Identification and Geolocation of Pavement Marking Issues Based on Artificial Intelligence and Mobile Phone

Artificial intelligence (AI) has been applied to detect faded pavement markings. However, previous studies only focus on particular types of pavement markings and do not involve the geolocations of identified marking issues in detection. Therefore, this paper proposes a lightweight approach leveraging deep learning and a mobile phone to identify and locate faded markings. First, videos are collected by a mobile phone mounted on the front windshield of a vehicle and then converted into images. A total of ~7,000 images with high quality capturing the faded markings are selected manually, and the faded markings are classified into two classes based on color and labeled as white faded and yellow faded markings. Then, a detection model is developed based on YOLOv8 and performs well in identifying the faded markings in white and yellow, including lane markings, arrow markings, delineators, and crosswalks, with an average precision of 0.90 and recall of 0.87. Additionally, geolocation is collected by an open-source GPX tracker along with the videos. Considering the heterogeneous sampling rates of video (30 frames per second) and GPX tracker (~2 times per second), this study develops another model to estimate and visualize the locations of the identified markings by time-based interpolation, which has a distance error of 0.27 m. The models proposed in this study offer an automatic and affordable solution to inspect pavement markings and locate identified marking issues, enabling efficient maintenance planning of pavement markings and ultimately improving safety for road users.