Lane Endpoint Detection and Position Accuracy Evaluation for Sensor Fusion-Based Vehicle Localization on Highways

Area of Science:

• Computer Vision
• Robotics
• Automotive Engineering

Background:

• Landmark-based vehicle localization is crucial for autonomous driving and ADAS.
• Existing landmarks like lane markings lack longitudinal position data.
• Lane endpoints offer a potential solution for improved localization.

Purpose of the Study:

• To propose and evaluate methods for detecting lane endpoints using monocular cameras.
• To develop a reliable system for assessing the positional accuracy of detected lane endpoints.
• To enhance vehicle localization capabilities for ADAS and autonomous driving.

Main Methods:

• Lane endpoint detection involves lane detection, candidate generation, and verification.
• Accuracy evaluation uses a mobile mapping system (MMS) for ground truth comparison.
• Methods were tested on a dataset collected from 80 km of highway driving in various conditions.

Main Results:

• The proposed lane endpoint detection method is computationally efficient.
• The accuracy evaluation method reliably measures positional accuracy.
• The system demonstrated effectiveness in both daytime and nighttime driving scenarios.

Conclusions:

• Lane endpoints are viable landmarks for vehicle localization.
• The developed detection and evaluation methods are suitable for real-world applications.
• This research contributes to more robust autonomous driving systems.