Date of Publication :29th March 2024

Abstract: We aim to design an accurate and robust lane detection system, especially curved lane detection, to increase safety and reduce accidents on structured roads. Intelligent cars may dramatically lower the number of traffic collisions and our focus has been to design an algorithm as accurate as possible to make vehicles intelligent. In this project, an improved lane detection system using Convolutional Neural Networks (CNN) is proposed. As part of the pre-processing, different filters are applied on the image dataset like Gaussian Blur to remove noise, Sobel and Laplacian operator to calculate the gradient followed by Canny Edge detection. The aim remains to detect the lane pixels, determine the curvature of the lane (curves) and vehicle position with respect to the center. The detected lane boundaries are then wrapped back onto the original images of roads to highlight the lanes and curves accurately. A numerical estimation of the lane curvature is also displayed.

Reference :