Detect Human Posture (RuleBased PPL)

1 VersionDetect

Rule-based posture gate: consumes Landmarks2d.Human22 keypoints plus the frame width and height and emits Bool true when the geometric score for the chosen posture clears a threshold. No ML model — a CPU-only deterministic gate over a pose estimator.

How it fits

Typical backends

Fall / lying alert: a pose estimator feeds keypoints, the posture gate flags the target pose, a filter drops the false ticks, and surviving alerts are emitted as JSON.

Sit-stand monitor: a YOLO pose backend supplies keypoints and the gate reports whether the subject is in the configured posture each tick.

Caveats

LimitationSingle-person input; multi-person scenes must be fanned out per pose upstream.
Hard constraintFrame width and height must both be positive; a zero dimension aborts the tick, and an unsupported target_posture aborts startup.
I/O contractA keypoint whose confidence is below visibility_threshold is treated as not visible, and any posture whose score needs a missing keypoint collapses to zero and emits false.
AccuracyScores are unit-less and roughly in the zero-to-one range but are not calibrated probabilities, so score_threshold must be tuned per deployment; the heuristics also assume a roughly upright camera and produce nonsense at side or rotated angles.
State lifetimeAll settings are captured once at startup; changing them requires a restart.

Versions

4f802bcclatestdefault
Automated release
6/25/2026