The scene renders at CSS resolution and is stretched on a high-density
display, so a count marker drawn as a scene sprite looked coarse next
to the camera callouts, which the world overlay paints at device
density.
Paint the markers through the overlay's custom paint lane instead, in
the callouts' typeface. Each paint reprojects the group centres; a
click is resolved against the rectangles painted in the last frame and
still zooms to the group without choosing a camera. A marker that would
cover HUD chrome is skipped, as overlay entries are.
With no aggregate sprites left, the contact sweep no longer needs to
skip them, so that exclusion is removed.
Zoomed out, the layer keeps one camera per ground cell and drew it as an
ordinary icon, so nothing showed that dozens or hundreds of other cameras
sat behind it. A cell with two or more cameras is now one round marker
with the count inside, at the centre of its cameras. Single cameras and
the active camera keep their icons.
The count covers the whole cell, so it does not change while the cell
crosses the screen edge. Markers that would overlap on screen merge into
one carrying their sum. Counts above 999 use the compact form of the UI
language and widen the marker into a pill. Clicking a marker zooms to
its group and does not choose a camera.
Ambient cards and coverage attach only to cameras drawn as icons; the
camera picker keeps one representative per visible cell as before.
Brings in the CCTV opening camera, the Iceland, Giroc HLS, Taiwan, Quebec,
Ottawa and world camera packs, the world clock and timeline, the zoom-out
limit and registered-only layer state, ported onto the renderer contract:
the sky follows the world time through the frame port's setTime in
preUpdate, the Quebec overlay reads the eye, projects and redraws through
the camera, query and frame ports, launches hide pending missions through
their draw records, and mission orbit rings turn on the scene clock.
The camera icons are now a draw-port icon set kept by an icon book; only
icons a view pass or restyle changes are written, in one write per pass.
Picks read the renderer's hit, the view pass and focus pass project
through the camera pose (geo/frustum.js poseProjector, matching the
renderer's projection and near clip) and the query port, and tile
readiness for ground sampling comes from the globe port.
The frustum wireframes, viewshed volumes and the active camera's monitor
plane are now line, mesh and quad items of the layer's own draw sets
(layers/cctv/drawSets.js), written from plain handles only as they change.
The plane takes its canvas double buffer or its video as a quad source.
Picks of lines and planes are matched by set, item id and the record that
draws them. The activation obstruction probe casts through the query
port's rayCast; its clamp lands a microtask later, rewrites the geometry
and asks again for the footprint ground, and a newer activation, a range
edit, a deactivation or a re-init drops an answer still to come.
The gizmo's rings and move arrows are draw-port lines that dim where
geometry hides them, and its handles and arrow tips are depth-test-free
draw-port points; their item ids keep the cctv-gizmo-<part> scheme the
layer's pick owner matches. Picks, rays and the cursor go through the query
port and the renderer's canvas, and the drag math runs on plain vectors.
The layer reports what the gizmo draws for the QA harness, which reads the
parts there instead of from the viewer's entities.
Turning CCTV on picked the first catalog record, so at King's Cross it
showed a Sydney camera. enable() now picks the camera nearest the viewer.
Until someone chooses a camera, each settled move re-picks the nearest,
so a share link whose flight lands after CCTV turned on still opens
nearby.
The frustum's view axis comes from geo/frames' east-north-up frame, and
the activation probe's mount point and hit distance from geo/ellipsoid and
geo/vector; a new test checks the axis and the monitor plane's orientation
against the renderer's own transforms, bit for bit. The mesh-floor
sampling reads the eye through the model's camera-port helper.
What the geometry part did with Cesium moves, unchanged, into the
rendering part, which owns the layer's Cesium drawing: the positions the
entities take, building the five coverage lines, building and removing
the viewshed volumes, the photoreal tileset readiness check, and the
probe's one ray pick, which now goes through a small rendering helper.
Geometry leaves the allow-list.
The layer's colours are renderer colours; the parts that still draw with
Cesium (icons, coverage lines, the monitor plane) turn them into Cesium
colours with the adapter's own conversion. The screen and matrix scratches
are plain objects, which is all the projections that fill them write.
The model part converts angles with geo/vector, reads where the eye is
from the camera port's pose, and builds the monitor plane's orientation
with geo/quaternion from the same axes; the projection part hands the
entity a Cesium quaternion of it. The photoreal tileset readiness check,
which looks through the scene's primitives, moves to the geometry part.
The ground pass moves a camera icon by writing geo/ellipsoid's position
into the record's own position object. The test plane's plain material
comes from the projection part, and the selection part only lost an
unused import.
All give the same numbers as the Cesium calls they replace. Policy,
state, selection, testing, model and ground leave the allow-list.
The CCTV layer kept every loaded camera live: each view change touched every
icon, every notification rebuilt every camera's state, and the picker listed
the whole catalog. With the default 4,000 cameras, a continental drag ran at
152-179 fps with 18-24 frames over 30 ms (worst 112 ms). It now runs at
235 fps with none. The layer now:
- shows icons only on or near the screen, and when zoomed out keeps one camera
per world-anchored cell (healthy feeds first), so panning keeps the same
cameras;
- grounds and draws coverage only for the cameras in view, the active camera
and its coverage neighbours, retrying a camera at most twice per view regime;
- reruns the view pass while the camera moves, paced by its own cost, and
projects with one matrix per pass instead of a Cesium call per camera;
- keeps the billboard buffer-usage hint fixed, so showing or hiding a few
icons no longer rewrites the vertex data of every billboard;
- lists the cameras in view (active camera first) in the picker, closing with
the count of cameras outside the view, and builds the full camera list in
the UI state only when it is read.