How these numbers are made
A phone at chest height films the walk. Two vision models label every half second of video (sidewalk, road,
curb, poles, vehicles, trees, sky) and estimate depth. The phone's motion tracking gives the camera height, the
distance walked and the slope; GPS places the walk on the map.
Each frame looks 1.5 to 15 m ahead. Everything it sees is filed at the spot it describes, not where the
walker stood, and every metre combines the ten or so frames that saw it, nearer views counting more. So a
lamp post 6 m ahead is a lamp post at that metre, not at the walker's feet.
Overhead cover is the share of the space 2.2 to 15 m above the path, within 2 m of it, filled by trees,
buildings, bridges or signs, against open sky seen through the same space. It does not depend on the hour.
Sun at walk time is whether the ground just ahead was sunlit when the walk was filmed, from the camera's
exposure: it depends on the hour, so it is not in the overall score.
Blocks come from OpenStreetMap: the street between two intersections. The side comes from where the video saw
traffic, which is more reliable than GPS for telling the two sides of a street apart. Metres inside an
intersection, including crossing the mouth of a side street, are crossings, not part of either block.
The overall score adds up the other measures, each scaled from 0 (the left end of its colour bar) to 100
(the right end), with these weights:
One walk is one moment: sun depends on the hour, puddles on the weather, parked cars on the day. Lines get more
trustworthy as more walks cover them.