Computer vision
Your images tell youmore than you think.
For manufacturing, warehouses, logistics and retail: we turn video streams and photos into useful signals, such as a missing object, a blocked area or a product with a visible defect. The system flags it, and people decide what happens next.
The cameras you have can do more than record
Many companies already have cameras installed, but the footage is usually viewed only after something has happened. Meanwhile, visual checks are done by hand and by spot-checking.
Hours of video, hard to use
Recordings pile up, but finding a particular event means scrolling through them by hand, minute by minute.
Quality control by spot-checking
On a fast line, the human eye gets tired. A visible defect can go unnoticed for an entire shift.
Inventory counted by hand
Pallets, parcels or products on the shelf are counted manually, with discrepancies that only surface at stocktaking.
Areas that need constant attention
Blocked access routes, restricted areas or loading bays are checked only when someone happens to walk past.
Processes without data on the real pace
You don't know exactly how long a lorry waits at the loading bay or where bottlenecks form, because nobody has time to time it.
Vision models trained for your task, not for everything
Computer vision means software that analyses images and video to recognise objects, states or events. It works well on clearly defined tasks, under stable filming conditions. Lighting, angle, resolution and obstacles in the frame directly affect the results, so no detection is perfect, and the system is designed to flag, not to decide on its own.
Object detection and counting
Models that recognise the objects relevant to you, from pallets to products or vehicles, and count them in an area or along a route.
Visual zones and rules
We draw the areas of interest on the image and define what counts as an event: entering an area, standing too long, an object left in the way.
Visual quality inspection
Based on examples of good and defective products, the model flags the items worth checking by an operator.
Alerts and a monitoring dashboard
Detected events arrive as notifications, with the associated image, and trends are tracked in a simple dashboard.
Local or cloud processing
Where appropriate, the analysis runs on an Edge AI device next to the camera, so the images don't leave the site.
Computer vision applications
Each application has its own image, infrastructure and privacy requirements. We always start with an assessment on real images.
Object detection
Models that recognise the objects that matter to your process in images and pass them on as structured data.
DetailsPeople and object counting
Anonymous counting of crossings, zone occupancy or the items in a batch, without identifying anyone.
DetailsIndustrial visual inspection
Cameras on the production line flag parts with visible defects, and the operator confirms every rejection.
DetailsVideo event detection
Visual rules that flag entries into restricted zones, missing protective equipment or fallen objects, for human review.
DetailsVideo analytics for retail
Aggregated visitor flows, queue length and empty shelves from in-store cameras, without identifying customers.
DetailsOperational monitoring
Visual tracking of process stages and of vehicle and pallet movement in yards and warehouses, so you can see where the flow gets stuck.
DetailsEdge AI for computer vision
Video analysis on devices installed on site: fast response, less network load and images that never leave the building.
DetailsLicence plate recognition
Automatic reading of licence plates at gates, barriers and car parks, with clear rules on access, retention and human review.
DetailsVisual smoke and fire detection
An additional warning layer that spots smoke and flames in camera images and quickly alerts the people who can check.
DetailsFall detection
Cameras that notice when a person falls and does not get up, so an alert quickly reaches someone on the team who can help.
DetailsDrone detection
Cameras and models that spot drones appearing over a site, track them in the frame and alert the security team.
DetailsContainer code recognition
Automatic reading of ISO 6346 codes on containers, with check digit validation and direct transfer to the terminal system.
DetailsNameplate reading
You photograph the nameplate of a motor or piece of equipment, and its data reaches your inventory or ERP sorted into fields.
DetailsCrack detection
Models that find and outline cracks and surface defects in photos and drone imagery, ready for an engineer to review.
DetailsCar park monitoring
Occupancy of every parking space or zone, read from cameras, with free-space counts kept up to date and no identification of drivers.
DetailsLivestock monitoring
Cameras and AI that count animals, follow activity in the barns and flag to the farmer the behaviour worth a closer look.
DetailsVehicle traffic analysis
Anonymous counting and classification of vehicles by direction, at gates, on access roads or on building sites, with data aggregated by interval.
DetailsVideo synopsis
Hours of footage condensed into a short summary of activity, searchable by object type, zone and time range.
DetailsCustom computer vision models
We develop computer vision models for your task end to end: data, labelling, training, evaluation, optimisation, deployment and retraining.
Details
What you can gain when images are analysed continuously
- 01
Faster reaction to events
An alert sent the moment an area becomes blocked can shorten the time until someone steps in.
- 02
Constant checking, not spot-checking
Automated inspection can analyse every product that passes through the frame, and operators focus on the flagged cases.
- 03
Data on real flows
Counts and measured times can give a more accurate picture of the pace of work and the bottlenecks.
- 04
Making use of existing cameras
Where image quality allows, the video infrastructure already in place can be used for new tasks.
Results are not guaranteed. They depend on image quality, lighting conditions and camera angles, the quantity and quality of training data and how well the task is defined.
Situations where computer vision can help
The examples below are conceptual scenarios, designed to illustrate realistic applications. They are not implementations delivered by WEB-AZ.
- Manufacturing
Flagging visible defects on the line
A fixed camera above the conveyor photographs each part, and the model marks scratches or missing labels for manual checking.
Conceptual example - Warehouses
Counting pallets at goods-in
At the unloading bay, the system counts the incoming pallets and compares the total with the delivery note, flagging any discrepancies.
Conceptual example - Logistics
Dwell time at the loading bay
The camera records vehicle arrivals and departures, and the dashboard shows average times and peak periods.
