Temporary Traffic Lights that can actually see the Queue

The idea in one sentence: Put a small camera roughly 20 feet above each end of a temporary roadworks site, use it to measure the true visible queue on both approaches, and let a safety-constrained controller adjust green time using better information.

The short verdict

This is a credible idea aimed at a real weakness in temporary traffic management.

Ordinary temporary lights are not blind. Most already use vehicle detectors, and newer commercial systems use modern radar and adaptive algorithms to alter green times. The broad idea of traffic-responsive lights therefore already exists.

The potentially useful distinction is this: detecting traffic is not the same as measuring the queue. A detector near the signal may know that both approaches are continuously busy, but not that one queue contains twelve vehicles while the other stretches several hundred metres and contains sixty. Once both local detection areas are full, the scale of the imbalance may exist mainly beyond the detector’s view.

A camera raised about 20 feet—or 6.1 metres—could look over ordinary cars and measure the stopped-queue tail, arrival rate, discharge rate and residual traffic after each green. Two cameras, one for each approach, would give the controller a much fuller picture of what is actually happening.

However, the camera system should never be allowed to invent its own traffic-light rules. An approved controller must retain authority over minimum and maximum green times, amber periods, all-red clearance, maximum waits and fail-safe behaviour. The proposed AI supplies measurements and requests; the safety system decides what is permissible.

The decisive question is not whether a raised camera can count vehicles. It can. The question is whether whole-queue information improves traffic flow enough to justify the masts, power, maintenance and approvals—and whether it beats the best adaptive radar systems already available.

The problem: both sides simply look “busy”

Imagine temporary lights controlling a single open lane through roadworks.

On one side, twelve cars are waiting. On the other, the queue contains sixty cars and is approaching a nearby roundabout. Both queues extend beyond the useful field of the local detectors, so both detectors remain continuously occupied.

To the controller, the two approaches may look broadly similar: traffic here, traffic there. It can infer demand from vehicle movements and timings, but it may not directly know how much remains beyond its measurement area.

The UK Department for Transport’s older guide to portable signals described conventional microwave detectors as seeing most moving motor vehicles up to about 40 metres away. It also warned that detectors cannot see around corners, parked vehicles, plant or materials. Operators were expected to observe changing flows and adjust settings when large queues developed. The booklet has now been withdrawn in favour of current guidance, but it explains the traditional limitation clearly (DfT portable-signals booklet).

Current local-authority requirements still reflect that practical problem. Merton Council, for example, says operators are responsible for adjusting green times during high traffic flows to minimise delays and balance queues (Merton Council).

The proposal is essentially to automate that act of looking down the road—continuously, from a better viewpoint and with recorded evidence.

What would the system actually do?

Each approach would have a compact telescopic mast positioned with the temporary signal equipment. At the top would be a small weatherproof camera looking back along the queue. The existing detector would remain in place, because radar and vision have different strengths.

The images would be processed locally at the roadside. Instead of constantly uploading identifiable video, the system would turn what it sees into a few useful measurements:

queue length in metres and approximate vehicles; position of the queue tail; vehicles arriving per minute; vehicles passing during green; how much queue remains after green; approximate numbers of HGVs, buses, motorcycles and cycles; whether the queue is about to block a junction or important entrance; and whether the view is clear enough to trust.

The two ends would share those measurements with the controller. It could then favour a substantially heavier queue, keep a productive green running slightly longer, avoid unnecessary direction changes and protect a queue that was nearing a roundabout or emergency access.

Fairness still matters. A small queue cannot be abandoned indefinitely because a larger one keeps arriving. The controller would therefore combine queue size with waiting time and guarantee every approach a bounded service.

If either camera became dirty, obscured, misaligned or unreliable, the system would report low confidence and return the lights to their approved conventional mode. “I cannot see properly” is an excellent answer from a safety system.

Why put the camera 20 feet up? Height changes the geometry.

A camera near the signal head mostly sees the fronts of approaching vehicles. A camera at roughly 6.1 metres looks down across vehicle roofs and gaps, making it easier to follow the stationary lane and locate the end of the queue. It may also see a tail approaching a junction that a local detector cannot observe.

This is physically realistic rather than a siege engine with a webcam. Mobile CCTV products already use six-to-nine-metre telescopic masts, while some modern traffic radars are designed for elevated mounting (Mobile VMS; Axis traffic radar).

Twenty feet does not grant omniscience. A bend, crest, tree or high-sided lorry can still hide traffic. The system must be able to say “queue continues beyond view” instead of producing a confident but fictional number. Difficult sites might need a small upstream sensor as well.

Has somebody already done it? Parts of it, certainly.

Traffic Group Signals’ AutoGreen uses modern FMCW radar and adaptive algorithms. The manufacturer reports journey-time improvements of up to 50% compared with conventional vehicle actuation and holds a UK patent covering aspects of radar-derived demand and portable signal control (AutoGreen; GB2535320 patent family).

