A Comprehensive Review on Lane Marking Detection Using Deep Neural Networks.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Automotive Engineering

Background:

• Lane marking detection (LMD) is critical for Advanced Driver Assistance Systems (ADAS) and autonomous driving features.
• Recent advancements in deep learning have significantly improved LMD performance.
• This paper provides a comprehensive review of current LMD techniques, emphasizing deep learning approaches.

Purpose of the Study:

• To review and analyze deep learning-based lane marking detection methods.
• To categorize LMD frameworks into single-stage and two-stage architectures.
• To discuss network architectures, loss functions, and optimization strategies for enhanced LMD performance.

Main Methods:

• Review of deep neural networks and conventional techniques for LMD.
• Categorization of LMD frameworks into single-stage and two-stage architectures.
• Analysis of network architectures (object detection, classification, segmentation) and loss functions.
• Comparative performance evaluation of five existing LMD techniques with visualizations.

Main Results:

• Deep learning techniques show promising improvements in lane marking detection.
• Analysis of different network architectures and loss functions reveals their contributions and limitations.
• Comparative results highlight the performance variations among existing LMD methods.

Conclusions:

• Lane marking detection faces challenges like generalization and computational complexity.
• Future research directions include efficient neural networks, Meta-learning, and unsupervised learning for robust LMD.
• Optimized deep learning models are essential for advancing ADAS and autonomous driving capabilities.