Ìý

Realistic Intersection-level Traffic Simulation Platform

By Randy Amparo
Slide 1: Title slide for Realistic Intersection-level Traffic Simulation Platform, presented by Randy Amparo.

Slide-1

Realistic Intersection-level Traffic Simulation Platform

Randy Amparo

Bachelor's Student in CIS, CUNY Lehman College

Mentors: Dr. Jang and Dr. Moshfeghi

July 30, 2026

Summer REU 2026

Slide 2: Presentation outline listing four sections, Introduction, Methodology, Results, and Conclusion, each with sub-topics.

Slide-2

Outline

  • 01 Introduction: Motivation & Importance; Definitions (SUMO, Homography, COSMOS); Background & Scientific gap; Objectives & Research Goal
  • 02 Methodology: Dataset Scale; Homography Process; Project Pipeline
  • 03 Results: Sumo Demo; Final Results
  • 04 Conclusion: Conclusion; Acknowledgements

Realistic Intersection-level Traffic Simulation Platform — Randy Amparo

Slide 3: Section divider slide reading Introduction.

Slide-3

Introduction

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 4: Motivation slide listing reasons for the research, alongside a photo of dense New York City traffic and a line chart showing traffic volume growth outpacing population growth.

Slide-4

Motivation

Growth of urbanization & concerns for transportation:

  • Time wasted every year to delays.
  • Testing new traffic strategies in the real world is expensive and disruptive.
  • Digital twins provide a safe environment for experimentation.
  • More accurate for making changes to infrastructure.

A photograph shows a busy New York City street filled with cars, taxis, and pedestrians crossing at an intersection.

A line chart titled "Urbanization Outpaces Population Growth in NYC Traffic" plots indexed growth (2005 = 100) from 2005 to 2024. Two lines are shown: Traffic volume (VMT), which climbs from 100 in 2005 to about 114 by 2019, drops sharply to about 90 in 2020, then recovers to around 116-117 by 2023-2024; and Population, which stays relatively flat, rising slowly from 100 to about 103 over the same period. The chart's caption states that vehicle traffic (VMT) has climbed far faster than population. A note below explains the chart shows daily traffic volume relative to population in New York City areas, showing that vehicles exceed population due to commuting.

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 5: Why This Research Matters, listing benefits of digital twins alongside two aerial images of urban road networks with labeled links and vehicle lanes.

Slide-5

Why This Research Matters

Current transportation systems require accurate traffic prediction with data-driven planning.

Digital twins make it possible to:

  • Test infrastructure changes
  • Analyze congestion
  • Improve traffic efficiency
  • Support smart city development

Two side-by-side aerial satellite images labeled (a) and (b) show urban road networks overlaid with colored annotations: image (a) shows a dense city block grid with red dots marking labeled nodes at intersections; image (b) shows a highway interchange with blue lines marking links, red lines marking vehicle lanes, and red dots marking labeled nodes, per the legend in the image.

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 6: SUMO, a screenshot of the SUMO 1.27.0 simulation software interface with callouts explaining the time delay slider and simulation time display, next to a definition of SUMO.

Slide-6

SUMO

What is SUMO?

SUMO or Simulation of Urban Mobility (SUMO) for short is an traffic simulation software that creates a virtual environment of real roads.

It allows researchers to simulate how vehicles, pedestrians, and traffic signals interact without testing in the real world.

A screenshot of the SUMO 1.27.0 application window shows its menu bar (File, Edit, Settings, Locate, Simulation, Window, Language, Help) and toolbar with a Time display, a Delay (ms) field, and a Scale Traffic field. Callout labels point to the "Delay (ms)" control, noting it slows down visualization, and to the Time field, noting it shows the specific time frame the simulation is at.

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 7: Homography slide explaining why camera coordinates and SUMO coordinates do not match, and defining homography as a transformation that converts pixel positions into real-world x,y coordinates, illustrated with a two-panel diagram of colored reference points.

Slide-7

Homography

Converting Camera metric data into Sumo:

  • The coordinate systems for camera images and SUMO simulations do not match.
  • Problem because COSMOS dataset comes from cameras.

What Homography Does?

