01 · Baseline input
4 views / point
11,032
Google car views
Four directional Street View images at 2,758 road-centre sample points form the traditional GVI baseline.
The camera problem
Google Street View drives the road.
Pedestrians walk the sidewalk.
01 / Move the camera
Traditional GVI inherited the driver’s seat.
Google Street View is usually captured from a vehicle in the road. Its Green View Index is therefore a car-level measure – affected by road width, parked vehicles and distance from sidewalk planting.
Google car-based GVI
Mapillary sidewalk PGVI
Two sampling positions. Two green views.
Traditional GVI observes the city from the road centre. PGVI moves the camera to the sidewalk, where planting, parked vehicles, road width and pedestrian-only space enter the frame differently.
0.56
Car-based GVI
perception correlation
perception correlation
0.59
Sidewalk PGVI
perception correlation
perception correlation
02 / Method
Move from the road to the sidewalk, then test the difference.
The workflow compares vehicle-based GVI with sidewalk-based PGVI, adds citywide spatial visibility, and validates all three against volunteers’ lived perception.
02 · Pedestrian input
camera on sidewalk
PGVI
Mapillary imagery
Pedestrian-level images capture sidewalk planting, cycle paths, pedestrian streets and narrow alleys.
03 · Viewpoint test
GVI
Δ
PGVI
same street · two views
9
street green types
The same typologies are read from both the vehicle lane and the sidewalk.
04 · Spatial model
geometry → visibility
VGA
spatial visibility
3D tree attributes and building geometry become a planar visibility field.
05 · Human test
7 perception indicators
183
volunteer evaluations
Seven streetscape indicators ground the computational outputs in perception.
03 / London field
The official City boundary frames 2,758 road-centre points.
The City of London boundary comes from the Greater London Authority’s official GIS service. Inside it, the study’s 50-metre sample points each represent four Google Street View directions; colour encodes vehicle-based GVI.
Official City of London boundary
2,758 points · 50 m interval · 11,032 images
Select a point to inspect its road-centre GVI.
Select a point to inspect its road-centre GVI.
The reason PGVI was needed
When the car and sidewalk disagree, whose green view are we measuring?
04 / Street lab
Nine streets, seen from the lane and the sidewalk.
Select a type to compare the actual Google car view with the pedestrian sidewalk view, then read the GVI-PGVI gap alongside spatial visibility and human ratings.
T1 · no greenery · buildings on both sides
Street type T1
Built edges, no greenery
A useful baseline: geometry remains visible, while almost no green enters the pedestrian frame.
PGVI · sidewalk
0.0004
GVI · Google car
0.003
VGA-GVI · spatial
315.50
Human rating
6.46
Bars are normalized within each metric across the nine sampled typologies. Values retain the published units.
Mapillary
Pedestrian imagery made a sidewalk metric possible.
Through collaboration with Mapillary, the study could capture sidewalk greenery, pedestrian-only streets, cycle paths and narrow alleys that conventional car-based imagery often misses.
The viewpoint changes the GVI.
Vehicle-level Google Street View and sidewalk-level Mapillary imagery do not frame the same greenery. The published segmentation table exposes the mismatch across all nine types.
05 / What aligned
The sidewalk view comes closest to human perception.
Mapillary-based PGVI showed the strongest relationship with perceived green visibility. Traditional car-based GVI can underestimate greenery when the vehicle camera is distant from or obstructed from sidewalk planting.
Correlation with volunteer ratings of perceived green visibility. A higher value means the computational measure more closely follows what participants reported.
0.72
Safety
Green is noticed more when the street also feels secure.
0.69
Transparency
Clear sightlines reinforce the visibility and legibility of greenery.
0.53
Walkability
A pedestrian-friendly street makes green experience easier to reach.
One relationship at a time.
Explore the full published correlation matrix. Hover or focus a cell to read the variables and coefficient.
Explore: choose a cell in the matrix.
positivenegative
06 / Design implications
01
Measure from the sidewalk, not the traffic lane.
Car-based imagery remains scalable, but any claim about pedestrian experience should be checked from the pedestrian’s actual position.
02
Mix heights and occupy both sides.
Trees combined with shrubs or grass, especially on both sides of a street, produce stronger pedestrian visibility.
03
Design the conditions around green.
Safety, visual transparency and walkability shape whether visible greenery becomes a positive human experience.