Pedestrian17.9%under 0.5 metres per second
A Street in One Minute
Watch one minute of people and vehicles move through a city block.
Try the project ↓How to use this project
Help
How to explore
Twelve cameras followed movement through four intersections near George Washington University for two hours on one sunny weekday afternoon. The data turn each detected path into a series of positions.
Play the minute
Watch the 4:12 p.m. replay, when all eight recorded road-user types appeared.
Choose who to follow
Select pedestrians, cars, bikes, or another type to update every view.
Read the full two hours
Compare the minute-by-minute count, movement speeds, and the places where tracks accumulated.
Keep the sample in bounds
Use the final notes to separate track IDs from people and this study period from typical traffic.
- Track ID
- A label for one detected path through the cameras; it is not a count of unique people or vehicles.
- Recorded moment
- One position and speed estimate for a track, sampled every tenth of a second in the source.
- Metres per second
- The source's speed unit; 1 metre per second is about 2.2 miles per hour.
- Automated test vehicle
- The one study car equipped with SAE Level 3 automation and driven through planned turns and stops.
Four intersections · one weekday commute
A street in
one minute
Watch people and vehicles move through one city block.
Twelve cameras followed movement around George Washington University from 3 to 5 p.m. on a sunny weekday. At 4:12, every recorded type appeared—from pedestrians to one automated test car.
60 secin the replay
8road-user types
2 hrin the full recording
One control changes every view
Whose movement do you want to follow?
Choose a type to filter the replay, the two-hour count, the speed bands, and the street footprint.
One minute at 4:12 p.m.
Each mark is a detected track at the selected moment. Press play or drag the time control. Changing the road-user filter removes the other types.
Local coordinates in metres4:12 p.m. + 0.0 sec
Pedestrian Bike or scooter Motor vehicle Automated test car
How the street count changed minute by minute
Each bar counts distinct track IDs seen during one minute. Taller bars mean more tracks, not necessarily more individual people or vehicles. The filter changes which type is counted.
3:01 p.m.4:00 p.m.5:00 p.m.
Busiest minute for all road-user types4:52 p.m.
294 distinct track IDs appeared. The striped bar marks the 4:12 replay.
How often each track was nearly still or moving
Each row splits the recorded moments into speed bands. Longer tracks contribute more moments. Use the filter to isolate one type.
Under 0.5 m/s0.5–2 m/s2–5 m/s5–10 m/s10 m/s or faster
Bicycle16.7%under 0.5 metres per second
Scooter0.0%under 0.5 metres per second
Passenger car48.7%under 0.5 metres per second
Automated test vehicle49.6%under 0.5 metres per second
Motorcycle0.0%under 0.5 metres per second
Bus42.8%under 0.5 metres per second
Truck69.5%under 0.5 metres per second
Read this as time in motion, not driving skill. The single automated test car was completing planned turns and stops. This recording cannot compare safety or typical behavior between groups.
Where the cameras kept finding movement
Each square groups recorded positions into a 10-by-10-metre area. Brighter squares contain more recorded moments. The type filter redraws the footprint.
0 metresLocal east–west position →200 metres
230 metresLocal north–south position ↓
How to read the source and its limits
The Federal Highway Administration project recorded four intersections near George Washington University with twelve synchronized cameras. Positions were estimated every tenth of a second.
A track ID belongs to one detected path. It is not a census of unique people or vehicles. The same physical road user can appear in more than one track; the project used one automated test vehicle even though the data contain several automated track IDs.
The data support a close look at this two-hour study period. They cannot describe a typical day, measure traffic volume for the neighborhood, or establish how automation changed anyone else’s behavior.
Why this exists
The question
What can a two-hour camera-derived trajectory sample show about the mix, timing, speed, and spatial footprint of movement through four Foggy Bottom intersections without treating one afternoon as typical traffic?
What to try
Play the 4:12 p.m. minute, then choose a road-user type to update the replay, two-hour pulse, speed bands, and street footprint together.
Method and limits
About the data
I found the Third Generation Simulation Data Foggy Bottom trajectories through Data.gov, then checked the underlying Federal Highway Administration dataset, its field definitions, the public-domain label, and the linked technical report. The report describes twelve synchronized 4K cameras covering four intersections and nearby road segments around George Washington University. The source says the footage covers 3 to 5 p.m. Eastern time on a sunny weekday and includes one test vehicle equipped with SAE Level 3 automation.
A reproducible Python reducer made nine HTTPS queries to the official dataset API on September 29, 2026. It reduced 2,492,261 position rows covering 7,200 seconds into a 134,075-byte static JSON file with SHA-256 651ff2ca101a1f40658278cf7dc5624019da28794349fd323b2886d379d132a0. The snapshot stores every query URL, response byte count, and response SHA-256. It contains type-level track and sample counts, distinct track IDs by minute, five speed bands, 10-by-10-metre position cells, and a half-second sample of 4:12–4:13 p.m. The browser reads only this checked-in snapshot and makes no network request.
The replay uses the source's Kalman-filtered local x and y coordinates. The source records positions every 0.1 second; the animation keeps one position every 0.5 second and rounds coordinates to 0.01 metre. The two-hour view counts distinct track IDs within each minute, so a track spanning two minutes appears once in each. Speed bands count recorded moments rather than tracks, which means longer tracks contribute more observations. “Nearly still” means a calculated speed below 0.5 metres per second. The footprint groups positions into 10-metre cells instead of drawing an aerial photograph or implying exact street geography.
A track ID represents a detected path, not a verified unique person or vehicle. One physical road user can contribute more than one track, which is visible in the multiple IDs assigned to the study's single automated test vehicle. Camera coverage, tracking loss, time in view, stops, repeat passes, and planned test maneuvers all affect the counts. This two-hour observation cannot establish typical traffic volume, compare safety between road-user groups, measure the neighborhood beyond the camera area, or show that automation caused another road user's behavior.
Skills used
- Data analysis
- Data engineering
- Statistical reasoning
- Systems design
- Visualization
- Information design
- Interaction design
- Visual design
- Creative coding
- Product thinking
- Editorial storytelling
- Accessibility
- Performance
What the data shows
- The source contains 20,511 track IDs: 15,307 labeled pedestrian and 4,547 labeled passenger car. These are detected paths, not verified counts of unique people or vehicles.
- At 4:12 p.m., all eight recorded types appeared in the same minute. Its 247 track IDs include 184 pedestrian tracks, 50 passenger-car tracks, and one automated-vehicle track; the busiest minute overall was 4:52 p.m. with 294 track IDs.
- Passenger-car positions were below 0.5 metres per second in 48.7% of their recorded moments. The single automated test car's tracks were below that threshold in 49.6% of moments, but planned maneuvers and repeated tracks make this unsuitable as a safety or performance comparison.