Drones Track 36,000 Vehicles as IIT Guwahati Builds Smarter Models for Indian Roads
The IIT Guwahati team used drones to observe traffic from above, where they could record many vehicles moving along the same stretch of road. This matters because Indian roads often have mixed vehicle types and less predictable movement than conventional traffic models assume. Detailed observations can help models represent those conditions more faithfully. Led by Professor Akhilesh Kumar Maurya, the researchers filmed daytime traffic on an eight-lane divided highway outside Delhi. Drones with high-resolution cameras recorded about five hours of footage in 4K. The team processed the videos with DataFromSky, software that automatically extracts vehicle movements. It produced records at 10 frames per second and classified vehicles into six groups. The resulting data covered 36,818 vehicles and nearly nine million positions. Researchers could examine how vehicles followed one another and changed speed as traffic became denser. The team used these observations to test traffic models, including machine-learning approaches. Such India-specific data could support more realistic simulations and inform road design, capacity assessment, and traffic management.
How did the IIT Guwahati team use drones to study traffic on Indian roads?
The IIT Guwahati team used drones to observe traffic from above, where they could record many vehicles moving along the same stretch of road. This matters because Indian roads often have mixed vehicle types and less predictable movement than conventional traffic models assume. Detailed observations can help models represent those conditions more faithfully.
Led by Professor Akhilesh Kumar Maurya, the researchers filmed daytime traffic on an eight-lane divided highway outside Delhi. Drones with high-resolution cameras recorded about five hours of footage in 4K. The team processed the videos with DataFromSky, software that automatically extracts vehicle movements. It produced records at 10 frames per second and classified vehicles into six groups.
The resulting data covered 36,818 vehicles and nearly nine million positions. Researchers could examine how vehicles followed one another and changed speed as traffic became denser. The team used these observations to test traffic models, including machine-learning approaches. Such India-specific data could support more realistic simulations and inform road design, capacity assessment, and traffic management.
How many vehicles and vehicle-position records did the researchers collect?
The researchers collected 36,818 vehicles’ movements and 8,938,920 vehicle-position data points. These figures show the scale of the dataset used to study how traffic behaves on an Indian highway. A large set of tracked positions lets researchers examine not just how many vehicles pass, but how they move and interact over time.
The team recorded about five hours of daytime traffic on an eight-lane divided highway on Delhi’s outskirts. Drones filmed in 4K, and DataFromSky software extracted vehicle movements from the footage. The resulting positions were captured at 10 frames per second. Researchers grouped the vehicles into two-wheelers, three-wheelers, cars, buses, heavy commercial vehicles, and light commercial vehicles.
This detailed dataset helped the team compare vehicle spacing and speed as traffic density changed. It also gave them evidence to test mathematical and machine-learning models. The researchers argue that data reflecting the diversity of Indian roads is needed because models developed elsewhere may not capture local traffic conditions. The findings could support more realistic traffic simulations and planning.
What is a stochastic fundamental diagram, and how is it different from a conventional traffic model?
A fundamental diagram relates traffic density to vehicle speed. In the conventional models described in the article, a given density generally produces one estimated speed. A stochastic fundamental diagram instead accounts for the range of speeds vehicles may actually travel at under those conditions. This distinction matters when traffic is variable and a single estimate would smooth over differences between vehicles.
For example, two vehicles in seemingly similar traffic may not move at the same speed. The IIT Guwahati researchers found that spacing also changed as traffic density rose: gaps were larger on relatively empty roads, narrowed as traffic increased, and changed again in severe congestion. A stochastic model can represent this kind of spread in observed behaviour rather than forcing all vehicles into one speed estimate.
The team developed stochastic fundamental diagrams using high-resolution drone data from an eight-lane highway near Delhi. Their aim was to make traffic models more realistic for Indian roads. The article does not give a single performance score for these diagrams, but says such India-specific models could support traffic simulations, road design, congestion management, and capacity assessment.
Why can vehicles travelling in similar traffic conditions still have different speeds on Indian roads?
