IIIT Hyderabad (IIITH) has been at the forefront of developing data-driven technologies to address the complex and often unpredictable realities of Indian roads. By applying advances in machine learning, computer vision, computer graphics, computational sensing and allied fields, the institute has been working on practical, scalable solutions that respond to real-world traffic conditions and driving behaviours unique to India.
A significant focus of IIITH’s research has been on anticipating driver and rider actions before they occur. On Indian roads, the ability to predict manoeuvres such as sudden turns, lane changes or abrupt braking can play a critical role in preventing accidents. To enable such early anticipation, researchers at the Center for Visual Information Technology (CVIT) curated the Driving Action Anticipation Dataset (DAAD), which captures both in-cabin and external views of vehicles operating in diverse traffic environments. The dataset spans varying weather, lighting and road conditions and includes sequences recorded before and during manoeuvres. This work culminated in the research paper “Early Anticipation of Driving Manoeuvres,” which was presented at the European Conference on Computer Vision in 2024.
Building on this research, the team extended its work to two-wheelers, where predicting intent is more challenging due to higher manoeuvrability and exposure. Anticipating rider behaviour before execution can help advanced driver assistance systems flag potentially unsafe actions in time. The resulting dataset on rider intention was submitted to the International Conference on Pattern Recognition 2024 as part of a global competition, marking an important step in addressing a segment that dominates Indian roads but has historically received limited attention in mobility research.
IIITH’s work also addresses infrastructure and governance challenges using AI-based approaches. In 2019, Hyderabad experienced unusually heavy rainfall that caused widespread damage to roads and related infrastructure. Traditional manual surveys to assess such damage are slow and resource-intensive. In response, researchers explored a low-cost alternative using mobile phones mounted on vehicles to capture road conditions and automatically identify damaged sections requiring attention. Similar AI-driven methods were later applied to estimate roadside tree cover at the request of government authorities, enabling faster and more accurate surveys.
Another extension of this approach focused on traffic enforcement. Visible surveillance cameras often alter driver behaviour temporarily, reducing their effectiveness. To address this, IIITH proposed unobtrusive systems using mobile cameras mounted on public vehicles to automatically detect violations and issue challans, supporting improved traffic discipline and compliance without intrusive enforcement mechanisms.
As the research evolved, the focus shifted from perception-based systems to predictive intelligence. While existing AI systems are adept at identifying objects and conditions, improving road safety requires the ability to anticipate what may happen next. This shift towards proactive understanding brings human behaviour back into the loop, recognising that every road user responds differently to traffic situations. One critical factor is driver attention, particularly where the driver is looking at any given moment.
Conventional gaze-tracking technologies rely on expensive and cumbersome equipment, limiting their real-world applicability. To overcome this, IIITH researchers introduced DashGaze, a low-cost driver gaze estimation system built using standard dash cameras. DashGaze is currently the largest dataset of mapped driver gaze in natural driving conditions, comprising approximately 0.9 million frames collected from 28 participants across multiple drives. The setup relies on a simple dual-camera mobile phone configuration, capturing synchronised views of the road, the driver and the driver’s egocentric perspective. The long-term goal is to enable attention monitoring and feedback directly on mobile devices, making the technology accessible and scalable.
Recognising the lack of datasets that reflect unstructured driving environments, IIITH partnered with Intel to create the India Driving Dataset (IDD). Unlike global datasets designed for structured traffic systems, IDD captures the realities of Indian roads, including heterogeneous traffic and inconsistent infrastructure. Data was collected using front-facing cameras mounted on vehicles driven across cities such as Hyderabad and Bengaluru. The dataset was released as a public resource for industry, academia and startups, and its widespread global adoption highlights the growing need for context-aware mobility solutions.
Beyond datasets, IIITH has played an enabling role by building platforms that support experimentation and innovation. A dedicated Data Foundation now hosts multiple open datasets, along with sensor-equipped data collection platforms for both four-wheelers and two-wheelers. Given that two-wheelers form the majority of vehicles on Indian roads, this focus fills a critical gap in advanced mobility research.
Since its initial release in 2018, the India Driving Dataset has expanded to include benchmarks for detection, classification, segmentation and 3D perception. While originally intended to support autonomous navigation research, the work has increasingly focused on addressing urgent road safety challenges. With approximately 1.5 lakh fatalities recorded annually on Indian roads, AI-driven prevention has become a necessity. Through initiatives such as iRaste, states like Telangana and cities like Nagpur are adopting data-led “safe systems” approaches, using predictive insights to improve vehicle safety, identify emerging risk zones and guide infrastructure interventions.
As AI-enabled mobility advances, the need for intelligent infrastructure, affordable tools and inclusive datasets will only grow. For India, meaningful progress will depend on understanding the full spectrum of road users, particularly two- and three-wheelers. By grounding innovation in real-world complexity, IIITH continues to contribute to the development of safer, smarter and more inclusive mobility systems for the future.
