A car brakes suddenly as a two-wheeler slips into the gap ahead. An auto-rickshaw swerves to avoid a bus, while another vehicle edges into the next lane. On Indian roads, where cars, buses, three-wheelers and two-wheelers often share limited space, traffic can change its rhythm in seconds.
For urban planners and traffic engineers, understanding this movement is no easy task. Many conventional traffic models assume that vehicles follow lanes and maintain relatively predictable speeds. But Indian roads rarely offer such orderly conditions.
Researchers at the Indian Institute of Technology (IIT) Guwahati are working to bridge this gap. By combining drone footage, automated vehicle tracking and machine learning, they have developed mathematical models that better account for the unpredictable ways vehicles move on Indian roads.
Tracking nearly nine million vehicle positions
Led by Professor Akhilesh Kumar Maurya from the Department of Civil Engineering, the research team studied traffic on an eight-lane divided highway on the outskirts of Delhi.
Using drones equipped with high-resolution cameras, they recorded around five hours of daytime traffic footage in 4K resolution. The videos were then processed using DataFromSky, an automated software that extracts vehicle movements from aerial footage.
Led by Professor Akhilesh Kumar Maurya, the IIT Guwahati research could help improve traffic simulations, highway planning and congestion management in India. Photograph: (IIT Guwahati)
The result was a dataset of 36,818 vehicles and 8,938,920 vehicle-position data points, captured at 10 frames per second. Researchers categorised the vehicles into six groups: two-wheelers, three-wheelers, cars, buses, heavy commercial vehicles and light commercial vehicles.
This gave them a detailed picture of how different vehicles moved, how closely they followed one another and how their speeds changed as traffic became denser.
Explaining why India-specific data is necessary, Prof Maurya says, “Traffic conditions in India are different from those in many countries where existing traffic models have been developed. We need traffic data that represents the diversity and characteristics of Indian roads.”
Drone-based data collection provides an opportunity to observe a large number of vehicles and their movements over a road section at high spatial and temporal resolution.
Teaching traffic models to account for unpredictability
One interesting finding was that the distance vehicles maintained from one another changed with traffic density.
On relatively empty roads, vehicles kept larger gaps. As traffic increased, these gaps became smaller. However, when congestion became severe, the spacing patterns changed again.
This variability is important because two vehicles travelling under seemingly similar traffic conditions may not necessarily move at the same speed.
Using automated vehicle tracking and machine learning, the researchers mapped 36,818 vehicles and nearly nine million vehicle-position data points. Photograph: (iStock)
To understand these differences, the researchers tested six mathematical traffic models, including established approaches such as Greenshields and Underwood. They also trained two machine learning models: Artificial Neural Networks (ANN) and Random Forest (RF).
But perhaps the most interesting part was their development of stochastic fundamental diagrams.
In simple terms, conventional traffic models generally estimate one speed for a given traffic density. Stochastic models go a step further by accounting for the range of speeds that vehicles may actually travel at under those conditions.
This allows the models to represent the randomness of Indian traffic more realistically, rather than smoothing over the differences between individual vehicles.
From better predictions to smarter road planning
The researchers found that the Underwood model performed most consistently among the mathematical models tested, achieving an R² value of 0.86. The two machine learning models achieved approximately 0.83.
The research could have practical applications in assessing highway capacity, designing intersections, estimating road speeds, and improving traffic simulations. The data could also help researchers calibrate car-following models, which study how drivers adjust their distance and speed in relation to vehicles ahead.
Prof. Maurya explains, “Reliable traffic management begins with models that represent how traffic actually behaves on Indian roads. The high-resolution trajectory data generated through drones allows us to capture interactions among different vehicle types and the large variability that conventional traffic models often overlook.”
Such India-specific models can support more realistic traffic simulation, better road and intersection design, congestion management, capacity assessment, and the development of intelligent traffic-management systems suited to Indian conditions.
Unlike conventional traffic models, the stochastic models developed by the team account for variations in vehicle speeds and distances, even under similar traffic conditions. Photograph: (Adobe Stock)
For anyone who has spent time on an Indian road, the challenge of predicting traffic is easy to understand. A car slows down, a two-wheeler slips into a narrow gap, an auto-rickshaw changes lanes, and the movement of hundreds of vehicles can shift in seconds.
By capturing these interactions in detail, the IIT Guwahati team hopes to make traffic models better reflect what commuters experience every day. The research is a step towards understanding not just how many vehicles use a road, but how they move around one another.
And that understanding could eventually help make the daily journey a little more predictable, whether someone is driving to work, taking a bus to school or navigating a crowded intersection.




