Hyderabad: Artificial intelligence may not be able to fix India’s potholes yet, but it can now spot them — along with distracted drivers, rule-breakers and erratic two-wheelers. That’s the vision researchers at IIIT-Hyderabad are working toward with a suite of low-cost AI tools designed specifically for Indian roads. At the recent TechForward seminar, Prof. C.V. Jawahar showcased how the Centre for Visual Information Technology (CVIT) at IIIT-Hyderabad is developing AI not just to detect, but to anticipate road risks. “Most global models are trained on neat, rule-following traffic. That doesn’t help much in India. Our goal is to build systems that work amid uncertainty, and on a phone, not just in a lab,” he explained.Among their latest innovations is Dashgaze, a tool that estimates driver attention using a regular dashcam or mobile phone camera. Unlike expensive eye-tracking hardware, Dashgaze uses naturalistic driving data — nearly one million frames — collected from real-world conditions in Hyderabad. “Distraction is a major cause of road accidents, but the tech to measure it has been out of reach. We’re trying to change that,” Prof. Jawahar said.The team is also working on predicting driver and rider behaviour through datasets such as DAAD (Driving Action Anticipation Dataset) and two-wheeler intention models submitted to global competitions. “If a rider is about to make a sharp turn without warning, the AI should catch the cues before the manoeuvre begins,” said a researcher on the project. “These few seconds can be the difference between a near miss and a crash.”IIIT-Hyderabad has applied AI to other critical road tasks — pothole detection, tree cover mapping and traffic violation monitoring — using mobile cameras mounted on public vehicles. The goal is to shift from bulky or visible surveillance systems to ambient, real-time monitoring. “People behave differently when they know a CCTV is watching. But roads need safer behaviour all the time,” Prof. Jawahar noted. The tools are powered by the India Driving Dataset (IDD), a publicly available dataset of Indian road footage created in collaboration with Intel. “We wanted startups and researchers to stop building on foreign data. The idea was: here’s how Indian roads actually look,” said a member of the team.This research is now finding its way into field applications like iRASTE, a project being piloted in Telangana and Nagpur. The system identifies accident blackspots, analyses mobility risks and supports real-time driver alerts. “Road safety can’t wait for autonomous cars to arrive. We need AI that helps today’s vehicles and today’s drivers make better decisions, especially on our kind of roads,” Prof. Jawahar said.
Source link
