Reference

Pneumatic tube, infrared or video: which counter fits the survey

A tube counter feels axles; a video counter looks at vehicles. Each fails in conditions the other handles, and the published evidence says where. We sell video counting, so read this with that in mind and check us against the sources at the end.

Two ways of deciding what passed
What a tube counter and a video counter sense, classify by and leave behind
Tube counterVideo counter
SensesA burst of air each time a wheel crosses the tubeVehicles in the image
Classifies byAxle count and axle spacingWhat the vehicle looks like
Leaves behindThe times each tube was struckThe recording itself
FHWA, Traffic Monitoring Guide, §2.1.5 and Table 2-10; McGowen and Sanderson (2011), p. 6.

How a tube counter works

It detects axles, then infers the vehicle

A pneumatic tube counter records a burst of air each time a wheel rolls over a rubber tube laid across the road. Its logger then applies rules, such as axle spacing and speed, to decide what vehicle passed. With two tubes laid parallel, the time an axle takes to cross from one to the other gives speed, and that speed multiplied by the time between hits gives axle spacing.

To place a vehicle in a class, the counter must know where the vehicle begins and ends, how many axles it has and how far apart they are. FHWA’s guide says paired tubes classify “under free flow traffic conditions”, and that they do not work well in very high volumes, below about 8 km/h, or in stopped traffic.

The tubes are nailed or clamped across the carriageway, two per lane in the multi-lane layout, and laying them sometimes needs a lane closure. Portable infrared counters also classify by axles, without touching the road, and detect motorcycles well; fog degrades them, and a large road crown can block the beam.

In two-wheeler-heavy traffic

Side by side, an axle sensor sees one vehicle

A Montana transport department account, reproduced in FHWA’s guide, sets out the limit. Road tube, infrared and acoustic sensors “cannot distinguish an individual vehicle when multiple vehicles pass through the sensor field at the same time in the same lane of travel”, and its systems saw such vehicles “as one vehicle”. With Montana’s tube layout, two vehicles less than 12.2 m apart were misclassified.

The same account, written around 2008, says the video and radar classification systems it reviewed had the same difficulty. For a video counter, what decides it is the view: from a high, steep mounting, vehicles side by side appear apart; from a shallow one they overlap in the image, which is the most common cause of under-counting in dense two-wheeler traffic.

IIT Madras researchers name the lack of lane discipline as a feature of Indian roads, and wrote in 2014 that none of the existing automated sensors could count all the vehicles moving in parallel within a lane in Indian conditions.

The clearest measurement in mixed traffic comes from Malaysia. A 2019 study at 4 sites, 1 hour each, compared a tube counter with counts taken from video. The tube counter was 7.6% low overall and 25.1% low on motorcycles: 224 against 299. Going back to the video, the researchers found that when one vehicle overtook another at the tubes, the pair was missed or recorded as one large vehicle. They still judged the counter usable for five vehicle classes.

The Indian evidence

What has been measured in India, and how

The Indian evaluation we could find is a one-page letter from the Department of Civil Engineering, RV College of Engineering, Bengaluru, dated 15 September 2021. It covers one location on the Bangalore University campus and compares a tube counter’s totals with a video count in 5 categories: 5,344 vehicles, 61.4% of them two-wheelers.

Each category figure is the counter’s total divided by the video total, so it is a ratio of counts, not a vehicle-by-vehicle match; two-wheelers came out at 98.45%. The letter does not say how long the count ran. It reports more than 95% where vehicles were more than 6 seconds apart; for gaps under 3 seconds it says accuracy “perhaps” will be 90–95%, which is an expectation rather than a measurement.

The letter also records a limit of classifying by axles: buses could not be told apart from two-axle trucks and six-wheeled LCVs, nor small cars from four-wheeled LCVs and shared autos, because their axle lengths are almost the same.

A 2016 paper from PEC Chandigarh shows the same pattern. In the 7 Indian classes it set a tube counter up for, buses and trucks share a class, and three-wheelers and cars share an axle-spacing band of 2.10–2.68 m. That paper did not test accuracy. We found no peer-reviewed Indian study that measures a tube or infrared counter against a reference count.

How video counting differs

It keeps the traffic, not just the count

A video counter finds vehicles in the image, so it needs a clear view, enough pixels on each vehicle and enough light. FHWA lists what degrades it: fog, rain, shadows, vehicles overlapping in the image, the day-to-night transition, poor contrast, and water or grime on the lens. Reliable night operation needs illumination. Video does not measure axle spacing, and classifies by axles only where the camera angle shows them.

A 2023 study of one video counting program at US strip-mall entrances found it undercounted overlapping and fast vehicles, and lost accuracy in rain, low light and evening recording. FHWA also warns that two video counters can give “vastly different results” depending on their image-processing algorithm, which is why a published method matters more than the technology’s name.

