Quantifying Roadside Evolution with Computer Vision and Deep Learning

Abstract 
This research investigates the use of deep learning-based 
computer vision techniques for monitoring road geometry 
changes to support urban planning and infrastructure 
management. Traditional road monitoring methods are 
often limited by time and cost, which necessitates an 
automated system capable of detecting and analyzing 
structural changes using video and image data. The 
proposed system consists of two custom-trained models: 
one for detecting Right of Way (ROW) and classifying 
surrounding land use types (residential, industrial, water 
bodies) around road boundaries, and another for 
identifying roadside vegetation. These models provide 
insights into unauthorized encroachments, vegetation 
distribution, and areas that need environmental

improvements. The approach involves processing input data 
through deep learning algorithms, converting 
video frames and images into quantitative insights 
that reveal structural changes over time. The 
models are trained using diverse urban road 
datasets and achieve reliable accuracy in detecting 
both road boundaries and vegetation. This project 
presents the system’s design, implementation, and 
performance, highlighting the potential of AI
driven solutions in transforming road 
infrastructure management.