Traffic Data Collection Methods Compared

Match the method to the question

There is no best traffic data collection method. There is a method that answers your question and a set that cannot. A tube count across a lane produces excellent volume and speed data for a week and will never tell you how many vehicles turned left. A licensed engineer watching an intersection for two hours produces turning movements, classification and pedestrian behaviour and cannot be in three places at once.

So start from the output. If the analysis needs turning movements, peak hour factors by approach, or pedestrian behaviour, you need manual or video-based observation. If it needs volume, speed and class at a cross-section over days or weeks, automated sensors are better and cheaper. If it needs travel time along a corridor, you need re-identification or probe data. Most real projects need two of these.

Manual counting

A trained observer records movements in real time on a tally sheet or a count board. Still the reference standard for turning movements, complex geometry, pedestrian and cyclist behaviour, and any situation where judgement about what happened is part of the count.

Strengths: handles occlusion, unusual manoeuvres, weather and clutter that defeat detection; captures behavioural detail such as non-compliance, gap acceptance and yielding. Limits: no auditable record after the fact, observer fatigue over long shifts, and cost that scales linearly with hours.

Video-based processing

Footage is recorded on site and counted afterwards by analysts, frame by frame. This is our core method and it is a deliberate choice, not a fallback: it keeps human judgement in the count while adding an auditable record.

• Auditable — a disputed number can be re-checked against the original footage without returning to site
• Re-analysable — one recording can yield turning movements, queue lengths, saturation flow and pedestrian counts
• Pausable — a congested cycle can be reviewed at quarter speed instead of estimated
• Consistent — the same analyst standard applies to hour one and hour twelve of the footage
• Camera-dependent — a bad camera position produces occlusion that no amount of analysis fixes

Fully automated computer vision counting sits adjacent to this and is improving. Its accuracy is strongly conditional: good on free-flowing separated traffic, materially worse in dense mixed traffic, heavy occlusion, rain and night conditions — which are exactly the periods a capacity analysis cares most about. Where an agency will scrutinise the appendix, human-verified counts remain the defensible option.

Pneumatic road tubes

Rubber tubes across the carriageway register axle pulses. Cheap, fast to deploy, and good for multi-day volume, speed and axle-based classification at a cross-section. Vulnerable to being dislodged, unreliable in stop-and-go queues where vehicles straddle or stop on the tube, and structurally unable to resolve turning movements.

Inductive loops

Wire loops embedded in the pavement detect metal mass passing over them. Permanent, accurate for volume and occupancy, and the backbone of signal detection and permanent count stations. Installation requires cutting the pavement, and failures are invisible until someone audits the data — a decommissioned or broken loop reporting plausible-looking numbers is a real hazard in historical datasets.

Radar and microwave sensors

Side-fired or overhead sensors measure volume, speed and lane occupancy without touching the pavement. Durable and weather-tolerant, well suited to permanent monitoring and speed studies. Lane assignment errors occur in dense traffic, classification is coarse, and turning movements at an intersection remain out of reach.

Bluetooth and Wi-Fi re-identification

Roadside readers detect device MAC addresses at two points and derive travel time between them. Excellent for corridor travel time, journey time reliability and origin-destination sampling. Sampling rates are low and variable, MAC randomisation on modern devices has degraded match rates, and it produces travel time, not volume.

Probe and connected-vehicle data

Commercial datasets aggregated from navigation apps, fleet telematics and connected vehicles. Unmatched spatial and temporal coverage — a whole network, historically, with no field deployment. But it is a sample of unknown and non-uniform penetration, it under-represents some vehicle types and neighbourhoods, and it delivers speed and travel time far better than volume. Most agencies will not accept probe volumes in place of counts for a capacity analysis.

Comparison table

Frequently asked questions

The service behind this guide