Axiom Traffic Vision
Most traffic counts in India are still done by hand at the roadside, or by a fixed camera that sees one point on the road. A drone hovering above a road section sees every lane at once. Axiom Traffic Vision turns that footage into per-lane counts, vehicle tracks and speeds.
The problem we solve
A classified traffic count usually means people with tally sheets at the kerb, or a camera on a pole. Both work, but both have limits that show up as soon as a road gets busy.
- Slow and error-prone: enumerators tire over a long shift, miss vehicles in dense traffic and have to be scheduled for every slot.
- One cross-section: a person or a fixed camera sees one point on the road, so five measuring points means five teams.
- No story between points: a manual count can't follow a vehicle as it changes lanes, and it can't measure speed along a section.
Seen from directly above, a road section becomes a map. Every lane, every vehicle and every lane change is in the same frame, recorded for as long as the drone holds its position.
How it works
- We fly the surveyAxiom pilots hover a DJI Mavic 4 Pro or Inspire 3 straight above the road, usually at about 45 to 60 m, once the location's permissions are in place.
- Every vehicle, trackedThe software detects each vehicle in every frame and follows it with an anonymous ID number for as long as it stays in view.
- Lanes and count linesWe draw the lanes and the measuring points (P1 to P5) on the frame. Each vehicle is counted once per point, in the lane it is in when it crosses.
- Measured in metresUsing the drone's altitude and lens, every vehicle's position becomes metres along and across the road. That gives speeds between measuring points, lane changes and acceleration.
- Flight checkThe drone's own flight log (DJI .SRT: altitude and drift) is checked against the clip, so the report says whether lane results and speeds are valid for it.
What you get
- Count sheet
- Vehicles per lane at each measuring point for every 20-minute slot, in your sheet's own layout, as Excel or CSV
- Segment speeds
- Average speed per lane on each stretch between measuring points, and how the same vehicles sped up or slowed down from one stretch to the next
- Lane changes
- Every lane change with where it started, where the vehicle crossed the lane line and where it finished, by clock time and metres along the road
- Annotated video
- The survey footage with each vehicle's class, ID and speed drawn on it, so anyone can check a count by eye
- Vehicle tracks
- A CSV with each vehicle's position and class, frame by frame, plus speed when the drone was hovering
- Flight report
- Altitude, drift and steadiness for each clip, and which results it supports
Privacy and your data
Traffic Vision counts vehicles, not people. It never reads or stores number plates, and it keeps no close-up images of vehicles. In the data, each vehicle is simply an ID number.
Footage and results are processed on our own machines in Bengaluru, not uploaded to a third-party cloud service. You receive the count sheets, the vehicle tracks and the annotated video.
Who it's for
- Traffic police and city or highway authorities planning signal timings, diversions and lane closures.
- Metro and infrastructure planners who need before-and-after counts around a construction site or a new junction.
- Traffic consultants who want a section-level survey to sit alongside their own counts and models.
Where it stands today
Traffic Vision is in pilot. We are running it on our own survey flights and proof-of-concept work, and we'd rather tell you its limits now than after the survey.
- It needs a hovering drone. Lanes are drawn on the frame, so lane results and speeds only hold while the camera stays still. The flight check flags clips where it moved.
- Small vehicles are the hardest case. Seen from above, a two-wheeler is a small, narrow shape that often rides close to larger vehicles, and some are missed. Improving this is our main work in progress.
- Longer sections need more drones. One drone covers part of a long section; a 500 m stretch needs several hovering side by side. The software can now join their clips into one road, matching their clocks and keeping one ID per vehicle across all of them. So far we have tested this on a single clip split in two, not yet on a real multi-drone flight.
Common questions
How much road can one drone cover?
By our estimate, a drone at about 60 m covers roughly 85 to 95 m of road in 4K with enough detail for lane-level counts. We plan the number of drones and their positions after a site visit.
Can you survey at night?
Not yet as a standard service. Night flights need their own permissions, and headlight glare makes tracking harder. We'd propose a test flight before promising night counts.
Can the count sheet match our format?
The export was built to match a real count sheet: one row per time slot, a block for each lane at each measuring point, and a total. The Excel file adds tabs for segment speeds, lane changes and vehicle speeds. Send us your layout and we'll tell you how closely we can match it.
How are speeds measured?
From how far each vehicle moves on the ground each second, using the drone's altitude from its flight log and the lens's field of view. Speeds are only reported for hovering clips. Vehicles cut off by the edge of the frame are left out, so they don't pull the averages down.
Request a pilot survey
Send us the road section, the measuring points and the time slots you need. We'll check the airspace, plan the flights and come back with what a pilot survey involves.