Title:	Deep Learning Based Road Geometry Estimation
DNr:	Berzelius-2024-391
Project Type:	LiU Berzelius
Principal Investigator:	Adam Lilja <adamlil@chalmers.se>
Affiliation:	Chalmers tekniska högskola
Duration:	2024-11-01 – 2025-05-01
Classification:	10207
Keywords:

Abstract

This project proposal aims to investigate the detection and tracking of lanes and road geometry for autonomous vehicles using cameras, using deep neural networks. The proposed research will focus on exploring the application of temporal fusion in enhancing the accuracy of the road model and incorporating uncertainty measures for the road geometry. Previous approaches for detecting and tracking lanes and road geometry have relied on traditional computer vision techniques, such as edge detection and Hough transforms. However, these approaches are limited in their ability to handle complex road environments and can be prone to errors. Some recent Deep learning techniques are only focusing on single frame detection of the road. These have shown promise in improving the accuracy of road detection. The proposed research will investigate how to train deep neural networks to enable faster inference and better accuracy in detecting and tracking lanes and road geometry. Additionally, the research will explore how to incorporate uncertainty measures into the road model to better handle noisy and ambiguous input data. Temporal fusion will be used to combine information from multiple time steps to make more accurate predictions about the road geometry. This technique has shown promise in other deep learning applications such as dynamic object tracking, but has yet to be extensively explored in the context of road geometry estimation. The outcome of this research will be a new approach to detection and tracking of lanes and road geometry using cameras for autonomous vehicles, as well as further understanding of the effectiveness of temporal fusion in enhancing the accuracy of the road model and the incorporation of uncertainty measures. This research has the potential to contribute to the development of autonomous vehicle technologies, and improve the safety and efficiency of road transportation.