Safe Driving on Indian Roads: IIIT-H expands Indian Driving Dataset

Traffic & Road Safety

Posted by AI on 2025-09-10 00:23:43 | Last Updated by AI on 2025-09-10 04:21:52

Share: Facebook | Twitter | Whatsapp | Linkedin Visits: 0

---

With Intel India and UC San Diego's support, IIIT-Hyderabad has grown the Indian Driving Dataset (IDD), first released in 2018, into a popular resource on unstructured traffic conditions. The dataset, which is now used by researchers in 30 countries, includes more than 46,000 tagged images and 12,000 LiDAR frames for studying how Western smart vehicles operate on Indian roads. The IDD was created to fill in categories absent from Western collections, such as auto rickshaws, cows, and unclear road boundaries, with 10,004 images taken in Hyderabad and Bangalore. It now has 15,000 downloads and 10,000 users across 88 countries. The research has expanded beyond merely self-driving cars. The same algorithms can identify potholes, count trees, and estimate flooding damage after heavy rain. The team also created the Bodhyaan research vehicle, which has cameras and sensors that collect multimodal driving data.

A recent addition is a dataset for two-wheelers, as 75% of India's vehicles are motorcycles. This has applications in electric vehicle range, fall detection, and rider safety, said Dr. Anbumani Subramanian, adjunct faculty at IIIT-H.

According to Professor C.V. Jawahar, who led the project, the goal was to construct something that accurately represented Indian reality. They wanted to demonstrate objectively how our roads differ from those in Western countries and to develop data that can be applied to Indian conditions.

Over time, the set of datasets has grown into numerous formats, such as IDD-detection, IDD-3D, and lighter versions for students and smaller labs.

Go!