Browse a neighborhood
Open the explorer, pick a neighborhood from the dropdown, and the map zooms to its detected intersections, traffic controls, and sidewalks.
MUSA Practicum · Spring 2026
Crossing conditions deserve their own map.
We built a city-scale screening workflow that pairs aerial imagery, Street View, and computer vision to surface crosswalks, crossing widths, and traffic controls across Philadelphia. So planners can see the street the way pedestrians actually meet it.
9,950
Street View images processed
3,745
Traffic-control detections
3,610
Crosswalk polygons generated
8
Philadelphia neighborhoods piloted
Why this matters
Walk Score stops at the curb.
Destination access is only half the story. If a crossing is missing, faded, wide, or uncontrolled, the route isn't truly walkable, even when the neighborhood scores high.
Cities usually have either coarse GIS inventories that age out quickly, or expensive field surveys tied to individual capital projects. There's very little in between. This work fills the gap: a repeatable screening layer that refreshes whenever the imagery does.
PASDA aerial imagery
Google Street View
YOLO + U-Net
GeoJSON · CSV
Who this is for
Planners, advocates, and DOT teams who need crossing-level evidence.
You're prioritizing pedestrian capital projects, writing a Vision Zero memo, or scoping a corridor study. You need to know where the missing crossings are, not just that some exist somewhere.
This site gives you the screening layer plus the methodology to defend it in a meeting.
How to approach this tool
Three steps, no GIS heroics required.
The workflow is built around inspect-and-export, designed to slot into how planning teams already work.
Verify in Street View
Click any point to see what the model saw. Each feature links straight to Google Street View so you can confirm before citing it in a memo.
Export & plan
The same outputs are available as GeoJSON and CSV. Drop them into ArcGIS or QGIS, join to your priority corridors, and act on the gaps.
Under the hood
Two computer-vision models, one planning interface.
The pipeline does the heavy reading; the site keeps the outputs in a form you can defend, share, and refresh.
Street View · Object detection
YOLO 101: finding stop signs & lights
How a single-pass detector sweeps intersection-facing photos, what it caught (and missed), and why F1 of 0.77 is good enough to plan with.
Read the explainer →Aerial imagery · Segmentation
U-Net 101: outlining the crosswalks
Why a "U-shaped" neural net is unusually good at marking every pixel that belongs to crosswalk paint, and how it produced 3,610 usable polygons.
Read the explainer →New to deep learning? Start with the planner-friendly overview →
Ready when you are
See the city the way pedestrians do.
Open the explorer, pick a neighborhood, and start auditing crossings in the next five minutes. No login, no install. Just the map.