Detecting of Pavement Marking Defects Using Faster R-CNN
Hani Alzraiee, Andrea Leal Ruiz, Robert Sprotte
Abstract
Pavement markings on roads and highways are used to guide the roadway users. They play an essential role in promoting efficient use of the roadway and drivers’ safety. Typically, pavement markings deteriorate at a higher rate and last between 0.5 and 3 years. Because of the short lifecycle, pavement markings require frequent inspection and maintenance. Traditionally, pavement markings have been assessed periodically by road inspectors. This manual method is time-consuming, subjective, and exposes the road inspectors to high safety risks. Therefore, this paper presents a deep learning framework for automated pavement marking defects identification. The proposed framework uses a photogrammetry data set collected from Google Maps. Images of pavement markings are processed by annotating the marking defects. A deep learning algorithm called faster region convolutional neural networks (R-CNN) has been utilized to identify the pavement marking defects. The proposed model went through three iterations of training and used 1,040 annotated images. In the final stage, the model was tested using 60 images and was run for 46,194 epochs. The model was able to identify the pavement marking defects with a confidence level ranging from 43% to 99%. The model result was validated visually by inspecting the condition of the road markings used in testing the model. The proposed automated process is capable of generating a summary report of the condition of pavement markings that can enhance the current practices.