#Industrial Visual Inspection Solution: A One Stop Combination Guide from Camera to Capture Card
>Target keywords: industrial camera solution, machine vision inspection solution, production line vision system construction
>Topic: 1 | Date: June 12, 2026 | Author: CC-4


At the end of last year, a factory that produces bearing outer rings approached us and said they spent over 300000 yuan on a visual inspection system, but after three months, the detection rate was less than 80%. Let's go to the scene and see that the problem is not with the algorithm - their cameras, lenses, and capture cards are products from three different suppliers, each with its own parameters, and when they come together, they fight. After changing to a unified plan, the detection rate was directly increased to 99.5%.
This case is not an isolated case. The most difficult aspect of implementing industrial visual inspection is not how to select a single parameter, but how to assemble the camera, lens, light source, acquisition card, and transmission link into a "coordinated" system. As long as there is a bottleneck between each link, the performance of the entire solution will be stuck there.
This article does not discuss single-chip selection (which has already been covered in the article "Industrial Camera Selection: A Complete Decision Framework from Parameters to Scene"), but focuses on how to combine these things together when facing a specific detection requirement.

##1、 The 'four piece set' for industrial visual inspection: Without any one, it won't work
A complete production line visual inspection system requires four core components:
Camera (imaging) → Acquisition card/interface (transmission) → Industrial computer (processing) → Execution mechanism (action)
↑                ↑                ↑              ↑
Camera+light source determines bandwidth limit algorithm+computing power elimination/alarm

Many people only focus on camera parameters, but overlook that the capture card is the bandwidth bottleneck of this link. A 5-megapixel global shutter camera generates approximately 150MB/s of data at 30fps, while the actual effective throughput of a gigabit Ethernet port is only around 115MB/s. No matter how strong the camera is, it is useless if the data cannot be transmitted.
So the first thing in industrial visual design is not to choose a camera, but to draw a data flow diagram: how much data each camera generates → what interface is used to transmit it → industrial control functions cannot process it.

##2、 Product combinations of three typical scenarios
###Scenario 1: Conventional size inspection+appearance defects (accuracy 0.05-0.1mm)
*Requirement portrait *: Check whether the dimensions of the parts are qualified, and whether there are scratches, indentations, and burrs on the surface. The production line speed is moderate (30-60 pieces/minute), and the workpiece size does not exceed 150mm.
Recommended combination:
|Step | Selection | Reason|
|------|------|------|
|Camera | 5-megapixel global shutter black and white camera | Global shutter is required for motion scenes, ensuring accuracy in black and white. 5-megapixel camera is sufficient to cover a 150mm field of view with a precision of 0.05mm|
|Lens | C-port telecentric lens (0.5-1x magnification) | Telecentric lens eliminates perspective distortion and is essential for size measurement|
|Light source | Circular light+backlight dual channel | Circular light for surface defects, backlight card size contour|
|Capture card | Gigabit Ethernet port (GigE Vision) | Single camera scene bandwidth sufficient, flexible wiring for 100 meters|
|Industrial control computer | i5+8GB memory+SSD | Conventional algorithms have sufficient computing power|
**Key points of combination: In this scenario, the acquisition card is not the bottleneck (a gigabit network can withstand 5 million black and white 30fps data), the key is to adjust the light source correctly. We have tested that the detection rates of the same 0.03mm scratch on the same camera with ring light and backlight are 62% and 98%, respectively - the choice of light source has a greater impact than camera resolution.
The global exposure camera solution (8081 series) from Shiduan Wei, paired with a high-definition capture card (EM38180 solution), achieves a measured repeatability accuracy of ± 0.005mm in size detection scenarios with a precision requirement of 0.05mm.

###Scenario 2: Online inspection of high-speed production lines (speed priority, 60-300 pieces/minute)
**Requirement Profile: High speed scenes such as filling lines, packaging lines, stamping lines, etc., where workpieces quickly pass through inspection stations and require high frame rate capture.
Recommended combination:
|Step | Selection | Reason|
|------|------|------|
|Camera | 1.3-megapixel global shutter, frame rate ≥ 120fps | High speed scenes do not require high resolution, they require frame rate|
|Lens | M12 fixed focus lens | Rapid deployment, controllable cost, and non extreme precision requirements|
|Light source | High brightness strip light/coaxial light | High speed exposure requires strong light compensation, and dark field environments rely on high-power light sources|
|Acquisition card | USB 3.0 (≤ 2 units) or dual gigabit network card | USB 3.0 effective bandwidth of 380MB/s, capable of withstanding high frame rate data streams|
|Industrial control computer | i7+16G memory+independent GPU | GPU acceleration required for high-speed processing|
**Key points of combination: In high-speed scenarios, the bandwidth of the acquisition interface is the first hurdle. For example, a 2-megapixel black and white camera running at 120fps produces a data volume of 2 million x 1 byte x 120fps ≈ 240MB/s. A gigabit network is not sufficient and must be connected to USB 3.0 or dual network cards for data splitting.
More importantly, the matching between exposure time and motion speed is crucial. A simple formula: exposure time ≤ minimum detection accuracy ÷ object motion speed. For example, the accuracy requirement is 0.1mm, the workpiece movement speed is 500mm/s, and the exposure time must be ≤ 0.1/500=0.2ms (i.e. 200 μ s). If the exposure time is not enough, a higher power light source is needed to compensate - this is a chain reaction.
Based on the global exposure sensor, the industrial camera solution of Shiduan Wei can achieve an exposure time as low as 10 μ s, coupled with a high brightness light source, to form stable and non trailing images in high-speed conveyor belt scenes.

