Semi-Supervised Image-to-Image Translation for Lane Detection in Rain

Page view(s)
62
Checked on Jan 21, 2024
Semi-Supervised Image-to-Image Translation for Lane Detection in Rain
Title:
Semi-Supervised Image-to-Image Translation for Lane Detection in Rain
Journal Title:
A*STAR Scientific Conference 2021
DOI:
Publication URL:
Publication Date:
30 November 2021
Citation:
NA
Abstract:
Impressive progress had recently been made in deep learning based lane detection for the autonomous vehicle (AV) domain using an in-car camera. However, relatively little attention was paid to lane detection in bad weather conditions. AV applications are highly affected by the bad weather conditions. A semi-supervised image-to-image translation has been proposed to enhance lanes while preserving background. Our method can achieve better results than unsupervised learning because the content (lanes) to be translated are explicitly given by a few labelled training samples. Furthermore, the knowledge about the road is used to define a new loss function to improve the lane-awareness. The experimental results on a large database collected from internet and our autonomous vehicle have verified that our semi-supervised lane-aware image-to-image translation can achieve better lane enhancement effect than the unsupervised translation and the lane detection rate can be improved significantly after the images are enhanced with our approach.
License type:
Publisher Copyright
Funding Info:
This research is supported by core funding from: I2R
Grant Reference no. : EC-2020-046
Description:
ISSN:
NA
Files uploaded: