Reference

Counting traffic from a drone, in India

A drone sees a whole junction or mid-block from above, and fewer vehicles hide behind others. It also flies under rules, for as long as a battery lasts, at a height that decides how many pixels each two-wheeler gets. This note takes each in turn, from the rules and the published studies.

The rules

What the Drone Rules, 2021 ask of a survey flight

The regime is the Drone Rules, 2021, as amended in February 2022 and September 2023; we found no later amendment. A draft Civil Drone Bill that would replace them was put out for consultation in September 2025, and we found no record of its enactment.

Where you may fly. The green zone runs up to 120 m wherever the map shows no red or yellow zone, and up to 60 m between 8 and 12 km from an operational airport’s perimeter. Flying there needs no permission. Above those heights, and from the ground up between 5 and 8 km from the perimeter, is yellow zone, which needs air traffic control’s permission; red zone needs the Central Government’s. Flying in either without permission is a cognizable, non-compoundable offence.

What a survey drone needs, by weight class
Type certificate, remote pilot certificate and insurance by drone weight class, Drone Rules 2021 as amended. Every class needs a unique identification number.
ClassType certificateRemote pilot certificateThird-party insurance
Nanoup to 250 gNot requiredNot requiredNot required
Microover 250 g, up to 2 kgRequiredRequired for paid workMotor Vehicles Act applies
Smallover 2 kg, up to 25 kgRequiredRequiredMotor Vehicles Act applies
Every class needs a unique identification number. Drone Rules, 2021, rules 5, 13, 14, 36 and 44, with the 2022 amendment.

The remote pilot must check Digital Sky for restrictions before every flight, because the map changes. A State Government, Union Territory or law-enforcement agency can declare a temporary red zone for up to 96 hours, through an officer not below Superintendent of Police. Registration and pilot certification now run through eGCA; the airspace map stays on Digital Sky.

What the rules do not say. As published, they contain no day-only or line-of-sight condition and nothing specific to roads or moving traffic. DGCA’s operating procedures, which we could not find, may add conditions. The duty is general: do not endanger any person or property, and other laws still apply. One Ahmedabad study flew with the city traffic police’s permission, which is a record of practice, not a rule.

On national highways

Drone video of every project, for construction rather than counts

Since June 2021, NHAI has required monthly drone video of every National Highway project, uploaded to its data portal, and PIB described the requirement as still in force in February 2026.

NHAI’s contract clause has engineers analyse that video for encumbrances, change-of-scope sites, progress, mobilisation and the rectification of non-conformances. Traffic is not on the list, and the recording specification is not public. We found no NHAI or MoRTH document that counts traffic from drone video.

What the view buys

Fewer hidden vehicles, not none

Seen from directly above, vehicles “cannot be occluded by other vehicles”, the authors of a German motorway study write, though height information is lost. Indian teams found the same benefit: a “non-occluded view” of an Ahmedabad roundabout, and less trajectory loss on a Chennai arterial where 70.84% of vehicles were two-wheelers. In Chennai, temporary occlusions still caused errors. A drone also fits a whole junction into one frame, which the Ahmedabad team says CCTV would have struggled to do.

Overhead views give up side detail. If your schedule splits trucks by axle count, as ours does, check that your view shows the axles.

The angle matters more than the aircraft. An IIT Delhi comparison checked one commercial extraction tool against manual counts on urban mid-blocks, by day, for 30 minutes per set-up. The largest class-count error in each:

Largest class-count error by camera set-up, IIT Delhi preprint, 2024
Set-upLargest error
Drone looking straight down, 150 m7.0%
Fixed camera on a 5.5 m foot-over-bridge6.1%
Drone at 60°, 50 m (buses)82%
Drone at 60°, 25 m (heavy vehicles)197%
The errors are that tool’s, not the camera’s alone. Flying at 150 m is above the green-zone ceiling and needs air traffic control’s permission.

Geometry

Altitude sets the pixels on every vehicle

In a straight-down photograph, scale is focal length divided by height above the ground. The ground covered by one pixel, the ground sample distance, is therefore:

GSD = (sensor width × height) ÷ (focal length × image width)

Over flat ground the scale is the same across a straight-down frame, unlike an oblique camera’s. Height is the variable you choose: double it, and the frame covers twice the width while each vehicle gets half the pixels across.

At junction height, the Ahmedabad team found two-wheelers small enough for tracking to break. We ask for 1080p or above; if the drone records more, export at the resolution it recorded.

The Ahmedabad study’s camera, worked through
Ground sample distance, pixels across a two-wheeler and frame width by flight height and export width
Height and exportcm per pixelTwo-wheeler, pxFrame width
80 m, as flown2.6783.5102.5 m
85 m, as flown2.8478.6108.9 m
120 m, the green-zone ceiling4.0055.7153.8 m
80 m, exported at 1080p5.3441.8102.5 m
85 m, exported at 1080p5.6739.3108.9 m
Sensor 13.2 mm wide, focal length 10.3 mm, 3,840 px wide, flown at 80–85 m; two-wheeler 223 cm corner to corner, seen from above (Bhavsar et al., 2023). The 120 m and 1080p rows are our arithmetic on the same camera.

These are top-view diagonals. Our own pixel evidence is median two-wheeler height at a counting line, from three sites not described as drone footage: two-wheeler accuracy was 94.2% at 94 px, 88.4% at 89 px and 88.9% at 86 px. Three points do not make a curve, and none is a threshold for drone footage.

Endurance

A battery is shorter than a count

A 15-minute interval fits inside one flight, only just at the short end. A peak hour takes at least 3 flights of 25 minutes, with a gap at every battery change unless a second drone takes over in the air.

The seven-day count a DPR traffic survey rests on is 10,080 minutes: more than 403 flights of 25 minutes, across seven nights, and none of these studies flew at night. A hovering drone also drifts; the German and Chennai teams both stabilised their footage in software.

Our reading: a drone suits the junction and the peak hour; a fixed camera suits the seven-day count.

Flight length in the published studies
IIT DelhiDelhi NCR mid-blocks
Battery 15–20 minDaytime; night, wind and fog untested
highDGerman motorways
17 min on averageSunny, windless, 8 AM–5 PM
pNEUMAAthens city centre
Up to 25 min with take-off and landingMorning peak; about 10 min without data in every 30
IIT Kanpur and IIT TirupatiChennai arterial
About 18 minDry, daylight

Where we fit

We count the footage; whoever flies it holds the permissions

Playcraft ATCC is software that counts footage you supply. The drone, its registration, the pilot’s certificate and the airspace check sit with whoever flies it: your team or your operator.

It accepts drone video as well as CCTV and phone-on-pole footage, at 1080p or above and a bitrate that does not smear small vehicles. You place the counting line on a reference frame in two clicks, and it reports the 19-class schedule by direction in 15-minute intervals, with a per-vehicle log and an annotated verification video. It runs offline on your own machines, so the footage never leaves your building.

It counts at one line per video. A junction turning-movement count as a finished deliverable is something we build on demand.

We have not published a measurement on drone footage, including how altitude and hover drift affect our counts. Our accuracy figures come from three sites not described as drone surveys, and say nothing about drone footage. Fifteen minutes of yours, at the resolution it was recorded, is the way to find out.

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