A queue that appears to extend 120 metres on a traffic operator’s screen may be 70 metres in reality, or substantially longer if vehicles are creeping, stopping irregularly or occupying multiple approach lanes. That difference affects every decision made from the data: signal timing, incident response, scheme appraisal and public reporting. Knowing how to measure queue lengths accurately therefore starts with a clear definition of what is being measured, rather than simply counting stopped vehicles.
For highways authorities and traffic signal teams, queue measurement should provide an operationally useful view of demand and delay. It must remain reliable across changing light levels, weather, traffic mix and approach geometry, without creating a maintenance burden or requiring intrusive carriageway works.
Define the queue before measuring it
A queue is not always a line of stationary vehicles. On a congested urban approach, vehicles may move at walking pace between signal cycles. At a roundabout, traffic may be continuously circulating but still be constrained by downstream conditions. On a rural trunk road, a short queue may form abruptly behind a turning movement and clear before a conventional count interval captures it.
The measurement specification should state whether a queue begins when vehicles are stationary, below a defined speed threshold, or both. A low-speed threshold is often more representative of operational congestion because it captures creeping traffic. However, setting that threshold too high can classify normal approach behaviour as a queue. The appropriate value depends on the road environment, approach speed, signal operation and purpose of the data.
It is also necessary to define the queue end. In most schemes, this is the furthest upstream vehicle that is stopped or moving below the selected threshold within a lane or detection zone. Where two lanes merge, or where lane discipline is inconsistent, a lane-by-lane measurement may be more meaningful than a single approach-level figure. The method should be agreed before commissioning so that reported values remain comparable over time.
Start with the decision the data must support
Queue data is most valuable when it answers a specific operational question. A signal engineer may need to know whether a queue reaches an upstream junction during the peak. A road safety team may need evidence of intermittent queuing beyond a bend. A network manager may be assessing whether a diversion route is creating unacceptable delay at a key node.
These use cases require different levels of precision and different reporting intervals. For signal optimisation, frequent measurements that identify maximum queue extent per cycle can be more useful than a five-minute average. For network performance reporting, a time series of average and 95th-percentile queue length may better show recurring congestion and unreliable journey conditions.
Avoid specifying a detector solely by a headline accuracy figure. Accuracy is meaningful only against a defined queue condition, road layout and validation method. A system that performs well for vehicles stopped at a signal line may need different configuration to detect slow, dispersed queues on a multi-lane approach.
Select detection technology for the site conditions
Above-ground detection gives highway teams a practical route to queue measurement without cutting the carriageway for inductive loops. The right sensor depends on the information required and the constraints of the location.
AI-powered video detection can provide detailed scene interpretation, including lane-specific vehicle presence, queue extent and classification. It is particularly effective where the camera view can clearly cover the full approach and where visual evidence assists scheme monitoring. Installation height, viewing angle, lighting transitions, glare, shadows and obscuration must be considered during design. A camera positioned for general surveillance is not automatically positioned for dependable queue measurement.
Radar detection is well suited to measuring vehicle presence, speed and range in challenging visibility conditions. It can provide dependable detection in darkness, rain and fog, making it a strong option for approaches where lighting varies significantly or where long-range coverage is required. Radar can also reduce sensitivity to visual scene clutter, although its field of view and lane separation capability need to match the road geometry.
Wireless traffic sensors can support rapid deployment or temporary studies where civil works would be disruptive. They may be appropriate for validating a problem location, monitoring works traffic or gathering baseline data before a permanent scheme. Their suitability depends on the required spatial resolution and duration of the survey.
In many locations, the best answer is not a single sensor type. Combining video-derived queue intelligence with radar speed or presence data can provide greater confidence where conditions are complex. The trade-off is additional integration and commissioning effort, which must be justified by the operational value of the information.
Design detection zones around real vehicle behaviour
Queue length accuracy is often won or lost in the placement of detection zones. A detector needs an unobstructed and correctly scaled view of the road section where vehicles are expected to queue, not just the stop line.
For signalised junctions, begin at the stop line and work upstream to the maximum queue extent that matters operationally. This might be the entrance to a nearby junction, pedestrian crossing, side road, bus stop or access point. Measuring a queue beyond the point at which it affects the network adds useful context, but only if the detector can do so reliably.
Zone boundaries should reflect lane markings, turn pockets, bus lanes and flare lanes. A left-turn queue can be operationally harmless if it is contained within a dedicated lane, while a short ahead-traffic queue may block all movements. Treating both as one approach queue conceals the condition that traffic engineers need to address.
At locations with horizontal or vertical curvature, survey the line of sight carefully. Parked vehicles, street furniture, signs, vegetation and large vehicles can all obscure the furthest queue point. Where full coverage is not possible from one mounting position, use more than one detection area or report the limit of observable queue length transparently.
Measure distance, vehicles and time together
A queue reported only as a vehicle count can mislead. Ten cars in a compact urban queue occupy far less road space than ten mixed vehicles including buses and articulated lorries. Conversely, a queue measured only in metres may not show how demand varies by vehicle type or lane.
A sound reporting method normally combines queue length in metres, number of vehicles and duration. For example, the operator may need to know that the eastbound approach exceeded 90 metres for four consecutive cycles, rather than merely that a peak queue of 110 metres occurred once. This distinction separates a brief fluctuation from sustained oversaturation.
Use maximum queue length per signal cycle where signal performance is the focus. Pair it with the time at which the queue reaches its maximum and the time taken to dissipate. If it does not clear during green, the residual queue is a direct indicator that demand exceeds available capacity for that movement.
Validate the system against ground truth
No queue measurement should be accepted solely because a detector is reporting plausible numbers. Validation compares the system output with observed ground truth, using synchronised video review, site observations or a controlled survey period.
Test more than one traffic condition. Peak and off-peak operation, daylight and darkness, wet weather, heavy vehicles, cyclists and unusual lane use can all expose configuration issues. A detector calibrated during a quiet midday visit may perform differently during a wet Friday evening peak.
Validation should check three things: whether a queue is detected when it exists, whether its end position is correctly identified, and whether false queues are being created by parked vehicles, shadows, opposing traffic or roadside activity. The acceptable tolerance should be linked to the decision being made. A scheme intended to prevent blocking back into an upstream junction may require tighter positional confidence than strategic trend monitoring.
Keep a record of the initial configuration, validation observations and subsequent changes. This is especially important when signal timings, lane layouts or road markings change. Data continuity matters, but it should not take priority over correcting a detector configuration that no longer reflects the physical layout.
Treat queue data as an operational input, not a dashboard metric
Accurate measurement becomes useful when it is linked to action. Persistent queues can inform signal timing reviews, stage sequencing, gating strategies, bus priority, active travel safety interventions or the case for a wider network scheme. Short-duration queue data can also identify incidents, signal faults and atypical demand before they become visible through public reports.
Context remains essential. A long queue is not automatically evidence of poor control if it clears efficiently and prevents a more harmful conflict elsewhere. Equally, a low average queue can hide severe but infrequent blocking-back events. Review maximums, duration, recurrence and downstream effects together.
For C & T Technology’s customers, non-intrusive radar and AI video detection can make this evidence easier to obtain, with reduced roadworks disruption compared with embedded detection. Yet technology alone does not create trustworthy results. Clear definitions, appropriate zone design and disciplined validation do.
The most useful queue measurement is the one that lets an engineer act with confidence at the next signal review, safety meeting or network control decision. Design for that moment, and the data will do more than describe congestion: it will help reduce it.