Enciclopedia Boulder v1.0

Inicio · Capítulos · 14. Señales para visión artificial

15.99 14.4.24 Pipeline de procesamiento

Etapa ID Acción
1 TIMELINE_SYNC Load the common monotonic timeline and synchronization uncertainty from 14.1.
2 QUALITY_MASK Create per-stream validity, occlusion and tracking-loss masks.
3 CONTACT_EVENT_INGEST Ingest ContactEpisodes, states, load estimates and graph changes from 14.3.
4 FEATURE_ALIGNMENT Align pose, CoM, scene-relative geometry, IMU, force and external markers.
5 FEATURE_FILTERING Filter and differentiate with declared causal/acausal estimators.
6 BOUNDARY_CANDIDATES Generate candidate events from contacts, dynamics, changepoints and learned boundary scores.
7 EVENT_FUSION Fuse candidates into probabilistic BoundaryEvents with latency and uncertainty.
8 LANE_INFERENCE Infer AttemptPrimary, WholeBodyDynamics and per-effector state posteriors.
9 CONSTRAINED_DECODING Apply transition, duration, anchor, nesting and simultaneity constraints.
10 EPISODE_ASSEMBLY Build source-to-target MoveEpisodes and preserve overlaps/composites.
11 OFFLINE_REFINEMENT Optionally revise boundaries with future context and sensor evidence.
12 METRIC_EXTRACTION Compute durations, rhythm, contact-transition and quality metrics without classifying movement prematurely.
13 QUALITY_GATE Validate coverage, grammar, uncertainty, source references and structural preservation before sealing.

El pipeline comienza después de pose, escena y contacto. No debe recalcular IDs de presas o contacto con semántica incompatible dentro del segmentador.