Conceptual example - Retail
Empty shelves and visitor flow
The system flags shelves that need restocking and anonymously counts footfall in certain areas, without identifying people.
Conceptual example
What the project can include
We choose the modules based on what needs to be observed, where the cameras are and who receives the information.
Detection and counting
- Detecting the objects relevant to the process
- Counting objects, or people without identifying them
- Detecting defined visual events
- Monitoring areas of interest
Quality and processes
- Visual quality control
- Monitoring industrial processes
- Measuring times and the pace of work
- Collecting images to train the model
Alerting and reporting
- Alerting by email, SMS or messaging apps
- Analysis of live or recorded video streams
- A dashboard with events and statistics
- Data export to other systems
Infrastructure
- Integration with existing video infrastructure
- Edge AI where appropriate
- Recommendations on positioning and lighting
- Assessing whether additional cameras are needed
From image to working system
Step 01
We define what needs to be seen
We clarify the event or object of interest, what happens after detection and what a useful result means for you.
Step 02
We check the cameras and conditions
We analyse real images from the site: resolution, angle, lighting throughout the day, obstacles. We confirm whether or not the equipment is compatible.
Step 03
We assess the legal framework
We establish the purpose of processing, how filmed people are informed and the retention period and, where necessary, carry out a data protection impact assessment together with you.
Step 04
We collect data and train the model
We gather and label representative examples from your site, then train or adapt the model for the defined task.
Step 05
A pilot in a single area
We run the system on one camera or one line, compare the detections with reality and adjust the alert thresholds.
Step 06
Expansion and maintenance
After the pilot, we expand to other areas and track performance over time, because changes in lighting or layout may require retraining.
We start from the equipment you have
In many situations we can take the video stream from your existing cameras, but each piece of equipment needs to be checked. Some of the technologies we can assess:
We do not guarantee compatibility with every camera or device. Compatibility, image quality and hardware capacity are checked for each project.
- IP cameras with RTSP
- ONVIF-compatible cameras
- NVR and DVR
- VMS systems
- NVIDIA Jetson edge devices
- Industrial mini PCs
- Industrial cameras for production lines
- Local or cloud storage
- REST API and webhooks
- MQTT
- Email and SMS notifications
- Telegram, Slack or Microsoft Teams
- Power BI or Grafana dashboards
Security & control
Images processed with restraint and transparency
Video can capture people, and that brings clear obligations. We design the system on the principle that only what is necessary for the agreed purpose is processed.
Data minimisation
Where possible, the system works only with the relevant part of the frame, blurs faces or keeps only the events, not the whole video stream.
Limited retention and access
We establish how long images are kept and who can see them, in line with your video surveillance policy and the applicable legislation.
Impact assessment where required
Systematic monitoring of premises or employees may require a data protection impact assessment (DPIA) and clear information for the people concerned.
No biometric identification by default
Facial recognition and other forms of biometric identification are not part of the standard offer. Any such request requires a separate assessment of legality, necessity and risks.
Useful guides
Before you decide, read
- Computer visionComputer vision without facial recognition: applications in manufacturing, logistics and retailAnonymous counting, visual inspection, event detection: what computer vision can do without identifying people and which conditions need to be met.Read the guide
- ComplianceThe AI Act explained for businesses: what the EU regulation on artificial intelligence requiresThe AI Act sorts AI systems into risk levels and sets different obligations for providers and users. What it means in practice for a business that uses chatbots or AI agents.Read the guide
Frequently asked questions
What is computer vision and what is it used for in a business?
Computer vision is a branch of artificial intelligence that analyses images and video to recognise objects, states or events. In a business it can count pallets or products, flag visible defects on a production line or a blocked area and send an alert, and a person decides what happens next.
Can I use the security cameras I already have?
It is possible, but it depends on resolution, angle, streaming protocol and image quality in real conditions. We check each camera before confirming anything.
How accurate is the detection?
Accuracy depends on the task, the filming conditions and the training data. No detection is perfect, which is why we measure the results in the pilot and set the system up to flag cases, not to make final decisions.
Do you offer facial recognition?
Not as a standard feature. Biometric identification raises serious legal and proportionality issues and requires a separate assessment of legality, necessity and risks. In most cases, the goal can be achieved without identifying people.
Do the images go to the cloud?
Not necessarily. Where appropriate, processing can be done locally on an edge device, and only events or statistics leave the site. We choose the architecture based on the data, cost and confidentiality requirements.
How many images are needed for training?
It depends on how varied the objects and conditions are. Sometimes an existing model can be adapted with a few examples; other times, collection and labelling sessions at your site are needed.
Can I monitor employees with this system?
Employee monitoring is strictly regulated and must be justified, proportionate and communicated. We recommend scenarios that track processes and objects, not people, and a legal review before any use of this kind.
What affects the cost of a project?
The number of cameras and areas, the complexity of the task, the volume of data to label, the hardware needed and the processing mode, local or cloud. After analysing a concrete case we can provide an estimate.
Can’t find the answer? Write to us and we’ll reply directly.
Tell us what your cameras should be watching for
Describe what needs to be observed and what video equipment you have. We will assess whether the task suits computer vision and what conditions would need to be met.
Related services
- AI automationAutomated workflows between the apps you already use, with AI where it helps and human review where it matters.
- Custom software & integrationsApplications, platforms and integrations built around your processes, so data flows between systems on its own.
- AI agentsAI assistants that follow the steps you set, use only approved tools and complete concrete tasks, with a person making the decisions.