Answers are grounded in this paper and cite their sources.
Introduction Lane detection is an essential feature of advanced driver assistance systems for reliable road scene perception. It allows the vehicle to determine the lane position and the orientation of its path within the lane. In recent years, deep learning has offered accurate approaches for lane detection in both the 2D image plane and 3D road coordinates. Methods This systematic review explores the advances in lane detection for autonomous vehicles from 2017 to 2026 using the PRISMA guidelines. Out of 2,839 records identified, 56 studies were selected based on the Critical Appraisal Skills Programme (CASP) quality criteria. These papers cover both 2D and 3D lane detection and include segmentation, transformer, key-point, and anchor-based approaches. The analysis focused on methodological innovations, public datasets, and lane representation, emphasizing the trade-off between accuracy and inference speed. Results This study demonstrates that transformer-based and anchor-based approaches consistently deliver state-of-he-art performance. Lightweight row-anchor models are well suited for real-time inference, and monocular 3D models improve 3D estimation. Discussion Challenges persist in cross-dataset generalization, robustness under rain, low light, and occlusion, and in the lack of confidence calibration and uncertainty estimation. Through PRISMA protocol-driven selection and quality appraisal, this work maps current lane detection trends, identifies the open problems, and outlines future directions towards building deployable and reliable lane detection systems in autonomous vehicles.
No takes yet. Share an insight, caveat, or question.
Maddiralla et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: