IIITH uses AI to tackle real-world road challenges in India
Hyderabad: At the recently concluded TechForward seminar on Artificial Intelligence in Vehicles, Prof. C.V. Jawahar from the International Institute of Information Technology, Hyderabad (IIITH), highlighted the institute’s research aimed at addressing real-world road challenges in India using AI and data-driven technologies.
The work focuses on practical solutions that can operate under the diverse, often unpredictable conditions found on Indian roads. These include efforts in driver behavior prediction, road infrastructure assessment, and scalable safety tools using cost-effective setups such as mobile phone cameras.
Understanding Driver and Rider Behaviour
One of IIITH’s major research areas is predicting driver and rider actions before they happen, such as turns, lane changes, or sudden stops. This helps improve response times for Advanced Driver Assistance Systems (ADAS). The Driving Action Anticipation Dataset (DAAD), developed by the Center for Visual Information Technology (CVIT), includes interior and exterior views across different road and weather conditions.
The work was presented at the European Conference on Computer Vision 2024. The research was also extended to two-wheelers, which are more challenging to study but critical in India. A dedicated dataset was submitted to the International Conference on Pattern Recognition 2024 to support further research in this area.
Smart Monitoring Using Simple Tools
In response to road damage caused by heavy rains in Hyderabad in 2019, IIITH proposed using mobile phone cameras mounted on vehicles to quickly assess road conditions. This method offered a fast, low-cost alternative to traditional surveys. The same mobile camera approach was used to estimate tree cover and monitor traffic violations without relying on visible surveillance. Another notable innovation was Dashgaze, a tool that estimates driver attention using dashcams. It uses dual mobile cameras to capture both the road and the driver’s face, providing insights into attention shifts. Dashgaze is based on the largest naturalistic driver gaze dataset created so far, with over 0.9 million frames collected from 28 drivers.
Creating Datasets for Real-World Conditions
To address the lack of datasets suitable for India’s unstructured driving environments, IIITH collaborated with Intel to develop the India Driving Dataset (IDD). Data was collected from cars driving through Hyderabad and Bangalore, including urban and rural areas. Unlike datasets built around structured environments and traffic compliance, IDD captures the unpredictability and complexity of Indian roads. It is publicly available and widely used by researchers and startups working on mobility solutions. IIITH also created sensor-equipped platforms for both four-wheelers and two-wheelers, allowing developers to test AI models in live settings.
Towards Safer Roads with iRASTE
The institute is also part of iRASTE, a project led in partnership with the Telangana government and Nagpur city. The initiative uses AI to enhance road safety by predicting and preventing accidents. The approach includes vehicle-based ADAS alerts, real-time mobility risk monitoring, and infrastructure assessments. These systems have helped reduce accidents by up to 60% in some areas. By identifying potential blackspots and designing timely engineering fixes, the project is contributing to proactive road safety management. The focus on integrating two-wheelers and three-wheelers, which dominate Indian traffic, is a critical step in building a more inclusive and effective mobility ecosystem.
