This session will feature a range of remote sensing and GIS projects primarily from our student attendees.
1:00 PM – 1:15 PM – Mapping Building Entry and Exit Points for Pedestrian Navigation: Technologies, Workflows, and Insights from Downtown Atlanta
In most navigation systems, pedestrian routes are oversimplified – typically assuming people can move freely across open spaces or along sidewalks. However, real-world pedestrian mobility is far more complex. Factors such as inaccessible building entrances, missing elevators, or closed stairwells can significantly alter travel time, especially for people with disabilities, parents with strollers, or those unfamiliar with the area.
To address these challenges and enhance pedestrian mobility and accessibility in complex urban environments, the Georgia Tech team is leading a data-driven initiative focused on mapping detailed building entry and exit points in Downtown Atlanta. This work is a foundational step toward building a comprehensive, impedance-aware pedestrian network that reflects real-world travel constraints often overlooked in traditional pedestrian planning.
Our presentation showcases a robust workflow that combines low-cost data collection technologies, machine vision (MV), and advanced spatial analysis to identify, classify, and integrate pedestrian-accessible infrastructure such as elevators, stairs, and ramp systems into the network. These vertical circulation elements are modeled as unique link types with variable impedance values based on accessibility, elevation change, and status.
A key innovation is the introduction of multimodal pedestrian paths (e.g., bridges, pedestrian plazas, tunnels) as a new, distinct class of network elements. These are separated from traditional at-grade paths to better represent path complexity and context-specific travel impedance.
This project not only improves route precision for navigation apps and urban planning tools but also enhances mobility by accounting for a wider range of physical access scenarios. The techniques presented are scalable and cost-efficient, making them practical for deployment in other dense urban areas.
Ira Pathak, Georgia Tech
1:15 PM – 1:30 PM – PINN-TSE: A Hybrid Model for Improved Traffic State Estimation and Communication
Traffic congestion and inefficiencies cause safety, time, and environmental problems. Traditional traffic management systems struggle to provide accurate and reliable data. This paper introduces the Physics-Informed Neural Network-Based Traffic State Estimator (PINN-TSE), a new framework that combines physics-based traffic flow models with machine learning and natural language processing (NLP). PINN-TSE uses both physical rules and data-driven techniques to predict traffic density and speed accurately. It balances accuracy and physical consistency using a custom loss function. Large Language Models (LLMs) are used to create user-friendly traffic insights through a chat-based web app, answering questions about specific locations and times. Tested on real-world data from the US-101 highway, PINN-TSE outperforms data-driven models by 60% for density and 76% for speed predictions. It also reduces shockwave speed prediction error to 8%. The system can identify traffic jams and suggest alternative travel plans, demonstrating its practical value for real-world traffic management. This approach contributes to smart transportation by improving safety, efficiency, and sustainability.
Tewodros Gebre, North Carolina A&T State University
Leila Hashemi Beni, North Carolina A&T State University
1:30 PM – 1:45 PM – Geographic Object-Based Image Analysis (GEOBIA) Techniques for High Urban Density Mapping in Downtown Chicago, Illinois, USA
Remote sensing data, including multispectral imagery (MSI) and lidar, are increasingly available at high spatial resolution (HSR). There is an expectation that remote sensing data analysis scales alongside increased resolutions. One type of analysis is geographic object-based image analysis (GEOBIA) which creates image objects from pixels using segmentation algorithms. This analysis provides object properties including reflectance values, spatial relationships between objects, object textures and patterns and elevation and intensity values. While GEOBIA techniques have become more well defined and are increasingly published, there is a gap in the literature exploring GEOBIA in high density urban environments for land use land cover (LULC) classifications and extraction of features within cast shadows. The downtown area of Chicago offers a unique area of interest to explore these techniques. This presentation will cover image segmentation, creation of image interpretation keys, ruleset development and thematic accuracy assessment of initial and refined classifications within Chicago, Illinois, USA.
Daniel Bartlett, The Pennsylvania State University
1:45 PM – 2:00 PM – Variations in Volunteer Identified Damage Types in Post Disaster Aerial Photos
After a natural disaster occurs, damage assessment can provide critical information to first responders and recovery agencies. Using volunteers to identify damages remotely, using aerial imagery, offers a potentially faster and more effective method of collecting damage assessment data. However, variations in the identified damages are common, and finding methods to reduce these variations can help in the production of more effective products. The goal of this research is to identify ways in which variations can be reduced through different labeling techniques and identify methods that lead to more or less variation.
Devon Borthwick, University of Wyoming
2:00 PM – 2:15 PM – 3D Change Analysis in Terrestrial Laser Scanning using Scan Geometry for Enhanced Progress Monitoring
Accurate detection and quantification of three-dimensional (3D) change is essential for reliable progress monitoring and for maintaining up-to-date digital twins of construction and demolition sites. Conventional approaches to change detection in point cloud data often rely on voxel-based differencing, surface modeling, or machine learning algorithms that depend on empirical thresholds or training data, making results sensitive to registration noise, occlusion, and scene complexity and limiting interpretability and generalization.
This study presents a fully geometric and data-driven approach for 3D change analysis for Terrestrial Laser Scanning (TLS) data that compares a reference epoch against an analyzed epoch, each composed of multiple scans, and operates directly in the measurement domain by exploiting scan geometry, including range, azimuth, and elevation angle. A panoramic reference grid is constructed to evaluate line-of-sight visibility and spatial occupancy, enabling each analyzed point to be classified into one of five physically interpretable change states: unchanged, data gap, lost, added, and revealed, with all classifications derived from explicit geometric reasoning without learned parameters. The approach is evaluated using a multi-epoch TLS dataset acquired at a demolition site, where it reliably distinguishes true structural removal, newly added material, visibility-driven exposure, and occlusion-induced data gaps. A voxel-based aggregation is further used to quantify the spatial prevalence of each change state across epochs, enabling comparative analysis of progress patterns over time. By treating 3D change analysis as a scan-geometry–constrained reasoning problem, this study provides a transparent and physically grounded framework for extracting actionable change information from TLS data and for systematically updating digital twins to reflect evolving site conditions.
David Abiola, Oregon State University
