feat: dual-modal WASM browser pose estimation demo (ADR-058)

[ruvnet]

Summary

• Dual-modal pose estimation demo combining live webcam video (CNN) with WiFi CSI in the browser
• ADR-058 documents the architecture: ruvector-cnn-wasm for CNN inference, attention-weighted embedding fusion, three operating modes (Dual/Video-only/CSI-only)
• Pre-built WASM package (~52KB) from ruvector-cnn-wasm with JS fallback for immediate deployment
• 8 JS modules: CNN embedder, video capture, CSI simulator, fusion engine, pose decoder, canvas renderer
• GitHub Pages deployment live alongside Observatory at /pose-fusion.html

Architecture

• Conv2D 3×3 → BatchNorm → ReLU → GlobalAvgPool → Linear → L2 Normalize (mirrors ruvector-cnn)
• Per-dimension attention fusion with dynamic confidence gating
• 17 COCO keypoints decoded from fused 512-dim embedding
• CSI pseudo-image encoding (amplitude/phase/temporal → 224×224)

Files Added

• docs/adr/ADR-058-ruvector-wasm-browser-pose-example.md — Architecture decision record
• ui/pose-fusion.html — Main demo page (dark theme, responsive)
• ui/pose-fusion/js/ — 7 JS modules (main, cnn-embedder, video-capture, csi-simulator, fusion-engine, pose-decoder, canvas-renderer)
• ui/pose-fusion/css/style.css — Full dark theme CSS
• ui/pose-fusion/pkg/ — Pre-built WASM package
• ui/pose-fusion/build.sh — WASM rebuild script
• ui/index.html — Updated with Pose Fusion nav link

Test plan

• Open pose-fusion.html locally — verify webcam prompt, skeleton rendering, CSI heatmap
• Toggle between Dual/Video/CSI modes — verify each works independently
• Check fusion confidence bars update in real-time
• Verify WASM auto-detection (loads JS fallback if WASM unavailable)
• Verify GitHub Pages deployment at https://ruvnet.github.io/RuView/pose-fusion.html

🤖 Generated with claude-flow