Infrastructure Systems: Digital Twins for Smart Streetscapes
REU Scholar: Vedant Sundriyal
REU Scholar Home Institution: CUNY Lehman
REU Mentor: Dr. Jinwoo Jang
Surface Reconstruction of Urban Photogrammetry for Wind Simulation: Developing QuASAR, a Quality Adaptive Reconstruction Method
View slide presentation for Surface Reconstruction of Urban Photogrammetry for Wind SimulationWind simulation over city scale 3D models is becoming more common as computing power and city scanning data improve, but these simulations need geometry that is closed and usable, not just realistic looking. Photogrammetric city scans, meaning 3D models built from photos, often fail this requirement. They can contain open gaps, broken pieces, and street clutter fused directly onto buildings. Existing reconstruction methods each solve only part of this problem. Poisson reconstruction closes the surface but rounds off sharp building edges. Alpha Wrapping produces a sealed surface but seals in the clutter along with it. Extruding building shapes from outside footprint data gives clean results but loses real captured detail and depends on outside data. This research presents QuASAR, short for Quality Adaptive Scan Aware Reconstruction, a method that treats different parts of a city scan according to how well that part was actually captured. Using a real scan of a city block in Miami, containing about four million faces and more than two hundred thousand disconnected pieces, our method uses shape-based point classification and whole object voting to separate buildings from clutter, then rebuilds only the weak or incomplete low rise buildings using footprints taken from the scan itself. Strong tower geometry is kept as captured. The full seven step pipeline runs automatically from one command. Our results show that the finished model is fully sealed and matches a hand tuned reference model within about zero point five three percent in volume. This quality adaptive approach gives an automated way to turn noisy city scans into usable geometry for future wind simulation work.
REU Scholar: Randy Amparo
REU Scholar Home Institution: CUNY Lehman
REU Mentor: Dr. Jinwoo Jang
Realistic Intersection-level Traffic Simulation Platform
View slide presentation for Realistic Intersection-level Traffic Simulation PlatformThe continued growth of urbanization has created significant transportation challenges in major cities across the United States, including traffic congestion, travel delays, and increased safety risks at busy intersections. Current transportation systems require accurate traffic data and data-driven planning to improve mobility and support effective infrastructure decisions. However, testing new roadway designs, traffic signal strategies, or safety improvements in the real world can be expensive, time-consuming, and disruptive. Digital twins provide a safe and controlled virtual environment for studying traffic behavior and evaluating possible transportation improvements before they are implemented. This research presents a realistic intersection-level traffic simulation platform that combines real-world vehicle trajectory data with the Simulation of Urban Mobility, or SUMO. The project uses the COSMOS trajectory dataset, which contains camera-based movement information for vehicles and pedestrians at an urban intersection. Unlike traditional SUMO simulations that rely mainly on randomly generated traffic, this approach uses observed trajectories to reconstruct realistic movements and traffic interactions within a virtual road network. The methodology consists of several stages: processing and cleaning the COSMOS trajectory data, transforming camera-based pixel coordinates into SUMO-world coordinates using homography, matching the transformed trajectory points to appropriate SUMO roads and lanes, generating simulation routes, and comparing the simulated movements with the original trajectories. The platform also measures distances between road users to identify close interactions, potential traffic conflicts, and locations where safety-related events frequently occur. The resulting digital twin demonstrates how real-world camera trajectory data can be integrated into a microscopic traffic simulation to reproduce intersection-level traffic behavior. This platform provides a reusable framework for analyzing vehicle and pedestrian movements, identifying potential conflict hotspots, and testing transportation improvements without affecting real road users. In conclusion, this research demonstrates the potential of combining computer vision, trajectory processing, coordinate transformation, map matching, and traffic simulation to support safer and more data-driven urban transportation planning.
REU Scholar: Mahfoudh Senhoury
REU Scholar Home Institution: CUNY Lehman
REU Mentor: Dr. Jinwoo Jang
Scenario Based Evaluation Of Urban Traffic
View slide presentation for Scenario Based Evaluation Of Urban TrafficUrban infrastructure decisions are costly and difficult to reverse, making it essential to evaluate the impact of proposed roadway changes before implementation. This study develops and validates a digital twin of the Glades Road × Airport Road corridor in Boca Raton, Florida, using SUMO microscopic traffic simulation calibrated against Florida Department of Transportation (FDOT) measurements. Peak-hour demand was derived directly from FDOT annual average daily traffic (AADT) counts using the standard peak-hour and directional-split factors, yielding target volumes of 2,619 vehicles per hour westbound and 1,881 vehicles per hour eastbound on Glades Road, and 403 vehicles per hour on Airport Road. The simulated baseline was validated against these targets using the GEH statistic, with all three approaches passing the FDOT/FHWA acceptance threshold (GEH < 5.0: 2.95, 1.84, and 3.76, respectively), establishing a validated baseline model. This baseline was then used to test a proposed infrastructure intervention — a short connector road linking westbound Glades Road directly to southbound Airport Road. While the intervention reduced westbound intersection volume by 16%, network-wide analysis revealed a 25% increase in mean travel time, with the majority of drivers who did not use the new connector experiencing increased delay. This outcome is consistent with Braess's paradox, demonstrating that a locally beneficial change can degrade network-wide performance. The results confirm the necessity of validated, data-driven simulation for evaluating infrastructure changes and establish a generalizable, GEH-validated workflow applicable to other FDOT-monitored corridors. Future work includes automated intervention search using surrogate models and Bayesian optimization, and a vehicle-counting pipeline for collecting real-time demand data from roads without existing traffic sensors.
REU Scholar: Cyrus Kahn
REU Scholar Home Institution: University of California - Riverside
REU Mentor: Dr. Jinwoo Jang
Quantitative Measurement of Driver Path Choice Changes Using Attributed Graph Networks
View slide presentation for Quantitative Measurement of Driver Path Choice Changes Using Attributed Graph NetworksTransportation systems continuously evolve as drivers adapt their route choices in response to changing travel patterns, personal habits, and external factors. Understanding these changes is important for analyzing long-term driving behavior, improving transportation planning, and supporting intelligent transportation systems. This research presents a graph-based framework for quantifying changes in driver path choice over time using attributed road networks. The methodology begins by enriching OpenStreetMap road networks with transportation data from the Florida Department of Transportation (FDOT), county Geographic Information System (GIS) datasets, and Mapillary to create a detailed representation of the roadway network. Raw GPS trajectory data are then processed using Fast Map Matching (FMM) to align noisy GPS observations with their corresponding road segments. The matched trajectories are represented as attributed graphs that capture both the structural connectivity of the road network and roadway characteristics. A Route Choice Change Index (RCCI) was developed to compare graphs across different time periods by measuring weighted changes in nodes and edges, providing a quantitative assessment of how driver route selection evolves over time. The framework was evaluated using multiyear vehicle trajectory data collected across South Florida. Monthly analyses and graph comparisons demonstrated the ability to identify stable travel patterns, detect significant route changes, and visualize long term behavioral evolution. These results show that attributed graph networks combined with map matching provide an effective and interpretable approach for measuring temporal changes in driver path choice. The proposed framework offers a scalable tool for transportation researchers to study driving behavior and may support future applications in travel behavior analysis, traffic management, and personalized mobility systems.