Geopavenet: Automated road pavement damage classifaction and severity identification with geolocation using jetson nao and yolov8/
Julius Nikolai D. Bernado, Victor Sebastian D. Bondoc, Renato Jr. T. Panis, Efren Jr. D. Pastores, Gary Clyde T. Rabe, and Jevon A. Silvano.--
- Manila: Technological University of the Philippines, 2025.
- ix, 138pages: 29cm.
Bachelor's thesis
College of Engineering.--
Includes bibliographic references and index.
In the Philippines, road pavement damage poses significant threats to road safety, often leading to accidents, vehicular damage, and increased maintenance costs. Existing road survey methods that check pavement damage can take time, require significant labor, can be expensive, and make subjective evaluations, hurting accuracy and efficiency through delay in maintenance and repair on the damage that is occurring. GeoPaveNet is designed to innovate the traditional method of inspecting road pavement damages by using deep learning models, and geolocation; creating an automated, real-time system shifting traditional road pavement assessments using artificial intelligence (AI). Intended to be a potential adaptation by the Department of Public Works and Highways (DPWH), offering an innovative approach in improving road condition assessments and streamline maintenance planning processes. Using a moving vehicle, a high resolution camera is mounted capturing images of road pavement damages in real-time, processed by the NVIDIA Jetson Nano, YOLOv8 as the deep-learning model classifying road pavement damage classes: potholes, cracks, alligator cracks, pumping and depression, with severity levels determined based on the criteria provided in the DPWH Guides and Standards for Labelling Road Damage, which include factors such as crack width, depth, deformation, and damage pattern. A VK-162 GPS module was included to geotag each detected road damage by taking longitudinal and latitude location coordinates in real time for accurate location mapping and comprehensive reporting of the conditions of the roads. The trained object detection model had a mean Average Precision (mAP) of 0.423 after 100 epochs of training, which reflects the model's ability to detect and classify multiple types of pavement defects for practical, real-world deployment.