###Scenario 3: High precision measurement (accuracy 0.005-0.02mm)
*Requirement profile *: Measurement of dimensions for shaft components, precision hardware, and optical elements, with extremely strict tolerance requirements.
Recommended combination:
|Step | Selection | Reason|
|------|------|------|
|Camera | 10-20 million pixel black and white, pixel ≤ 2.2 μ m | High resolution pixel reduction ensures single pixel accuracy|
|Lens | C-port telecentric lens (1-4x magnification)+low distortion | distortion<0.05%, precise measurement of baseline|
|Light source | Far center parallel backlight source | Eliminate edge diffraction, clear contour|
|Acquisition card | USB 3.0 or 10GigE | High resolution and large data volume require high bandwidth|
|Industrial control computer | i7+32GB memory+professional GPU | High resolution image processing requires large computing power|
*Example of bandwidth calculation *: A 100 megapixel black and white camera with a measurement tempo of 15fps.
-Data volume: 10 million x 1 byte x 15fps=150MB/s
-USB 3.0 (380MB/s effective bandwidth) has sufficient margin, while gigabit network (115MB/s) is just stuck at the boundary
-Conclusion: USB 3.0 or 10GigE is recommended for this scenario
The 4K solution of V-Well, combined with high-resolution industrial cameras, provides a complete image acquisition chain in precision measurement scenarios, with a unified solution from sensors to acquisition cards to avoid compatibility loss between modules.

##3、 How to calculate the bandwidth of the collection card? (One formula to solve it)
This is the most practical skill in visual design - calculating the amount of data and deciding which interface to use.
Formula: Required bandwidth (MB/s)=horizontal pixels x vertical pixels x frame rate x color coefficient
-Black and white camera color coefficient=1 (1 byte per pixel)
-Color coefficient of color camera=3 (3 bytes per pixel, RGB)
Actually calculate once:
Rule of thumb: The final effective bandwidth of the interface should have at least 30% margin. If an interface runs at a utilization rate of 90%, the risk of latency and frame loss will sharply increase.

##4、 List of stages that need to be covered by a complete plan
Pull it from beginning to end to ensure there are no omissions:
Camera selection (shutter speed+resolution+frame rate+black and white/color)
□ Lens selection (focal length+image plane+distortion+interface)
Light source scheme (type+angle+wavelength+power)
□ Collection interface (bandwidth calculation+transmission distance+multi camera splitting)
□ Industrial computer configuration (CPU+memory+GPU+storage)
□ Trigger synchronization (triggered by photoelectric sensors/encoders, synchronized by multiple cameras)
□ Software platform (image acquisition SDK+algorithm framework+result statistics)
□ Implement linkage (PLC communication+exclusion/sorting+data recording)
□ Environmental protection (dustproof and waterproof+heat dissipation+anti vibration installation)

The triggering of synchronization is often overlooked. Simultaneous shooting with multiple cameras requires hardware trigger signal synchronization, otherwise the two cameras will capture images of the workpiece at different times, resulting in direct errors in subsequent stitching and comparison.

##5、 Why is it not recommended to 'save' your own computer
Returning to the case at the beginning of the article. The root cause of the problem at that bearing factory is the "accumulation thinking" - buying cameras, capture cards, and software from three suppliers separately, hoping that they can work seamlessly together. The actual result is:
-The driver of the camera is incompatible with the SDK of the capture card, resulting in a random ± 15ms delay jitter every time it is triggered
-The lens image size does not match the camera sensor, and the edges of the image appear dark
-The light source controller does not support the camera's trigger signal and can only remain on. The lifespan of the light bulb has been reduced from one year to three months
The value of a unified solution is not that the parameters are higher, but that each link has been validated to work together. What you received from a supplier was a 'usable system', not a 'bunch of parts that need to be adjusted by yourself'.

##Summary
The essence of industrial visual inspection solutions is not to match parameters, but to match system integration capabilities. A good plan will leave enough margin between each link -30% bandwidth, 3 times resolution, and 20% light source power. These margins are the confidence that the system will run on the production line for three years without crashing.
Next time you do visual design, why not first draw a data flow diagram, marking the bandwidth and latency of each link, and then see where the bottleneck is. This habit is more useful than remembering a hundred parameters.

V-Well Technology provides complete solutions for industrial cameras, capture cards, and embedded modules, supporting full process technical services from selection to deployment. To learn more about the industrial visual product portfolio, please visit videowell-work