SRL’s Multiphase Adaptive Detection System also combines radar with adaptive control. The company says it detects vehicles and cyclists up to 60 metres away. Published industry reporting on a modelled tidal-flow site claimed substantial reductions in queue length and journey time, although modelling for one scenario is not a universal field guarantee (SRL Multiphase ADS).

SRL’s REMOS includes radar, CCTV and remote human operation. Portable elevated CCTV and separate roadworks queue-warning systems are also commercially available. Patents already exist for CCTV-based road-queue detection.

Therefore DQI has not invented intelligent temporary lights, cameras, queue analytics or tall poles. The possible gap is the particular combination: a calibrated elevated camera on each approach, measuring the visible queue beyond the local detector and feeding comparative whole-queue information into constrained two-way control.

That may be a useful product feature rather than an entirely new industry. A patent professional and traffic-signal specialist would need to check novelty and freedom to operate before anybody cheerfully announces an invention.

Where it could make a real difference

The strongest sites would have visible but uneven queues: commuter traffic that changes dominant direction during the day; factories, schools, events or nearby junctions producing sudden platoons; long residual queues that need more than one green to clear; queues threatening to block a roundabout, side road or depot entrance; and

locations otherwise requiring repeated manual observation and adjustment.

The system could also produce anonymous reports showing queue lengths, delays, fallbacks and interventions. Dorset Council’s current guidance notes that authorities can struggle to verify whether supposedly intelligent signals are actually being used effectively. A measured performance record could be valuable in its own right (Dorset Council).

What it cannot fix

If both directions continuously demand more traffic than one open lane can carry, the machine cannot manufacture road capacity. Nor can it remove the all-red clearance period required to let the restricted section empty safely.

If both queues are equally enormous, whole-queue vision may improve fairness or spillback management more than total throughput. Complex junctions, pedestrians, abnormal loads, emergency access and unpredictable site activity may still require a competent human controller.

The mast also creates practical trouble. It needs ballast or outriggers, a safe wind rating, power, secure positioning, protection from theft and a plan for overhead lines, trees, bridges and narrow footways. Cameras acquire rain, road dirt, insects and bird deposits with impressive efficiency. A self-check and simple cleaning regime are not glamorous, but they are essential.

Safety and privacy

The AI should not directly command red, amber and green lamps. It should send demand information to an established controller that independently prevents conflicting phases and enforces every safety timing. In the first product, the camera must never shorten an approved all-red clearance because it thinks the road looks empty.

Privacy is manageable because the system does not need identities. Processing can happen locally, with raw frames discarded and only anonymous measurements retained. No number-plate recognition, facial recognition or audio is required. Any diagnostic video buffer would need strict access, encryption, limited retention and a proper data-protection assessment. The ICO treats number plates as personal data in most circumstances and expects surveillance operators to consider necessity and proportionality (ICO video-surveillance guidance).

Cost and the sensible route forward

A paired, non-controlling field demonstrator is likely to cost roughly £25,000–£60,000 once engineered masts, cameras, edge computers, batteries, radios, software and field support are included. A controller-integrated pilot with a manufacturer, safety work and live-site evaluation could reach £100,000–£300,000 or more. These are concept estimates, not supplier quotations.

The sensible route is not to create an entirely new traffic-light company. It is to work with an established manufacturer or traffic-management provider whose signal heads, controllers, maintenance system and authority relationships already exist.

The first experiment should use two six-metre cameras in shadow mode. They would observe a busy, directionally imbalanced site but exert no control. Human-reviewed video would establish the true queue, while the software estimates it automatically. The proposed controller would log what it would have requested.

The trial should then answer three questions:

Can the elevated cameras measure visible queue tails reliably in daylight, darkness and ordinary British weather?

Does that information improve simulated delay, throughput and spillback compared with properly configured conventional signals?

Does it add enough value beyond a current adaptive radar system to justify the extra hardware?

Only after those answers are positive should the system advise a human operator, and only after that should it be permitted to request limited green extensions from a live controller.

The DQI verdict

The central thought survives scrutiny.

Modern temporary signals already know that traffic is present. What they may not always know—especially when both local detection areas are saturated—is how much traffic remains beyond their view. A raised camera could provide that missing comparison.

The idea is technically plausible, commercially adjacent to products that already exist and burdened by entirely manageable but very real safety, mast, privacy and intellectual-property questions. It will not abolish queues. It may stop some queues being managed with unnecessarily poor information.

Final status: worth a controlled paired pilot.

First buy the evidence, not a fleet of enormous poles.

Or, in the language of the original proposal: give the lights a proper view of both queues and—if the data support it—bosh, get everybody moving.