Technology // Overview

How a machine vision inspection system is built

The parts of an inspection and reject system, what each one does, and how it is chosen for your product and line, as we build stations for plants in South Africa.

Technical review: Frank GuoReviewed 25 September 2026

Discuss your line integration

// Short answer

A machine vision system for inspection is more than a camera. It combines a trigger, an industrial camera and lens, lighting designed around the defect, inspection software that turns the image into a pass or fail decision, a link to the line PLC, and a reject mechanism that removes failed products, ideally with a sensor that confirms each removal. Reliability depends on getting every link right on your own product and line.

// In this section

Technology pages

The platform we build on, how to choose HIKROBOT devices, and how a project is delivered and supported.

Platform

HIKROBOT VisionMaster integrationThe inspection software we build stations on, and what we deliver around it.

HIKROBOT selection guides

VisionMaster algorithm toolsWhich tool for which inspection: locating, measurement, codes and OCR, defects.
Smart camera selectionVision sensor, smart camera or vision controller, and how each is configured.
Code reader selectionID series, resolution, lighting and PLC protocols.
Typical configurationsSix station configurations and the checks each one fits.

Delivery & support

How we deliver a projectSix stages from requirement and sample test to handover.
Support & changesWho maintains the station after handover, and how changes are made.

// Architecture

The inspection chain, end to end

Most stations follow this chain; web and line-scan systems mark, map or stop the line instead of rejecting single parts. The numbers in the drawing match the component table below.

[Schematic]Block diagram · not to scale

Swipe sideways to see the whole drawing →

[1–3]: Capture: trigger, camera and lens, lighting

[4]: Decision: inspection software

[5]: Line PLC: tracking and SKU selection

[6–7]: Reject and confirmation

[8]: Records: reject images and counts

// Components

What each part does and what decides the choice

None of these is chosen from a catalogue alone. Each is confirmed on your samples and your line.

Components of a machine vision inspection system
Component	Role	What decides the choice	Typical question
[1] Trigger	Tells the station a product is in position.	Product spacing, speed changes, product shape and transparency.	Photo-sensor, or encoder for variable speed?
[2] Camera	Captures the image to be inspected.	Smallest defect vs field of view (resolution), speed, colour or monochrome, area-scan for discrete parts or line-scan for continuous material.	How many cameras to see every face that matters?
[2] Lens	Projects the scene onto the sensor.	Field of view, working distance, distortion; telecentric lenses where measurement must ignore small height changes.	Standard lens or telecentric?
[3] Lighting	Makes the defect visible and the image repeatable.	Surface (glossy, matte, clear), defect type; geometry such as front, low-angle, back, dome or coaxial light; strobing for fast lines.	Which angle separates the defect from the background?
[4] Inspection software	Turns the image into a decision and data.	Rule-based tools for defined faults; deep-learning tools for variable appearance defects; code reading and OCR.	Smart camera, PC or vision controller?
[5] PLC link	Delivers pass/fail and receives the running SKU.	Your PLC make and model and the protocols the chosen camera or controller supports.	Digital I/O or industrial network?
[6] Reject	Removes failed products from the line.	Product weight and stability, spacing, available air, space on the conveyor.	Air jet, pusher or diverter?
[7] Confirmation	Proves the failed product actually left.	Criticality of the defect and audit requirements.	What happens if a reject is not confirmed?
[8] Records	Keeps reject images, counts and results.	Traceability needs, storage, whether results go to MES or a dashboard.	How long are images kept, and who reviews them?

// Hardware

Machine vision cameras, lenses and sensors in plain terms

The hardware words that appear in quotations and datasheets, what each means, and when it matters for your station.

