Lane detection algorithm for autonomous driving using deep learning

Abstract

In the decade of Industrial 4.0, autonomous driving has been a popular and controversial topic. Autonomous vehicle has become more advanced during recent years due to the increase in the amount of available technology and computational power. The application of deep learning algorithms to perform lane detection for autonomous vehicle has been gaining positive feedbacks during these few years. However, the utilization of deep learning algorithms in lane detection systems has not been fully developed yet. Hence, this project aims to develop a low-cost robust lane detection system by deep learning which can deal with various road conditions and increase the accuracy of its detection results in real time. The deep learning algorithm (Semantic Segmentation: FCC-VGG16) was trained on a PC for detecting lanes. Data and images were collected from NTU MAE Robotics Research Centre. The trained model was installed on a Raspberry Pi which was used to control a donkey car. The donkey car was a prototype of autonomous vehicle and it was used to perform real-time lane detection. This final year project serves as a platform of work for future research on the feasibility of the usage of deep learning algorithm to build a lower-cost lane detection system with high robustness.