Modern Advanced Driver Assistance Systems (ADAS) and autonomous driving (AD) solutions depend heavily on strong perception capabilities. These systems must not only identify moving objects like vehicles and pedestrians but also detect static road hazards that can impact how safely and comfortably a vehicle can operate. In India, poor road infrastructure—such as potholes, uneven patches, and worn surfaces—presents a serious safety concern. Unlike the more structured and well-maintained roads in developed countries, Indian roads often lack lane markings, have chaotic traffic mixes, and suffer seasonal damage from monsoons. Because of this, detecting road surface issues becomes a crucial part of the perception system, yet it remains underexplored.
This research introduces a simulation-driven framework to tackle this problem. Using the CARLA simulator, the framework generates multi-sensor datasets combining camera, LiDAR, and radar data. Potholes and damaged road surfaces are added into traffic scenes that resemble real Indian road conditions. To reduce the sim-to-real gap, the synthetic data is complemented by blending with real publicly available Indian datasets such as the Indian Driving Dataset (IDD/IDD-3D) for unstructured environments and RDD2020 for annotated road damages. The method merges deep learning on images with elevation data from LiDAR and reflections from radar, is data-driven relying on AI models and designed to better handle tricky environments like rain, dust, darkness, or visual blockage.
Hazard locations are marked using SLAM to help build smarter maps that allow for warnings, speed adjustments, or safer lane choices. Initial tests show that combining multiple sensors gives better results than using just vision alone. Overall, the work provides a flexible and repeatable process for detecting road hazards, tailored for the unique and messy reality of Indian roads, with the goal of improving safety for ADAS and autonomous vehicles in developing regions.
Keywords:
ADAS/AD, Potholes, Road Hazard Detection, Multi-Sensor Fusion, CARLA, SLAM/Geotagging, Indian Roads, India