A missed HGV, a cyclist counted as a motorcycle, or a queue recorded as free-flowing traffic can distort the decisions that follow. The best vehicle classification tools are therefore not simply those with the longest feature list. They are the systems that produce dependable, usable data in the real road environment, while fitting the installation, maintenance and safety constraints of the site.
For highways authorities, consultants and contractors, the choice has moved well beyond a basic vehicle count. Modern classification can support signal control, active travel planning, speed management, freight analysis, junction design and evidence-led road safety interventions. The technology must be assessed against those operational requirements, not selected solely on the basis of a headline accuracy figure.
What makes a classification tool the right one?
Vehicle classification identifies traffic by type, rather than merely detecting that an object has passed a point. Depending on the technology and configuration, outputs may distinguish cars, light goods vehicles, buses, rigid lorries, articulated lorries, motorcycles and, increasingly, cyclists and pedestrians. Some systems can also estimate vehicle length, speed, direction, lane occupancy, headway and queue characteristics.
The right level of detail depends on the decision being made. A rural speed reduction scheme may need reliable speed and vehicle-length data. A signalised urban junction may need real-time detection by lane, including cyclists, buses and turning traffic. A freight route assessment may require consistent differentiation between HGV classes over several weeks.
Accuracy matters, but it should be considered in context. A classifier may perform strongly in clear daylight yet need careful camera positioning to retain performance in darkness, glare, shadows or heavy rain. A radar may provide highly reliable speed and range data but offer less visual context when an unusual road layout must be investigated. The best specification accounts for the full operating condition, including traffic density, geometry, weather exposure, available mounting points and communications infrastructure.
Best vehicle classification tools by application
AI-powered video detection
AI video detection is often the strongest choice where a site needs rich classification data and visual understanding of road user behaviour. A properly configured camera can classify multiple road user types across several lanes and approach arms, while also measuring movements, occupancy, queue formation and conflict indicators.
This makes AI video particularly effective at complex junctions, pedestrian crossings, cycle routes, bus priority locations and urban networks where vehicle type alone does not tell the whole story. The ability to define detection zones in software is a practical advantage when lane use changes, a stop line is moved or a new cycle facility is introduced. It also avoids cutting into the carriageway, reducing disruption during installation and future changes.
The trade-off is that camera placement is critical. Mounting height, viewing angle, lighting, occlusion from high-sided vehicles and scene complexity all affect results. A camera should not be treated as a fit-and-forget device. Commissioning needs to validate detection zones against actual movements, and the system should be reviewed after major layout or signal staging changes.
Radar vehicle detectors
Radar is highly effective where dependable detection, speed measurement and vehicle profiling are required in difficult weather or low-light conditions. Above-ground radar can detect approaching and departing traffic over a defined range, making it well suited to rural approaches, speed management schemes, queue detection, traffic-actuated signals and locations where a camera view would be compromised.
For classification, radar systems commonly use measured length, speed and signature characteristics to group vehicles into useful categories. This can provide strong operational data for separating cars from larger vehicles, monitoring freight movements or identifying slow-moving queues. Radar is also a practical answer where privacy considerations limit the use of visual imagery.
Its limitation is one of detail rather than reliability. A radar-based classification may not always distinguish every vehicle body style required for a highly granular survey. Where the brief requires differentiation between several HGV configurations, or the detection of cyclists and pedestrians alongside vehicles, combining radar with another sensing method or using AI video may be more appropriate.
Wireless and above-ground traffic sensors
Wireless traffic sensors are valuable where civil works must be kept to a minimum and the objective is targeted data collection or detection. They can support vehicle counts, direction, speed and classification, depending on the sensor type, and are useful for temporary surveys, works monitoring and locations where installing loops would create disproportionate disruption.
Their main benefit is deployment speed. There is no requirement to cut slots into the carriageway or reinstate the road surface after installation. This reduces lane closures, exposure of installation teams to live traffic and the risk of future failure caused by road movement or resurfacing.
However, a shorter-term or compact sensor solution should still be assessed for its data capacity, battery life, communications method and suitability for the required vehicle classes. It is a good fit when the survey question is clearly defined. It is less suitable when a scheme requires continuous, detailed multi-modal intelligence at a complicated junction.
Pneumatic tubes and inductive loops
Pneumatic tubes remain familiar tools for temporary traffic surveys. They can provide counts, speed and axle-based classification, and may be appropriate for straightforward, short-duration studies on lower-risk roads. Their limitations are practical: tubes are vulnerable to damage, require carriageway installation and can be affected by traffic behaviour, multiple axles and lane discipline.
Inductive loops have long been used for stop-line detection and signal actuation. They can offer dependable presence detection, but installation and replacement involve cutting the carriageway. That brings traffic management, reinstatement requirements and a maintenance burden that above-ground alternatives are designed to avoid. Loops also provide limited classification capability unless used as part of a more specialised arrangement.
Neither method is automatically wrong. For a defined legacy estate or a simple one-off survey, they may remain proportionate. But when authorities are seeking adaptable detection, quicker delivery and reduced roadworks, non-intrusive radar, video and wireless technologies generally offer a more sustainable route.
Match the tool to the traffic decision
A useful specification starts with the action the data will support. If the goal is to retime signals around peak-period demand, the system needs lane-specific detection, reliable presence information and outputs that can integrate with the controller strategy. If the goal is to understand rat-running, count duration, direction, classification confidence and time-stamped data become more important.
For road safety, consider the vulnerable road users at the site. A vehicle-only classifier may be insufficient near a school, town centre crossing or cycle route. AI video can identify cyclists, pedestrians and turning interactions, enabling a more complete assessment of exposure and behaviour. Radar can provide an excellent complementary measure where approach speeds and early detection are central to the intervention.
Freight studies need a different lens. Ask whether the system can distinguish vehicle lengths and classes consistently, whether it can identify lane and direction, and whether the output can be aggregated by hour, day and route. A generic ‘large vehicle’ category may not answer a question about articulated lorry movements through a constrained settlement.
Specify the data, validation and maintenance plan
The most effective procurement documents define the required outcomes before naming a technology. State the vehicle classes to be reported, the lanes and movements to be covered, the minimum reporting intervals, the expected operating conditions and whether the data is for real-time control, a temporary survey or long-term monitoring.
It is also sensible to define how performance will be tested. Validation should compare classifier outputs with a representative manual or video-reviewed sample across the traffic conditions that matter at the site. A short test during quiet daytime traffic is not enough for a busy urban junction, a school peak or a wet winter evening. Agreeing the validation method early avoids ambiguity at handover.
Data ownership and accessibility deserve equal attention. The platform should make it practical to inspect raw and aggregated data, export the fields needed for analysis, identify device health issues and retain evidence for scheme evaluation. A traffic counter that collects data but makes it difficult to interrogate will not deliver its full value.
Finally, consider maintenance in operational terms. Above-ground equipment is easier to access than road-embedded detection, but it still needs a clear support arrangement, secure mounting, communications monitoring and periodic review of detection performance. C & T Technology’s approach is to pair non-intrusive detection with the technical guidance needed to ensure the selected system works for the road, rather than only on a product datasheet.
The strongest classification project begins with a precise question: what traffic behaviour must be understood or changed? When that question is clear, the appropriate sensing technology, data outputs and validation plan become far easier to specify – and the resulting evidence is far more likely to improve the network.