Vehicles can travel at different speeds even when traffic seems similar because Indian roads are highly variable. Cars, buses, three-wheelers, and two-wheelers share limited space, and traffic can change quickly. This matters because a model that assumes one predictable speed for each traffic condition may miss differences among individual vehicles.
The researchers found that vehicle spacing changed with traffic density. On relatively empty roads, vehicles kept larger gaps. As traffic increased, the gaps became smaller, but spacing patterns changed again when congestion became severe. The study also notes that two vehicles travelling under seemingly similar conditions may not move at the same speed. These observations show why a single speed estimate may not capture traffic’s full range of behaviour.
The IIT Guwahati team used drone footage and automated tracking to measure movements on a highway near Delhi. They developed stochastic fundamental diagrams to represent the range of speeds possible at a given density. The models are intended to reflect the variability conventional approaches may overlook. More realistic models could help improve traffic simulations, road planning, and congestion management.
Which tested model performed most consistently, and how did its result compare with the machine-learning models?
Among the six mathematical traffic models tested, the Underwood model performed most consistently. It achieved an R² value of 0.86. The two machine-learning models, an Artificial Neural Network and a Random Forest, each achieved approximately 0.83. These results give a direct comparison between the tested mathematical and machine-learning approaches on the researchers’ traffic data.
The team studied traffic recorded by drone on an eight-lane divided highway outside Delhi. Automated tracking generated nearly nine million vehicle-position data points, covering six vehicle groups. Researchers used the dataset to examine speed and spacing as traffic density changed, then tested established mathematical approaches, including Greenshields and Underwood, alongside the two machine-learning models.
The reported figures show that Underwood had the highest result among the models described, though the article does not claim it will be best for every road or situation. The research also developed stochastic fundamental diagrams to represent the range of speeds vehicles may travel at a given density. The team says India-specific data and models could support more realistic traffic simulations and planning.
How could more realistic traffic models improve road design, congestion management, and traffic simulations?
Traffic models can help planners estimate road speeds, assess capacity, and design intersections. If a model reflects the way vehicles actually interact, its simulations may give a more realistic picture of traffic. This matters on Indian roads, where different vehicle types share space and movement may not follow orderly lanes or predictable speeds.
The IIT Guwahati study used drone footage to track vehicles on an eight-lane divided highway near Delhi. Researchers found that spacing changed as traffic density rose, and that vehicles in seemingly similar conditions could still travel at different speeds. Their stochastic fundamental diagrams represent a range of possible speeds at a given density, rather than just one estimate. That can help simulations preserve more of the observed variability.
The team says its data and India-specific models could support intersection design, highway capacity assessment, speed estimation, congestion management, and intelligent traffic-management systems. The data could also help calibrate car-following models, which examine how drivers adjust distance and speed in relation to vehicles ahead. These are potential applications, not reported outcomes of a road redesign.
What do traffic density and vehicle speed mean, and why does their relationship matter for understanding how traffic flows?
In standard traffic-engineering usage, traffic density means the number of vehicles occupying a given length of road at a particular time. Vehicle speed is how quickly a vehicle moves along the road. Their relationship matters because it helps describe how traffic behaves as more vehicles occupy limited space. The article does not give formal definitions, but it uses density and speed to explain the models tested.
The study found that spacing changed as traffic density increased. Vehicles kept larger gaps on relatively empty roads, narrowed those gaps as traffic grew, and showed different spacing patterns in severe congestion. Yet vehicles in seemingly similar conditions could still travel at different speeds. A conventional model generally estimates one speed for a given density, while a stochastic fundamental diagram represents a range of possible speeds.
The IIT Guwahati team developed these models using drone-tracked traffic on a highway near Delhi. Its results could help make simulations and planning more realistic for Indian roads. The researchers identify applications including highway capacity assessment, intersection design, congestion management, and traffic simulation. The goal is to better represent how traffic actually moves.
This brief was written by AI from the original reporting and checked by other models. Names, figures and quotes come from the source; read it for full context.
Read more in the JupiteX app
Pulse is free. New stories every 4 hours, each one broken into the questions that explain it.
Or read more news on the web