Improving Nighttime Driving Safety: Machine Learning for Street Luminaire Detection and Localization

Abstract: Ensuring road safety is a central concern in urban planning and transportation management, where traditional methods for monitoring and maintaining road infrastructure can be labor-intensive, costly, and prone to human error. Recent advancements in artificial intelligence (AI) provide new opportunities to address these challenges by automating tasks that previously required significant manual effort. This study focuses on the application of AI, specifically machine learning (ML) and image segmentation techniques, to improve the development and maintenance of roadway inventories, with a particular focus on street luminaires, which are essential for nighttime driving safety.
The core innovation of this research lies in the use of ML-driven image segmentation applied to Street View Images (SVIs). By leveraging AI to automatically detect and localize street luminaires, this study demonstrates how AI can enhance roadway infrastructure management in a way that is more efficient and scalable than traditional methods. Image segmentation, a technique that divides an image into multiple meaningful segments, is used here to isolate and identify streetlights from the complex background of urban environments. This process enables transportation agencies to quickly and accurately monitor critical infrastructure without the need for manual surveys or expensive hardware installations.
The detection module developed in this study achieved impressive results, with precision rates exceeding 0.98 and recall rates over 0.90. These metrics demonstrate the model's effectiveness in identifying street luminaires across various urban settings, even when faced with diverse lighting conditions, environmental factors, and camera perspectives. This level of accuracy underscores the potential of AI to outperform traditional methods in terms of both speed and reliability.
However, while the detection module was highly successful, the localization of street luminaires faced challenges in more complex scenarios. These included variations in camera angles, the presence of uncommon objects like steel bridges, and the impact of lane positioning. Despite these difficulties, the AI system still showed promise, and addressing these challenges through refined data collection methods and expanded training datasets is likely to improve performance.
The application of AI, particularly machine learning techniques such as image segmentation, offers transformative potential for roadway inventory management. By automating processes like the detection of streetlights, AI can significantly reduce the time and cost associated with manual inspections, while providing more consistent and accurate data. This, in turn, directly enhances road safety by ensuring that critical infrastructure, such as streetlights, is properly maintained, reducing the risk of accidents due to poor lighting conditions.
In conclusion, the study demonstrates that AI can play a pivotal role in advancing road safety through improved infrastructure management. Future research should focus on optimizing AI models, expanding training datasets, and conducting further testing in diverse environments to fully harness the potential of AI-driven solutions for road safety. By adopting these technologies, transportation agencies can develop more efficient, accurate, and cost-effective systems for managing critical road infrastructure.