Drop a queue-geometry.json to begin.
queue-geometry.json here — and optionally a
preview-cloud.json for the 3D replay.Looking down, north up, equal scale. The corridor is coloured by where each lane width came from. It opens out at U-turns because the apex has no rail beside it — the ray runs on down the next lane.
Shaded spans are interpolated — no usable depth there, so the width is inferred from its neighbours.
Everything here is drawn from measurements in queue-geometry.json and
preview-cloud.json — nothing is invented.
The place as it was photographed, with the lane laid over it in the same colours the phone draws live while you walk: translucent floor, ankle-high curbs, dashed where the width had to be inferred. Drag to turn, scroll or pinch to zoom, two fingers or right-drag to move.
The lane is on by itself to start with. Solid surfaces rebuilds the place around it in 10 cm × 10 cm × 5 cm cells, each drawn only where three depth returns from two different frames agreed something was there, coloured by the photographs those returns landed in. Nothing spans a gap and nothing is smoothed: a hole in the blocks is a hole in the measurement, and what stays as loose dots under The scan is measurement no second frame confirmed.
Eye level stands you on the lane centre at 1.62 m and points you down it. Click into the view, then ↑ and ↓ walk the path, ← and → look around, shift hurries, drag turns your head, escape leaves. You are on rails on purpose: the path is the thing being checked, and a free camera lets you drift off it without noticing.
What each edge was made of — rope, rail, wall, curb — is in the plan view above and the stations table below. The curbs here are drawn one height on purpose: standing walls at their real height walled the space off from the camera looking into it. Phone & aim adds the depth camera’s frustum every sixth pose, and the playback head carries a full one — that pyramid is the volume anything could be measured in, so a rail outside it was never seen however close the walk passed to it.
Floor, rails and posts, reduced from the depth points as the walk progresses. Drag to orbit.
preview-cloud.json loaded. Generate one with
python -m queuewalk.preview <recording> --out preview-cloud.json, then drop it here.
The walk is measured; where it sits on the Earth is fitted, and the fit can be out by metres and turned by degrees. Click a recognisable spot on the walk, then the same spot on the map. Two pairs is enough to solve it, three or more tells you how well.
Straight runs and turns, in walk order. Banks are groups of parallel runs.
The table view — every value in the charts above, readable without colour.
What the pipeline could and could not do at the time. This is how you tell months later whether a number can be trusted.