Detecting Urban Road Changes using Segmentation and Vector Analysis

Abstract

The rapid growth of urbanization is driving increased road infrastructure development. Detecting and monitoring changes in urban road areas is challenging for city planners. This research proposes using semantic segmentation and vector analysis on high-resolution images to identify road network changes. The U-Net model performs semantic segmentation, pre-trained on a Massachusetts road dataset, predicting labels for a specific area with temporal data and co-registration to reduce distortions. Predicted labels are converted to shapefiles for vector analysis. Satellite images from Google Earth archives demonstrate the change detection process. The outcome of this predictive phase was the transformation of projected labels into shapefiles, thereby facilitating vector analysis to pinpoint and characterize alterations.