Development of Lane Marking Detection and Classification for Perception System in Autonomous Vehicles

Lane marking detection is a necessary system in autonomous vehicles as well as a challenging one to develop. It forms the basis for intelligent systems in autonomous vehicle driving and has always been a concern in the field of computer vision, especially in perception systems. In fact, there are a lot of machine learning methods used for lane marking detection including CNN-based lane marking detection with various models and datasets for model training. In this research, a lane marking detection method based on instance segmentation is presented. The proposed method is trained using the TUSimple dataset. The lane marking detection results are then classified into 2 types of classes i.e., solid and dashed lines based on training using deep learning with the same dataset. Furthermore, the proposed lane marking detection and classification method is conducted in real experiments in urban area at Bandung, Indonesia. The method gives a promising result as it can well-classified the lanes with more than 70% accuracy.