PaveCLIP: Automated Detection and Localization of Pavement Distresses Using CLIP and GPS Integration

Abstract: The timely identification and assessment of pavement distresses are vital for maintaining road infrastructure and ensuring traffic safety. Traditional methods of pavement inspection typically involve manual surveys and image analysis, which are labor-intensive, time-consuming, and lack the scalability needed for widespread applications. To address these limitations, we introduce PaveCLIP, an innovative system that utilizes CLIP (Contrastive Language–Image Pretraining) to automatically detect and classify pavement distresses in video data. By harnessing CLIP’s powerful vision-language model, PaveCLIP can process video footage in real-time, accurately identifying different types of pavement distresses based on visual input paired with descriptive labels.

A key feature of PaveCLIP is its integration with GPS data to localize the detected distresses precisely. The system extracts and fuses geospatial coordinates from the video metadata, allowing for the mapping of pavement conditions with high accuracy. This combined approach not only improves the efficiency of pavement monitoring but also enhances data granularity, enabling transportation agencies to respond to road maintenance needs more effectively.

Our evaluations demonstrate that PaveCLIP excels in distress classification and geolocation, showing marked improvements over traditional video analysis techniques in terms of both speed and accuracy. The system’s ability to process video data with automatic language-based recognition provides an accessible and scalable solution for road management practices. We anticipate that PaveCLIP will set a new standard for automated, data-driven infrastructure assessment, promoting proactive maintenance strategies and contributing to safer and more resilient transportation networks. This advancement underscores the potential for machine learning and multimodal models to revolutionize civil infrastructure monitoring and management.