A watch area is designed to refer to a place in the scene rather than a fixed patch of the screen as the camera moves.
This page documents the intended registration method. The public route plays a clip with per-frame zone geometry prepared ahead of time; it is not registering the zone live.
This distinction sounds small and is everything. A screen-space region silently changes meaning every time the drone drifts a meter. “Count cars entering this area” is only a coherent instruction if the area refers to a fixed place, so anchoring is treated as a core competency of the runtime rather than a rendering detail.
When the operator finishes drawing, the runtime samples the region: distinctive keypoints, a spread of trackable corner features, and a small template image of the patch itself. This bundle is the zone's visual fingerprint.
2
Follow, every frame
Optical flow tracks feature points from frame to frame and re-derives the polygon from their motion while the scene moves smoothly.
3
Reacquire, on failure
Flow breaks under fast pans, motion blur, or occlusion. The runtime then falls back to matching the stored keypoints against the current frame, and past that, to searching for the template patch directly. Whichever succeeds re-seats the polygon.
4
Sanity-check, always
Every proposed update has to pass plausibility tests before it's accepted: the implied transform must be geometrically reasonable, and the patch under the new polygon must still resemble the fingerprint (similarity above 0.64). A zone that can't verify itself reports lost instead of guessing.
The intended design can also use camera pose when telemetry is available. Visual registration is the baseline in this document; telemetry is an additional input, not a public compatibility claim.
The design rule for zones is that a wrong answer is worse than no answer. A zone that has lost registration greys out and stops producing events until it reacquires. It does not slide to wherever the pixels went. Operators learn to trust the overlay precisely because it admits when it doesn't know.
The challenge loads a prepared zone record for real boulevard footage with a moving camera. During playback, the browser draws the saved polygon for each frame. The prepared entry records were produced against that saved geometry, and you can download the polygon for every frame if you want to check the registration yourself.