What is vehicle counting and classification in video surveillance?

What it measures

Vehicle counting and classification combines two functions. Counting tallies how many vehicles cross a virtual line or occupy a defined zone over time. Classification identifies what kind of vehicle each is — passenger car, two-wheeler, three-wheeler / auto-rickshaw, bus, light commercial vehicle, or heavy truck — and usually adds direction of travel, lane, and an approximate speed estimate. The output is per-class volumes and trends that quantify how a road, junction, gate, or facility is actually used.

How it works

• Detection and tracking: the model detects each vehicle and tracks it across frames to avoid double-counting.
• Line and zone logic: counts increment when a tracked vehicle crosses a tripwire or enters a counting zone.
• Classification: a vision model assigns a vehicle class and can capture colour and direction.
• ANPR pairing: linking to automatic number plate recognition ties each count to a readable plate for enforcement and evidence.
• Output: dashboards, APIs, and reports feed traffic-management, BI, and command-centre systems.

Where it is used

Vehicle counting and classification supports traffic planning and junction design, congestion and signal optimisation, toll and parking automation, axle/class-based tolling, fleet and logistics yard management, and enforcement when paired with ANPR and traffic-violation analytics. For a city it converts the existing camera estate into a continuous, objective traffic-survey instrument rather than relying on periodic manual counts.

How VMukti delivers it

VMukti provides vehicle counting and classification as one of its 26+ AI models, running over any ONVIF camera across 1,000+ supported models, at the edge for low-latency gate and lane decisions or in the cloud for citywide aggregation. It works alongside ANPR — which reads plates at speeds up to 200 km/h with 95%+ accuracy across Indian, US, EU, and Gulf formats — and traffic-violation detection, all feeding the city Integrated Command and Control Centre (ICCC). The analytics are proven across smart-city and transportation deployments within VMukti's 900+ projects processing 1B+ camera feeds annually.

Accuracy considerations

Counting and classification accuracy depend on camera height, angle, and field of view, and on handling occlusion in dense traffic. VMukti tunes camera placement and counting geometry per site, and its in-house R&D team can train classification on local vehicle types and signage where the standard model needs adaptation.