Road Parameter-Based Driver Assistance System for Safe Driving

Abstract

One of the key aspects of Advanced Driver Assistance Systems (ADAS) is ensuring the safety of the driver by maintaining a safe drivable speed. Overspeeding is one of the critical factors for accidents and vehicle rollovers, especially at road turns. This article aims to propose a driver assistance system for safe driving on Indian roads. In this regard, a camera-based classification of the road type combined with the road curvature estimation helps the driver to maintain a safe drivable speed primarily at road curves. Three Deep Convolutional Neural Network (CNN) models viz. Inception-v3, ResNet-50, and VGG-16 are being used for the task of road type classification. In this regard, the models are validated using a self-created dataset of Indian roads and an optimal performance of 83.2% correct classification is observed. For the calculation of road curvature, a lane tracking algorithm is used to estimate the curve radius of a structured road. The road type classification and the estimated road curvature values are given as inputs to a simulation-based model, CARSIM (vehicle road simulator to estimate the drivable speed). The recommended speed is then compared and analyzed with the actual speeds obtained from subjective tests.