Machine vision hardware terms and when they matter
Term	What it is	When it matters
Machine vision camera (industrial camera)	A camera built for continuous use on a line, triggered by the station and read by software; it has no screen or buttons.	Always; its resolution and speed are chosen from the smallest defect and the line rate.
Area scan camera	Captures a whole rectangular image at once.	Separate products that pass one at a time.
Line scan camera	Captures one line at a time and builds a continuous image as the material moves.	Webs, sheet and fast, wide product; see web inspection.
HDR imaging (high dynamic range)	Bright and dark areas both usable in one image, by combining exposures or a sensor mode.	Shiny and dark areas in the same view, such as foil next to print.
Global shutter	Every pixel is exposed at the same moment.	Moving product: a rolling shutter, which exposes row by row, distorts fast-moving parts.
Frame rate	How many images per second the camera can take.	Fast lines and several images per product.
GigE camera (GigE Vision)	A camera that sends images over Ethernet using the GigE Vision standard, often with power over the same cable (PoE).	Cable lengths of tens of metres and several cameras on one PC.
Machine vision lens and C-mount	A fixed focal length lens; C-mount is the common screw mount on industrial cameras.	Field of view and working distance set the focal length.
Telecentric lens	A lens whose magnification does not change with small changes in distance.	Measurement where part height or position varies.
Photoelectric sensor (photo eye)	A light-beam sensor that detects that a product is present.	Triggering the camera; counting spaced products.
Encoder	A sensor on a conveyor shaft that reports distance travelled.	Variable speed, stop-start lines and line scan cameras.
Fixed barcode scanner (code reader)	A compact reader that decodes barcodes and 2D codes.	Code reading only; it does not inspect print or labels.
IP65 / IP67 enclosure	Protection against dust and water jets (IP65) or temporary immersion (IP67).	Washdown areas on food lines and dusty plants.

More terms in the machine vision glossary →

// Processing

Smart camera, vision sensor, PC or vision controller

Where the inspection software runs changes cost, capacity and maintenance. The software itself can stay the same.

Comparison of vision processing architectures
Architecture	What it is	Suits	Watch out for
Vision sensor	A compact camera with lighting and a small set of built-in tools that gives a pass or fail (HIKROBOT lists its SC1000, SC2000 and SC3000 series as vision sensors).	One simple check, such as presence or a pattern, from one view.	Few tools; not suited to many variants or several views.
Smart camera	Camera, lighting control and inspection software in one housing (HIKROBOT SC5000X / SC6000 run SC_VisionMaster).	One or two checks from a single view; compact stations.	Processing headroom for many tools or deep learning on one part.
PC with industrial cameras	Separate GigE Vision or USB3 Vision cameras connected to an industrial PC running VisionMaster.	Several cameras, heavier processing, operator screen and image storage.	PC hardening for the plant environment and its lifecycle.
Vision controller	A purpose-built industrial computer for multi-camera stations with camera ports and digital I/O.	Multi-camera stations that need a robust, cabinet-mounted unit.	Port and I/O count for later extensions.

HIKROBOT publishes a dedicated SC_VisionMaster build for its SC5000X and SC6000 smart cameras alongside VisionMaster for PC and vision-controller stations, and states that VisionMaster is compatible with the GigE Vision and USB3 Vision standards (HIKROBOT download centre and Machine Vision Smart Product Catalog 2025 Q4, both checked 24 September 2026). GigE Vision and USB3 Vision cameras expose a common programming interface through the GenICam standard hosted by EMVA.

// Decision tools

Rule-based tools and deep learning

Most stations use both. The choice is made per check, not per project.

Rule-based

Defined, measurable faults

Position, presence, dimensions, code reading and print verification against a reference. Predictable, explainable and quick to validate.

Works best whenThe defect can be described as a measurable rule

Deep learning

Variable appearance defects

Scratches, marks and contamination that vary in shape, or natural products whose appearance varies. Learns from labelled images of good and bad parts.

Works best whenEnough representative images of each defect exist

How to choose between rule-based and deep learning →

// PLC communication

Protocols depend on the model, not the brand

We never assume a protocol. We check the exact camera, reader or controller model against your PLC before quoting.

PLC communication options by product type
Product type	Protocols listed by the vendor	Source
VisionMaster software (PC)	TCP/IP, Modbus, serial port, UDP, EtherNet/IP and other common industrial protocols	HIKROBOT Smart Product Catalog 2025 Q4
Smart code reader (example: IDH7000 row)	SmartSDK, TCP, FTP, PROFINET, EtherNet/IP, Modbus, UDP, serial	HIKROBOT Smart Product Catalog 2025 Q4
Any station	Hard-wired digital I/O for trigger, pass/fail and reject	Standard practice

Protocol support differs between smart cameras, code readers and PC software, and between models. For the full integration method, see PLC & reject integration.

