Achieving High-Precision Localization in Self-driving Cars Using Real-Time Visual-Based Systems

Localization is a critical component for the safe and efficient operation of autonomous vehicles, as it enables precise positioning within the environment. In this paper, we present a novel approach to real-time visual-based localization for autonomous vehicles that combines the strengths of deep learning-based lane detection for robust perception with map matching algorithms for accurate localization against a priori map data. Through extensive experimentation and evaluation, our system achieves a remarkable mean Euclidean error of approximately 0.5 m, demonstrating high precision in localization tasks. Moreover, our approach operates at an impressive frame rate of approximately 35 frames per second (fps), ensuring timely and responsive localization updates crucial for real-time decision-making in autonomous driving scenarios to out-perform previous works on visual-based localization systems. These results underscore the efficacy and feasibility of our method in addressing the visual-based localization challenges in autonomous vehicle navigation, paving the way for enhanced safety and reliability in autonomous driving systems.