  • Homography is a transformation that aligns camera images with a real-world map by using reference points.

A diagram shows two trapezoidal panels, each containing five colored dots (cyan, yellow, green, blue, red, magenta) in similar but skewed arrangements. An arrow labeled "H: C sub px → C sub w" points from the left panel to the right panel, illustrating the transformation. The left panel is labeled "pixels" and the right panel is labeled "x,y coordinates."

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 8: COSMOS Dataset slide with a description of the camera-based traffic dataset and a satellite image of the Amsterdam Avenue and 120th Street intersection in New York City.

Slide-8

COSMOS Dataset

  • COSMOS is a camera-based traffic dataset that records the movement of vehicles and pedestrians in a New York intersection (Amsterdam Ave and 120th Street).

A satellite image shows the intersection of Amsterdam Avenue and West 120th Street in New York City, with a red pin marking the intersection and surrounding buildings labeled, including Fu Foundation School of Engineering, Mudd Hall, Lowell Hall, Massawa Ethiopian restaurant, Hartley Pharmacy, and Columbia University Institute for research buildings nearby.

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 9: Background slide describing prior use of homography in traffic research, alongside an illustration of a busy multimodal intersection and a screenshot of a SUMO intersection simulation.

Slide-9

Background

  • Most existing research uses homography to convert vehicle positions from camera pixel coordinates into real-world coordinates.
  • However, they do not integrate the data into a realistic digital twin for visualization.

An illustration depicts a busy urban intersection with cars, a bus, cyclists in a bike lane, and pedestrians in crosswalks. Below it, a screenshot of a SUMO simulation shows a simplified four-way intersection with vehicles rendered as small colored rectangles on gray road lanes.

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 10: Scientific Gap slide explaining the lack of a bridge between traffic simulations and real-world datasets, illustrated with an aerial highway interchange image with colored vehicle trajectory lines connected by an arrow to a simulated intersection with colored vehicle icons.

Slide-10

Scientific Gap

  • Simply, there isn't a bridge to connect simulations to real world datasets.
  • Simulations like SUMO visualize synthetic traffic, which doesn't fully reflect real-world driving behavior.
  • But, incorporating real world data can make them more useful.

An aerial image of a highway interchange overlaid with multiple colored trajectory lines (pink, blue, green, yellow, red) tracing vehicle paths, with several vehicles labeled "Car." A red arrow connects this image downward to a second image, a top-down simulated view of an intersection showing vehicles as colored rectangles with small red dots marking tracked positions on crosswalks and lanes.

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 11: Objective and Research Goal slide describing the plan to build a realistic digital twin in SUMO using real camera-based trajectory data, illustrated with a side-by-side comparison of a camera view with tracked vehicle boxes and its corresponding Digital Twin simulation view.

Slide-11

Objective & Research Goal

  • Develop an realistic intersection-level digital twin in SUMO by integrating real camera-based vehicle trajectory data.
  • Used to evaluate traffic behavior, study safety-related scenarios, and test potential improvements.

A composite image shows, on the left, an overhead camera view of an intersection with yellow bounding boxes tracking two vehicles, and on the right, a labeled "Digital Twin" simulation view of the same intersection showing simplified vehicle icons and a traffic signal at a four-way crossing.

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 12: Section divider slide reading Methodology.

Slide-12

Methodology

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 13: Project Pipeline, titled From Camera Data to a Digital Twin, showing a ten-step numbered workflow list with descriptions, next to a matching six-image vertical sequence from raw camera image through homography, SUMO road network, SUMO simulation, to the final digital twin overlay.

Slide-13

Project Pipeline: From Camera Data to a Digital Twin

  1. COSMOS Camera Dataset: Raw intersection video from COSMOS dataset.
  2. Extract Trajectories (Vehicles & Pedestrians): Detect and track vehicles and pedestrians to obtain trajectories in image space.
  3. Trajectory Cleaning & Motion Analysis: Remove noise and outliers, compute speed, direction, acceleration, etc.
  4. Homography Transformation (Pixel → SUMO Coordinates): Transform trajectories from image coordinates to real-world (SUMO) coordinates.
  5. Map Matching (Trajectories → Road Network): Match trajectories to the SUMO road network.
  6. Route Generation: Generate routes for all matched agents in SUMO.
  7. SUMO + TraCI Simulation: Run microscopic simulation and control via TraCI.
  8. Simulation Validation (Compare Real vs Simulated): Compare trajectories and metrics to validate realism.
  9. Conflict & Traffic Analysis: Compute conflicts, TTC, density, speed, and other traffic metrics.
  10. Final Digital Twin: Realistic, data-driven digital twin of the intersection.

