camera will do the job. If you are inspecting at high speed on a fast production line, the additional investment in the 5GigE model is justified.

The sensor size also affects your lens cost. The cameras with a 1/1.8" sensor use smaller, less expensive lenses than those with a 2/3" sensor. Choosing the smaller-sensor camera when your resolution allows it saves money on both the camera and the lens.

The principle: specify what you actually need, not what sounds impressive on paper. A camera that exceeds your requirements is not a better camera. It is a more expensive one.

Invest more: software (usually)

Commercial machine vision software platforms cost more than open-source alternatives, but the total cost of ownership calculation almost always favours the commercial option for production systems. When something goes wrong at 2am on a production line, commercial software comes with a support team. Open-source comes with a forum.

Platforms like Zebra's Design Assistant also allow non-software engineers to build and maintain applications using flowchart-based programming. This reduces your dependency on specialist software developers and means the people closest to the production line can support the system themselves. That long-term self-sufficiency is worth paying for.

Where open-source tools like OpenCV make sense is in R&D and prototyping, where flexibility matters more than support, and in organisations with strong internal software teams who can maintain the codebase indefinitely.

Invest more: industrial PCs

This is a short section because the answer is simple. Do not use consumer PCs in production environments. They overheat, they lack real-time I/O, and they are not rated for continuous operation. The price difference between a consumer PC and an industrial equivalent with fanless cooling, real-time I/O, and appropriate environmental ratings is a few hundred pounds. The cost of a production line stopping because the PC overheated is considerably more.

Invest more: specification and feasibility

The cheapest component in a machine vision system is always the engineering time spent getting the specification right before you buy anything. A feasibility study that costs a few thousand pounds can save tens of thousands in avoided mistakes. A system that was properly specified works the first time. A system that was guessed at gets redesigned, often more than once.

Clearview's ClearviewFormula methodology and Insights test lab exist for exactly this purpose. For a detailed breakdown of what machine vision systems cost at different complexity levels, see our pricing page.