Prof. Dr. Levent Kuzu, a faculty member at Istanbul Technical University's (ITU) Faculty of Civil Engineering, Department of Environmental Engineering, stated that they are able to calculate vehicle emissions using traffic camera footage through an AI-supported method.
The TÜBİTAK-supported project led by Kuzu and his team at ITU aims to predict air pollution from vehicles in real time by combining artificial intelligence and computational fluid dynamics. The developed method detects vehicle density and types through traffic cameras and uses this data to generate high-resolution air quality maps, enabling more accurate forecasts of pollution levels in cities.
Kuzu noted that traffic, industry, and residential heating are the three main pollution sources in cities, and that traffic significantly affects air quality.
He emphasized that combustion-based emissions such as carbon monoxide, particulate matter, and nitrogen oxides are observed in all major cities, and that there are various challenges in tracking these pollutants.
Highlighting the importance of easy access to data for modeling and forecasting studies, Kuzu said most existing data is general or average and that obtaining region-specific detailed data is not always possible.
"We wanted to more realistically calculate the contribution of vehicle emissions to air pollution by using artificial intelligence and deep learning as a sub-field. Our predictions are very close to observed values, meaning we can very accurately estimate ambient air concentrations," he said.
Kuzu explained that the system functions in three main stages:
"First, images from traffic cameras are analyzed using deep learning algorithms to classify vehicles and determine their speeds and types. In the second stage, emission factors specific to each vehicle type are used to calculate the estimated emissions generated by each vehicle group. In the final stage, the contribution of these emissions to ambient air is calculated using computational fluid dynamics models, taking meteorological data into account. This allows us to accurately estimate pollutant concentration at any given location. This method enables air quality predictions using only camera footage, without the need for fixed monitoring stations."
He noted that their method allows for emission estimation anywhere—from main roads to side streets—making it possible to obtain traffic-related pollution data as needed.
Kuzu said they used traffic monitoring cameras in Istanbul for the project and trained the model under different meteorological conditions. He confirmed that they could apply this model to cameras monitoring traffic in Istanbul.
In the pilot study, traffic cameras on Beşiktaş Barbaros Boulevard were used.
"As a result, we can predict vehicle types with over 95% accuracy. Emissions are calculated using this method, and we can determine pollutant concentrations at any point with video footage. The model was run for the existing air quality monitoring station at Yıldız Technical University's Beşiktaş campus, and the accuracy of the results was verified," he added.
Kuzu said the most important part of their software is detecting the real-time number and types of vehicles, which is made possible with image processing.
"In cities, we already know that traffic is the primary source of emissions. Globally, the two major pollutants that exceed concentration limits in cities are particulate matter and nitrogen oxides. Therefore, we must accurately identify emissions from traffic. Once defined and calculated, we can determine how to improve or prevent them. There are many areas where this can be applied—as long as you have the footage, it can be processed for any purpose," Kuzu concluded.
