Lane Detection Using Deep Learning Approach

Abstract

The recent advancements in vehicle detection and self-driving capability of autonomous vehicles has become a challenging trend in recent years. Though humans are prone to errors while they drive the car, these automated vehicles are more critical in identifying and ensuring safety. The autonomous detection capability is quiet an interesting and more importantly be an demanding topic of research. The study on detection of appropriate LANE for identifying the road layouts directs for proper route identification. In the study presented here, Convolutional Neural Network was used for effective lane detection. Convolutional Neural Network (CNN) deals with images looking for deep pixel level analysis. For the study presented here, image frames were extracted from running length videos of the driving vehicles which of reasonable length (several minutes of less that 5 in number) from which image frames were extracted and processed. The procedure involves dealing with OpenCV functions, camera calibration and perspective transformation. The results obtained is quiet promising with accuracy level 87.5% as compared to existing algorithms and leads to proper detection of LANE using the deep learning algorithmic procedure used in the study.