Hybrid Method: Automatic Crack Detection of Asphalt Pavement Images Using Learning-Based and Density-Based Techniques

Pavement distress is a major contributor to overall road quality degradation. Maintaining roadways with the use of a Pavement Management System requires an extensive and up-to-date inventory of the many types of roadway distress. Such inventory for a road network can be labor-intensive and time-consuming to create. This study introduces a novel method, called Hybrid Method, to detect asphalt pavement surface cracks based on 2D images. Hybrid Method consists of two separate techniques, proposed by this study, which are combined to detect cracks. The first is a supervised learning-based technique which requires annotated images. This approach utilizes the Local Binary Pattern (LBP) as a texture descriptor algorithm to encode image data. Random Forest is the classifier, employed to learn from the extracted LBP features and predict new observations. The second, density-based technique, relies primarily on thresholding. This technique searches for crack objects iteratively while considering their spatial and geometric properties. These techniques were merged to create a hybrid crack detection method that was found to be effective for detecting pavement surface cracks. Four datasets were analyzed using the proposed Hybrid Method, which resulted in an average precision, recall, and F1 scores of 90%, 78%, and 84%, respectively. The results suggest that the Hybrid Method performed well in challenging situations, such as images with shadows, road markings, and oil stains. Its performance on images of different resolutions, pavement textures, and lighting conditions was remarkable as well.