Asphalt pavement crack detection and classification using deep convolutional neural networks
Desai, Akshata Arvind
MetadataShow full item record
The Georgia Department of Transport (GDOT) maintains around 12,000 miles of interstate roadways. Annual pavement surveys are conducted to report the condition of the pavement. Survey engineers from GDOT conduct manual surveys and record their observations on the location, type and severity of pavement distresses like cracks, potholes, rutting, ravelling etc. Pavement cracks are among the major distresses that need to be serviced through the lifetime of the pavement to avoid frequent replacements that accrue high costs. The transportation departments spend millions of dollars to preserve pavement conditions. Therefore this task of roadway maintenance calls for an upgrade to automatically detect and classify pavement cracks that can save huge time and costs. In this era the technological industry is going through a revolution lead by Artificial intelligence. Any manual tasks such as detecting pavement cracks and classifying them based on the level of severity can be automated using machine learning and AI. The aim of this research work is to automate the task of crack classification based on GDOT’s protocol using deep learning techniques that can learn by leveraging information from huge image datasets. Convolutional neural networks are a class of deep neural networks that can be used for extracting specific features from an image. These features are used by the network to further detect and classify pavement cracks. This research develops a complete method to use the detected crack patterns from the image and output the type, severity and extent for pavement cracks. Crack patterns are detected using the Faster RCNN model and they are post-processed to match the crack type and severity definitions in Computerized Pavement Condition Evaluation System (COPACES).