A parallel vertical sequence of six images illustrates the pipeline output at each stage: a "Camera Image" showing an overhead intersection view with cars and pedestrians; a "Homography (Transformed)" image showing colorful trajectory lines on a black background; a "SUMO Road Network" image showing the constructed road network for the intersection; a "SUMO Simulation" image showing simulated vehicles and trajectories on the road network; and a "Final Digital Twin (Overlay)" image showing the simulation overlaid with real trajectory paths for visual comparison.

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 14: How the Dataset was organized (Step 1+2+3), describing the 1.2 million-observation COSMOS trajectory dataset and the resulting DataFrame columns.

Slide-14

How the Dataset was organized (Step 1+2+3)

The COSMOS trajectory dataset has over 1.2 million trajectory observations, representing the movement of vehicles and pedestrians captured across multiple scenes.

After processing

The data have been organized into a DataFrame with the following columns:

  • scene_id (id of the observation)
  • track_id (specific road)
  • frame (specific moment in order)
  • time (when did it happen)
  • agent_type (pedestrian or vehicle)
  • pos_x and pos_y

Also, transformed coordinates acquired from homography: (x_world, y_world) and matched edges.

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 15: Homography Process (Step 4), a four-panel diagram showing selecting reference points on a camera image, computing the homography matrix H, transforming trajectories from pixel to world coordinates, and the resulting trajectories aligned on the SUMO road network.

Slide-15

Homography Process (Step 4)

A four-panel horizontal diagram illustrates the homography workflow:

  1. Select Reference Points: An aerial photo of an intersection with four colored dots (red, yellow, cyan, magenta) marking chosen points visible in both the camera image and the map. Caption: Choose matching points that are visible in both the camera image and the map (real world).
  2. Compute Homography Matrix (H): Two trapezoidal grids connected by an arrow labeled "H," representing the image plane transforming to the real-world plane. Caption: Homography finds the best transformation H that maps points from the image plane to the real-world plane.
  3. Transform Trajectories (Pixel → World): Four colored trajectory paths (red, green, blue, orange) made of connected dots, representing trajectories after transformation. Caption: Apply H to every trajectory point to obtain real-world coordinates (meters).
  4. Aligned on SUMO Road Network: The same four colored trajectories now overlaid correctly onto a dark rendered SUMO intersection network. Caption: Trajectories now align with the correct roads in the SUMO network.

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 16: Four numbered panels showing a reference point coordinate table, an annotated overhead intersection photo with matching numbered points, a sample data table with agent positions and world coordinates, and an XML snippet of SUMO road network edge definitions.

Slide-16

Homography and network detail panels

Four numbered panels illustrate supporting data for the homography and network-building process:

Panel #1 is a table titled "Positions" with columns sumo_x, sumo_y, cosmos_x, cosmos_y, and Title, listing eight reference points (P1 through P8) with their SUMO and COSMOS pixel coordinates and descriptive titles, including "Left Turn Marking Am North," "Road Divider Am North," "Road Divider 120th East," "Road Stop Line Am South," "Road Divider Am South," "Road Divider 120th West," "120th Sidewalk (Lowell Hall corner)," and "Intersection Center."

Panel #2 shows two side-by-side overhead images of the intersection, the left an aerial photo and the right a rendered SUMO map, each with matching numbered yellow markers (P1 through P8) at the same physical locations, showing the reference points used for alignment.

Panel #3 is a data table with columns agent_id, agent_type, time, x_world, y_world, and matched_edge, listing sample rows of pedestrian and vehicle observations with their transformed world coordinates and the SUMO edge each was matched to.

