Fast-ICA Based Lane Detection Method for Autonomous Vehicles

Abstract

Lane detection is an important process in autonomous vehicle systems. Noise in the image, such as object shadows and terminating lane lines, make lane detection difficult. This study proposes a Convolutional Neural Network architecture with a dimension reduction method that has not been used before in lane detection. The proposed method has been tested with the open-source TuSimple dataset. The results showed that the proposed Fast-Independent Component Analysis based model training improved performance in lane detection and reduced the mean percent error by 42.2%.