// FAQ

Technology questions

What is a machine vision camera?

A machine vision camera is an industrial camera made to be triggered by a machine and read by inspection software, not by a person. It is chosen by resolution, speed, sensor type, colour or monochrome, and interface such as GigE Vision or USB3 Vision.

What is a smart camera?

A smart camera is an industrial camera with a processor and inspection software built in, so it can make a pass or fail decision and talk to the PLC without a separate PC. It suits one or two checks from one view.

What is a vision sensor?

A vision sensor is the simplest camera-based device: a compact unit with lighting and a few fixed tools, such as presence, pattern match or count, that gives a pass or fail. It is easier to set up than a smart camera but has fewer tools.

What is a GigE camera?

A GigE camera sends its images over a Gigabit Ethernet cable using the GigE Vision standard. It allows long cables and several cameras on one PC or vision controller, and many models take power over the same cable.

What is a global shutter camera?

A camera whose sensor exposes all pixels at the same moment. Moving products stay undistorted, which is why global shutter sensors are the norm for inspection on moving lines.

What components make a reliable vision inspection and reject system?

A trigger that tells the station when a product arrives, a camera and lens matched to the smallest defect and the field of view, lighting chosen so the defect stands out, inspection software that turns the image into a decision, a link to the line PLC, a reject mechanism that removes failed products, and ideally a sensor that confirms each removal. A weakness in any one link, for example a reject that sometimes misses, undoes the rest.

Smart camera vs PC-based system: which should we use?

A smart camera suits one or two checks from one view and keeps the station compact. A PC or vision controller with separate industrial cameras suits several cameras, heavier processing such as deep learning, or stations that need an operator screen and image storage. HIKROBOT publishes a dedicated SC_VisionMaster build for its SC5000X and SC6000 smart cameras alongside VisionMaster for PC and vision-controller stations; whether a recipe moves between them is confirmed per project.

When are telecentric lenses needed?

Mainly for measurement. A normal lens changes the apparent size of a part when its distance to the camera changes; a telecentric lens keeps the magnification constant within its working range, so dimensions read correctly even if the part height varies slightly. Presence, label and code checks rarely need one. Whether your measurement does is decided during the sample test.

Do we need an encoder on the conveyor?

If line speed varies or the conveyor stops and starts, yes: the encoder lets the PLC track a failed product to the reject point by distance rather than time, so the reject still hits the right product, provided products move with the belt without slipping or accumulating. On lines that run at a constant speed, a time delay is often enough.

Can we use cameras from other brands?

VisionMaster is stated to be compatible with the GigE Vision and USB3 Vision standards, so cameras from other makers that follow those standards can be acquired (HIKROBOT Machine Vision Smart Product Catalog, 2025 Q4). Device control is most complete with HIKROBOT hardware. We confirm the exact combination for your station before quoting.

Sources

HIKROBOT Machine Vision Smart Product Catalog 2025 Q4 (PDF) (vendor document, checked 24 Sep 2026)
HIKROBOT VisionMaster product page (vendor page, checked 24 Sep 2026)
HIKROBOT download centre (VisionMaster, SC_VisionMaster) (vendor page, checked 24 Sep 2026)
EMVA — GenICam standard (standards body, checked 24 Sep 2026)

For machine vision inspection systems in South Africa, addaScan scopes the complete station around the product and production line. This page explains the component choices; the service overview connects applications, sample testing, budgets and delivery.

Typical HIKROBOT configurations →Choosing a HIKROBOT smart camera →Camera and lens selection →HIKROBOT VisionMaster integration →How we deliver a project →PLC & reject integration →All applications →Cost & budget guide →

// Line integration

Talk through your line with an engineer

Tell us the product, the check and your PLC. We will tell you which architecture fits and what has to be confirmed on site.

Request a Vision Assessment