Advances in Vision-Based Lane Detection: Algorithms, Integration, Assessment, and Perspectives on ACP-Based Parallel Vision

Xing, Yang; Lv, Chen; Chen, Long; Wang, Huaji; Wang, Hong; Cao, Dongpu; Velenis, Efstathios; Wang, Fei-Yue · 2018 · OpenAlex-citations

DOI: 10.1109/jas.2018.7511063

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This review paper addresses the critical need for robust and accurate vision-based lane detection systems, which serve as the foundation for Advanced Driver Assistance Systems (ADAS) like lane departure warning and lane keeping assistance. Motivated by the limitations of single-sensor systems in diverse driving conditions and the lack of standardized evaluation metrics, the authors provide a comprehensive analysis of existing methodologies. The study aims to categorize and evaluate current approaches to identify gaps and propose future directions for improving system reliability and safety. The authors conduct a structured literature review covering three primary domains: detection algorithms, integration methodologies, and assessment techniques. Regarding algorithms, the paper classifies methods into conventional image processing (feature-based and model-based) and machine learning approaches, including Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs). Feature-based methods rely on edge and color detection, while model-based methods use geometric fitting techniques like RANSAC and splines. Machine learning methods are highlighted for their superior accuracy and end-to-end capabilities, though they often incur higher computational costs. For integration, the authors categorize strategies into three levels: algorithm level (combining multiple detection algorithms), system level (integrating lane detection with object or road recognition), and sensor level (fusing camera data with radar or LiDAR). Finally, the paper reviews evaluation methods, distinguishing between offline metrics using ground truth data and online methods based on real-time confidence calculations. Key findings indicate that while machine learning algorithms significantly improve detection accuracy compared to traditional methods, they require substantial computational resources and training data. Model-based conventional methods offer better robustness against noise but are less flexible with complex lane shapes. The review highlights that integration strategies, particularly at the sensor and system levels, are essential for overcoming the inherent vulnerabilities of vision-only systems, such as false positives caused by shadows or similar-colored obstacles. The authors note that current evaluation practices suffer from a lack of uniform metrics and ground truth data, making comparative assessment difficult. The significance of this work lies in its systematic classification of lane detection technologies and its proposal of a novel framework based on Artificial Society, Computational experiments, and Parallel execution (ACP) theory. This ACP-based parallel vision framework aims to construct virtual parallel scenarios for model training and evaluation, addressing the challenges of real-world data scarcity and system validation. The paper concludes that future developments should focus on multi-level integration and parallel execution systems to enhance the robustness and reliability of lane detection, thereby supporting the advancement of higher-level autonomous driving capabilities.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• lane positioning