vision_ai_pose 0.1.0

On-device body pose detection (33-point skeleton) with pure-Dart fitness/posture analytics and skeleton overlays for Flutter, built on the vision_ai engine.

0.1.0

Initial release — on-device body pose detection for Flutter, built on vision_ai ^0.4.0.

Detection

• 33-point body pose — per frame, a full skeleton via the MediaPipe Pose Landmarker (LIVE_STREAM). Each PoseResult carries normalized landmarks (image coords + visibility/presence) and worldLandmarks (metres, origin at the hip midpoint).
• PoseVision.create(...) ergonomic session wrapper (start / stop / results / updatePoseConfig / switchCamera / dispose); PoseModels.ensureLoaded() copies the bundled model into app storage and returns its path.
• PoseLandmarkType index constants (nose=0 … rightFootIndex=32) and poseConnections (the canonical MediaPipe bone topology) for drawing skeletons.

Analytics (pure Dart, zero model weight)

• jointAngle(a, b, c, {use3D}) and PoseAngles.of(pose) — elbow/knee/hip/shoulder angles (left + right), 2D or 3D; torsoTiltFromVertical(pose) for torso inclination.
• RepCounter — repetition counting from a joint-angle down→up oscillation with hysteresis (RepState count + phase).
• PostureClassifier — upright / leaning / crouching / lyingDown / unknown.
• FallDetector — latched fall detection (horizontal torso + rapid hip drop).
• isPoseReliable / landmarkVisible — gate analytics on landmark visibility.

Overlays (pure Flutter, zero model weight)

• PoseCameraView — camera Texture + live skeleton overlay (with an overlayBuilder escape hatch).
• PoseSkeletonPainter — bones + landmark dots, mirroring + visibility-aware fading; PoseOverlayStyle tunes colors/widths/min-visibility.
• RepCounterOverlay — binds a RepCounter to a live count/phase badge.

Bring your own model / variant

• PoseVision.create(customModelPath:) swaps in a different Pose Landmarker variant (lite / heavy) or a fully custom .task (skips the bundled-asset copy).

Bundled model

• pose_landmarker_full.task (the full variant) — the accuracy/latency middle ground for live fitness/posture. Swap to lite or heavy via customModelPath.

Platforms

• Android and iOS (symmetric with the rest of the vision_ai ecosystem).