Panel #4 shows a snippet of SUMO network XML code defining internal edges and lanes for a clustered intersection node, including edge IDs, lane indices, disallow and allow attributes (such as pedestrian, tram, rail, urban, delivery, bicycle), lengths, and shape coordinates.

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 17: Map Matching (Step 5), a data table showing pedestrian trajectory records with source file, scene id, agent id, track id, frame, time, position, and world coordinate columns.

Slide-17

Map Matching (Step 5)

A large data table shows trajectory records with columns source_file, scene_id, agent_id, track_id, frame, time, agent_type, pos_x, pos_y, x_world, and y_world. The visible rows show two pedestrian tracks (track_id 0.0 and 645.0) from the same source file "cosmos_231212_l20_f2p5_51054369864.npy," with sequential frame numbers and times (0.0 through roughly 7.6 seconds), their raw pixel positions (pos_x, pos_y), and their transformed real-world coordinates (x_world, y_world), illustrating how each trajectory point is mapped to the road network over time.

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 18: Section divider slide reading Results.

Slide-18

Results

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 19: SUMO simulation screenshot of a four-way intersection with vehicles approaching, annotated with a red arrow and text reading Accident about to happen.

Slide-19

SUMO simulation demo

A screenshot of the SUMO 1.27.0 application shows a "real world" rendered view of a four-way intersection with a roundabout-like center, with several yellow vehicle icons approaching from multiple directions along tree-lined roads on a green background. A red arrow points toward the center of the intersection with the annotation "Accident about to happen," highlighting a simulated conflict scenario.

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 20: Conflict Analysis slide listing three simulation capabilities, tracking agent movement, measuring close interactions, and locating conflict areas, alongside intersection screenshots and a diagram highlighting a conflict zone.

Slide-20

Conflict Analysis

With my new setup I am able to…

  • Track Agent Movement: Use TraCI to record each vehicle's and pedestrian's position, speed, and route during the simulation.
  • Measure Close Interactions: Calculate the distance between agents at each time step to identify when they come within a safety threshold, such as 3 meters.
  • Locate Conflict Areas: Map these close interactions to show where potential conflicts occur most often around the intersection.

Two side-by-side images at the top show a rendered SUMO intersection and a matching real overhead camera photo of the same intersection with pedestrians. Two arrows point downward from these images to a third image showing the intersection outlined in yellow with numerous small vehicle icons and labels clustered along the roads, representing tracked agent positions used for conflict analysis.

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 21: Section divider slide reading Conclusion.

Slide-21

Conclusion

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 22: Conclusion slide summarizing three project outcomes: reconstructing a real intersection from camera data, building a reusable digital twin pipeline, and demonstrating the benefits of combining digital twins with real world data.

Slide-22

Conclusion

In my project…

  • Reconstructed a real urban intersection from camera data.
  • Built a reusable digital twin pipeline.
  • Demonstrated the benefits of combining digital twins and real world data.

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Slide 23: Acknowledgement slide thanking the National Science Foundation, Center for Smart Streetscapes, I-SENSE, and CUNY Lehman College, with their logos displayed.

Slide-23

Acknowledgement

We want to express our gratitude the National Science Foundation (NSF) and Center for Smart Streetscapes (CS3) under NSF Cooperative Agreement No. EEC-2133516, and the Institute for Smarter Cities, Spaces and Health (I-SENSE), and CUNY Lehman College.

Logos shown: National Science Foundation (NSF), Center for Smart Streetscapes, 91ÖÆƬ³§ I-SENSE: The Institute for Smarter Cities, Spaces, and Health, and Lehman College.

Randy Amparo — Realistic Intersection-level Traffic Simulation Platform

Last slide: Contains plain text stating 'End of presentation. Click the right arrow to return to beginning of slide show.'

End of Presentation

Click the right arrow to return to the beginning of the slide show.

For a downloadable version of this presentation, email: I-SENSE@91ÖÆƬ³§.

Additional Information
The Institute for Smarter Cities, Spaces, and Health was established in early 2015 to coordinate university-wide activities in the Sensing and Smart Systems pillar of 91ÖÆƬ³§â€™s Strategic Plan for the Race to Excellence.
Address
91ÖÆƬ³§
777 Glades Road
Boca Raton, FL 33431
i-sense@fau.edu