
Picks discrete peaks from a DensityMap given a requested UInt64 count, locating local maxima as weighted centroids and emitting, by deploy-time arm, the top-N [Point<Double>], a peak count, or a [Landmark] list with density as confidence.
A crowd density map is reduced to per-head coordinates which are drawn back over the source frame for a live points overlay.
Recovered head points are tested against zone geometry to count or gate individuals per region instead of over the whole frame.
UInt64 input; when the map has fewer distinct local maxima than requested, fewer come out and there is no padding to the target.[Landmark] arm each peak's density value is carried as the landmark confidence; on the UInt64 arm only the count of found peaks is emitted.DensityMap input must be 2-D with shape (H, W) and a data length of exactly H times W; any other shape aborts mid-tick.window_size is the neighbourhood radius used to suppress non-maximal points: lower it to separate closely-spaced targets, raise it to merge noisy clusters; a value of zero is rejected at startup.Automated release