Road measurement software

We publish the count and the miss.

Divided carriageway, both directions

26 JUL 2025 · 06:15
POS00:00:00 / 00:00:10
Four vehicles tracked at once; the panel increments per class as each crosses. Real survey footage, not staged.

Three sites, Indian arterial roads, live uncontrolled traffic

11 August 2026
Hand-counted
16,560
Counted
15,729
Weighted
92.1%
Volume
95.0%

We headline 92.1%, the lower figure. Per-site table ↓

Traffic counting, classification and pavement distress measurement from recorded video. Runs on your own machines, offline. Every number audits back to the frame.

Send us 15 minutes of footageRead the accuracy method

Back within a working day: the classified count, the annotated verification video, and where your camera position is costing you. You keep the output.

The counting line — 19 canonical classes

18 vehicle categories plus Other — see all 19 with their PCU values.

What you licence

Tools that run on your machines

Licensed software, not a metered service. Each installation is bound to its machine and verified offline: no login portal, no internet connection needed, nothing sent out of your office.

Seen running

The class that is easiest to get wrong

A model trained on car, truck and bus puts a five-axle truck and a multi-axle vehicle in the same column. The classes that carry the heaviest loads are the ones a count sheet exists to separate.

Five seconds of a production survey. Nothing here is staged or rendered for this page.

Five-axle articulated truck

10 JUL 2025 · 09:43
POS00:00:00 / 00:00:05
Axle-count classes are where generic detectors fail — a five-axle truck and a multi-axle vehicle look alike to a model trained on car, truck and bus. Here it lands in the right column, and the tally shows it. Bystanders at the roadside are blurred; we do not publish identifiable people.

Why the number holds

Accuracy is a method, not an adjective

Anyone can put a percentage on a page. The question a reviewing engineer asks is of what, measured how, against what.

Two figures, both disclosed

Volume accuracy answers capacity and PCU questions. Weighted class accuracy asks whether every vehicle landed in the right column. We publish both and headline the stricter one.

Ground truth is hand-counted

16,560 vehicles counted by hand by an independent survey firm — a separate company, not a separate process of ours. Accuracy is stated against those counts, never against another model’s output.

Thin categories excluded

Any class with fewer than 50 manual counts is marked statistically unreliable and kept out of the headline, rather than quietly averaged in.

We name what degrades it

Pixels on target at the counting line. Between Site A (94 px) and Site C (86 px), 8 pixels of median two-wheeler height correspond to 5.3 points of two-wheeler accuracy.

Read the full methodWhat we do not measure, and why →

Three sites, Indian arterial roads, live uncontrolled traffic

11 August 2026
SiteVehiclesAccuracy
A4,579 manual4,595 counted93.0% weighted99.7% volume
B5,824 manual5,375 counted91.2% weighted92.3% volume
C6,157 manual5,759 counted92.3% weighted93.5% volume
All16,560 manual15,729 counted92.1% weighted95.0% volume

Manual — hand-counted by an independent survey firm Counted — by the software

We lead with 92.1% weighted, not 95.0% volume. Weighted accuracy asks whether every vehicle landed in the right column. Volume accuracy lets over- and under-counts cancel.

For the reviewing engineer

What you hand the reviewer

Four files: two tied to a standard, two that let the reviewer check them back to a frame. These are the real ones from a production survey.

Also applied in builds on demand: IRC:108-2015 (AADT) · IRC:37-2018 (MSA) · MoRTH reporting. See what we build on demand.

Next step

Send us 15 minutes of your footage

  • Classified count, per class and per direction
  • Annotated verification video
  • A note on what your camera position is costing you

No obligation. You keep the output. Back within a working day.

Send us footageSee it run on a call