Semi-supervised lane detection with Deep Hough Transform

TL;DR

This paper introduces a semi-supervised lane detection method that leverages a novel Hough Transform-based loss to utilize unlabelled data, significantly reducing annotation dependency and improving detection performance.

Contribution

The paper proposes a new loss function based on geometric knowledge in Hough space, enabling lane detection with minimal labeled data and effective use of unlabelled images.

Findings

Significant performance improvement on CULane and TuSimple datasets.

Effective lane localization via global max-pooling in Hough space.

Reduced reliance on manual annotations for lane detection.

Abstract

Current work on lane detection relies on large manually annotated datasets. We reduce the dependency on annotations by leveraging massive cheaply available unlabelled data. We propose a novel loss function exploiting geometric knowledge of lanes in Hough space, where a lane can be identified as a local maximum. By splitting lanes into separate channels, we can localize each lane via simple global max-pooling. The location of the maximum encodes the layout of a lane, while the intensity indicates the the probability of a lane being present. Maximizing the log-probability of the maximal bins helps neural networks find lanes without labels. On the CULane and TuSimple datasets, we show that the proposed Hough Transform loss improves performance significantly by learning from large amounts of unlabelled images.