What video keeps is the traffic itself. Both mixed-traffic tube evaluations above took their reference count from video, and in the Malaysian study the recording is what showed why the tube had erred. A tube counter’s raw record is a list of the times each tube was struck; unless a camera ran alongside, there is nothing to recount from. Counting a recording by eye is, in the Montana account’s words, “extremely labor intensive”, which is the work automated video counting does.

Where each fits

Match the instrument to the survey

Each row is drawn from the sources at the end. Where a cell says no source was found, none was.

Tube or infrared counter against video counter, by survey condition
A week unattended on a lower-volume, lane-disciplined road
Tube or infrared: What tubes are built for: short counts run up to a week, and tubes work well on lower-volume rural roads.
Video: Storing the whole run is a cost, and a portable camera mast is slow to set up.
A high-volume, multi-lane carriageway
Tube or infrared: Tubes “do not work effectively” there; the traditional layout covers only the lanes beside a shoulder.
Video: One camera covers several lanes; a view from the side loses far lanes behind trucks.
Two-wheeler-heavy traffic with no lane discipline
Tube or infrared: Vehicles side by side in a lane register as one; 25.1% short on motorcycles in the Malaysian study.
Video: Vehicles overlapping in the image are undercounted; a high, steep view keeps them apart.
Stop-go or very slow traffic
Tube or infrared: Does not work well below about 8 km/h or in stopped traffic.
Video: No general source found; one study counted slow, queued vehicles well.
Night
Tube or infrared: No lighting limit is listed.
Video: Needs illumination or street lighting.
Fog or rain
Tube or infrared: Snow is listed against tubes, not rain or fog; fog degrades infrared.
Video: Degraded by fog, rain and water on the lens.
Speed and axle spacing for each vehicle
Tube or infrared: Measured directly between the two tubes.
Video: Neither, in the shipped software: it counts at one line, so it gives no speed, and axle spacing is not measured.
Buses apart from two-axle trucks; three-wheelers, cycles, rickshaws
Tube or infrared: Not separated by axle spacing in the Indian evaluations; the US 13-class scheme has no three-wheeler or cycle class.
Video: Classified by appearance. Ask for per-class accuracy, not a total.
A count someone may dispute later
Tube or infrared: The raw record is a list of axle hits.
Video: Any interval can be recounted from the recording.
Corridor CCTV already recording
Tube or infrared: Not applicable: tubes have to be laid.
Video: Only if it sees the full width with enough pixels per vehicle. Most corridor CCTV was not mounted for counting.

Reading the numbers

Comparing a tube figure with a video figure

A total can be right while the intervals are wrong. In four Montana case studies, tube totals were within less than 4% of the reference, but individual 15-minute counts were off by 7–12% against hand counts. From either technology, ask for interval-level and per-class figures, and for how many sites, vehicles and classes stand behind them — seven questions do the job.

We have no head-to-head measurement. Our figures are measured against hand counts, not against a tube counter on the same road. On its one site, the Bengaluru letter’s two-wheeler ratio was 98.45%; ours, across three sites, is 90.1%. Those are different roads and different class schedules — 5 categories there, 19 classes here — so neither number tells you which instrument would do better on yours.

Where Playcraft fits

What we ship, and what we do not claim

Playcraft ATCC counts and classifies recorded CCTV, drone or phone-on-pole video into a 19-class schedule, with a per-vehicle log carrying timestamps and confidence, and an annotated verification video for manual audit. Against 16,560 vehicles counted by hand by an independent survey firm, at three sites on Indian arterial roads where two-wheelers were 60% of traffic, it measured 95.0% by volume and 92.1% weighted by class.

Its weakest published classes are Mini LCV at 86.3%, Cycle at 89.4% and 2-Wheeler at 90.1%. Two-wheeler accuracy was 94.2% where the median two-wheeler stood 94 px tall at the counting line, and 88.9% at 86 px, so camera position decides much of the result. Our published figures do not separate day from night, so they say nothing about a night count.

The shipped software does not measure speed: it counts at one line, and spot speed needs a second line at a surveyed distance. Speed, headway and queue studies are built on demand, not a shipped deliverable. The output carries no axle spacing, and an axle-load survey needs weigh-in-motion data that video cannot provide.

So if the survey is a week on a lane-disciplined road and needs speeds, rent the tubes. If the traffic is two-wheeler-heavy, the schedule needs buses and trucks apart, or the count may be disputed, send us fifteen minutes of footage, taken from the hours you will actually count, including night if the survey runs through it.

Sources

Next step

See it run on a call

  • The tool running on footage you choose, or ours
  • Every output file opened and explained
  • A plain answer on fit before commercials
See it run on a callSend us footage instead