Computer vision · Vehicle traffic
Every crossingbecomes a number.
Traffic analysis with computer vision means detecting, classifying and counting vehicles in a camera's images, using lines and zones set in advance. We build systems that show how many cars, vans or lorries pass a point, in which direction and at what times, without reading licence plates and without identifying drivers.
Events
- 08:30:17Line L1 · 2 in, 2 out
- 08:30:17Classes: car, truck, motorbike
- 08:30:18No identification · counts and types only
- 08:45:0015-minute interval saved to the report
Without data, access decisions are made by eye
How busy a gate is, when the queue builds up or how many lorries come in each day is often estimated from impressions or from manual counts done rarely. Simple in-road sensors count crossings, but they do not tell you what kind of vehicle passed or which way it went.
- Manual counts are done occasionally, over short periods, and are hard to repeat
- Nobody knows what share of traffic is lorries, vans or cars
- Peak hours and queues at the gates are only noticed once they become a problem
- Solutions that read licence plates are too much for a simple need for statistics
Who it is for
- Logistics parks and industrial sites that want to understand lorry flows at the gates and on internal roads
- Managers of shopping centres, campuses or office buildings with their own access roads
- Contractors and site managers who track machinery and haulage traffic
- Consultants and technical teams who need traffic data for studies, planning or reorganising access
Step by step
How a vehicle ends up in the traffic report, step by step
The five steps on the animation follow one crossing: from the raw camera image, through detection, classification and tracking, to the counting line that turns it into a number and the person who checks the totals.
Events
Events
- 08:30:17Line L1 · 2 in, 2 out
Events
- 08:30:17Line L1 · 2 in, 2 out
- 08:30:17Classes: car, truck, motorbike
Events
- 08:30:17Line L1 · 2 in, 2 out
- 08:30:17Classes: car, truck, motorbike
- 08:30:18No identification · counts and types only
- 08:45:0015-minute interval saved to the report
Events
- 08:30:17Line L1 · 2 in, 2 out
- 08:30:17Classes: car, truck, motorbike
- 08:30:18No identification · counts and types only
- 08:45:0015-minute interval saved to the report
Step 1 of 5
The camera image
The best view is from above and slightly in front, with every lane in the picture and vehicles clearly separated from one another. A camera mounted low or almost parallel to the road lets a lorry hide the car beside it. We check footage by day, by night and at peak times, because headlights and backlighting change a great deal of what can be seen.
Technical options and when each fits
Line counting
One virtual line per lane or access point, with crossings counted in both directions. It is the simplest and most robust option, suited to gates, access roads and car park entrances.
Direction matrix
Several entry and exit zones at a junction or fork, with the system recording where each tracked vehicle came from and where it went. It suits traffic studies and access reorganisation, but needs a view that covers the whole junction.
Processing at the camera or on a server
An edge device mounted next to the camera sends only the counters, which reduces network traffic and avoids moving images around. A local server or the cloud makes sense when there are many cameras on the same network or when the footage needs reviewing anyway.
What you receive at hand-over
- The detection and classification model, fine-tuned on images from your measuring points and exported for the chosen hardware
- The lines, zones, directions and aggregation intervals configured for each camera
- Integration of the counters with the database, dashboard or spreadsheet where you want to see the data
- Notifications for queues or congestion, where they were requested
- A measurement report on footage from your site, comparing the automatic count with a manual one by direction and category
- Documentation for changing the lines and zones and, typically, a short training session for the team that checks the data
No strings attached: tell us what the cameras should see and we'll tell you honestly whether it can be done.
Let's talkWhat the traffic analysis system can measure
- 01
Counts crossings over lines
Vehicles crossing a virtual line are counted separately for each direction of travel, for every lane or access point.
- 02
Classifies the vehicle type
The model tells apart the categories we agree together, for example cars, vans, lorries, buses, motorcycles or bicycles.
- 03
Tracks directions of travel
At a junction or a fork, the system can count how many vehicles go from one direction to another.
- 04
Flags queues and congestion
When the number of vehicles in a zone exceeds a set threshold, for example at a gate, the person responsible gets a notification.
- 05
Estimates time spent in a zone
For a waiting area, the system can measure how long a vehicle stays on average, without knowing which vehicle it is.
- 06
Produces aggregated reports
Counts are grouped into intervals of minutes, hours or days and reach dashboards or spreadsheets, without keeping images unnecessarily.
How it works
From camera image to traffic report
The steps marked in yellow are approved by a person.
- Input
- AI
- Action
- Human in the loop
- Result
- 01
Input
Camera stream
A camera facing the road or gate sends the image.
- 02
AI
Detection and classification
The model finds the vehicles in the frame and estimates the category of each.
- 03
Action
Tracking and rules
Each vehicle is tracked while it crosses the frame, and line crossings are added to counters.
- 04
Human in the loop
Spot-check review
A team member regularly compares the counts with video clips reviewed by hand.
- 05
Action
Aggregation by interval
Counters are grouped by hour, day, direction and category.
- 06
Result
Traffic report
You get totals and trends, without data that identifies vehicles or people.
What data and camera placement are needed
- Video footage from the measuring point, at peak and quiet times, by day and by night
- The list of vehicle categories that matter and the rules for telling them apart, for example where a van ends and a lorry begins
- A definition of the lines, zones and directions to be counted
- A camera mounted high enough, with a clear view of the lanes, to reduce vehicles hiding one another
- Access to the camera stream (for example via RTSP or ONVIF, where the equipment supports them) and processing resources: edge device, local server or cloud
Possible integrations
- IP cameras with an RTSP stream
- ONVIF-compatible cameras
- NVRs and VMS systems
- NVIDIA Jetson edge devices
- PostgreSQL databases
- Power BI or Grafana dashboards
- Exports to Google Sheets or Microsoft Excel
- MQTT
- REST API and webhooks
- Email or Microsoft Teams notifications
- IP cameras with an RTSP stream
- ONVIF-compatible cameras
- NVRs and VMS systems
- NVIDIA Jetson edge devices
- PostgreSQL databases
- Power BI or Grafana dashboards
- Exports to Google Sheets or Microsoft Excel
- MQTT
- REST API and webhooks
- Email or Microsoft Teams notifications
Each integration's compatibility is checked during the analysis phase, for the systems and versions you use.
Control and transparency
What stays with people
- Setting the vehicle categories, lines and zones, and reviewing them when traffic patterns change
- Regular spot checks of the automatic counts against a manual count
- Interpreting the data and the decisions based on it, for example reorganising access or the delivery schedule
- Confirming that filming has a legitimate purpose, that it is signposted and that images are kept only as long as necessary
Limits and operating conditions
- In heavy traffic, vehicles that hide one another can be miscounted; a camera mounted higher reduces the problem but does not remove it
- Similar categories, for example a large van and a small lorry, are harder to tell apart, especially from side angles
- At night, in rain, fog or against the light, detection quality drops; headlights can hide the shape of the vehicle
- The system produces operational statistics; it is not type-approved equipment for official measurements or for detecting traffic offences
How we measure the accuracy of traffic analysis
We agree the indicators together before launch and track them afterwards. The values depend on the process, the data and the volume.
- Counting error by interval
- The difference between the automatic total and a manual count on the same video footage, over intervals chosen by the team
- Classification accuracy
- The confusion matrix between vehicle categories, on an independently labelled sample from the site
- Accuracy by direction
- Crossings checked separately for each direction and each movement, to see whether errors cluster in one place
- Stability across conditions
- Results compared separately on footage taken by day, by night, in rain and at busy times
How we implement traffic analysis, in 4 steps
- 01
We agree the points and the questions
We choose where to measure, which categories matter and which decisions will be based on the data.
- 02
We check the cameras and the angle
We review real images and recommend, if needed, repositioning or an additional camera.
- 03
We calibrate on real data
We compare the automatic counts with manual counts and adjust the lines, zones and model.
- 04
We connect the reporting
The aggregated data reaches the chosen dashboard or system, and quality is monitored over time.
Use cases
The scenarios below are conceptual examples, designed to illustrate realistic applications. They are not implementations delivered by WEB-AZ.
- Logistics
Lorry flow at the gate of a logistics park
A camera at the gate counts the lorries and vans coming in and going out, by hour. The site manager sees when the queue builds up and can discuss staggering some deliveries with tenants.
Conceptual example - Construction
Haulage traffic on the site access road
Heavy vehicles entering the site are counted by day, and the data goes into a table checked by the site manager and used to plan deliveries.
Conceptual example - Retail
Entrances to a shopping centre car park
Cameras at the car park entrances count vehicles at each access point and in each direction, so the management can see which entrances are overloaded at weekends.
Conceptual example
Useful guides
Before you decide, read
Frequently asked questions
What is AI vehicle traffic analysis?
It is a computer vision application that detects vehicles in a camera's images, classifies them into categories and counts them using set lines and zones. The result is totals by interval, direction and vehicle type, used for planning and operational decisions.
Does the system read licence plates?
No. Traffic analysis works with anonymous vehicles: each one is tracked only while it crosses the frame and then remains only in the totals. Licence plates are personal data, and reading them is a separate solution with its own assessment of purpose and proportionality.
Which vehicle categories can it tell apart?
The categories are agreed together, based on your needs, for example cars, vans, lorries, buses, motorcycles and bicycles. The more two categories look alike, the more labelled examples and the better camera placement are needed.
Can I use an existing CCTV camera?
Sometimes. Cameras mounted low or at a sharp angle to the road often give poor results, because vehicles hide one another. We review real images before recommending whether to use them or add a camera.
Can it measure vehicle speed?
An approximate speed estimate is possible after careful camera calibration, but it remains indicative information for analysis. The system is not type-approved equipment and cannot be used for penalties or official measurements.
Is the video footage kept?
Not necessarily. The system can process the stream and keep only the counters. If images need to be kept for review, we agree the retention period and access together, in line with your video surveillance policy and data protection law.
Can’t find the answer? Write to us and we’ll reply directly.
Want to know what passes through your gate?
Tell us which points you want to measure and which decisions depend on this data. We review the cameras, angles and filming conditions and tell you what can realistically be measured.
Related solutions
- People and object countingAnonymous counting of crossings, zone occupancy or the items in a batch, without identifying anyone.
- Licence plate recognitionAutomatic reading of licence plates at gates, barriers and car parks, with clear rules on access, retention and human review.
- Car park monitoringOccupancy of every parking space or zone, read from cameras, with free-space counts kept up to date and no identification of drivers.