android/FadCam
1
0
Fork 0
mirror of https://github.com/anonfaded/FadCam.git synced 2026-03-13 09:00:17 +08:00
Code Issues Packages Projects Releases Wiki Activity
Files
master
FadCam/DIGITAL_FORENSICS_ROADMAP.md
Faded 7da57ddedd Add new drawable resources and update layouts for forensic features 
- Introduced new vector drawables for forensic icons: `ic_forensics_pet` and `ic_forensics_vehicle`.
- Replaced info icon TextView with an ImageView in `fragment_faditor_mini.xml`.
- Updated layout in `fragment_forensic_intelligence.xml` to include an ImageButton for lab menu and adjusted constraints.
- Modified padding and layout properties in `fragment_forensics_events.xml` for better UI alignment.
- Created new layout `fragment_forensics_export_center.xml` for the export center functionality.
- Added `fragment_forensics_gallery.xml` layout for displaying evidence gallery with sorting options.
- Implemented `fragment_lab_sidebar.xml` for lab sidebar navigation with new sections.
- Enhanced `fragment_settings_digital_forensics.xml` with new UI elements for motion lab dependency and evidence collection settings.
- Created item layout `item_forensics_gallery.xml` for individual gallery items.
- Updated `strings.xml` with new strings for evidence gallery, export center, and digital forensics features.
2026-02-15 06:20:59 +05:00

7.6 KiB
Raw Permalink Blame History

Digital Forensics Roadmap

Summary

Add a new Advanced subsection named Digital Forensics (separate from Motion Lab) to deliver:

  • AI-based event intelligence
  • Reliable video-to-data linkage across rename/move/paste-back
  • Optional live preview AI overlay
  • Future cloud-sync-ready architecture

This roadmap is modular, realtime-oriented, and privacy-first.

Lab Tab Consolidation (Locked 2026-02-14)

  • Add 6th bottom-nav item: Lab
  • Add dedicated tab-root surface titled Forensic Intelligence
  • Move forensic browsing into Lab (Events + Insights sections)
  • Remove Events/Insights navigation rows from Digital Forensics settings (settings now control-only)
  • Keep single source-of-truth flow: detector -> recorder -> Room DB -> Lab UI
  • Bump forensics DB schema and use latest clean destructive reset policy for this refactor
  • Keep English-first strings for new UI labels in this phase

Hard-Cut Runtime Stack (Locked 2026-02-15)

  • Use EfficientDet-Lite1 as the only detector path in recording analysis
  • Remove overlay payload compatibility fallback logic in renderer path (strict LABEL|CONF|cx|cy|w|h|TYPE)
  • Shift forensic event persistence to real-time heartbeat writes (no stop-only dependency)
  • Add multi-event active aggregation keyed by media+type+class
  • Add persistent snapshot evidence table (ai_event_snapshot) + DAO
  • Add lifecycle metadata fields to ai_event (status, seen-range, sample count, peak confidence, alert fields)
  • Bump forensics DB schema to latest (version=5, destructive migration policy)
  • Move Lab root from tabbed pager to events-first single surface with top-right hamburger
  • Add Lab sidebar shell with Insights/admin destinations

Architecture Decision (Locked 2026-02-13)

  • Single detector source of truth: EfficientDet-Lite1
  • Legacy person-only detector removed
  • No detector fallback paths to legacy models
  • Preview overlay only (never burned into encoded video)
  • Detector runtime uses TFLite Task Vision API (metadata-driven labels/output parsing, no manual label file)

Product Positioning

  • Keep Advanced section unchanged.
  • Add a new feature card: Digital Forensics.
  • Keep Motion Lab as separate feature.
  • Discord alerts: Coming Soon placeholder for now.

Primary Requirement: Persistent Video Identity Linkage

Goal

A video should retain linked forensic/event data after:

  • rename
  • folder move
  • export/import back into FadCam

Identity Strategy

  1. Exact Fingerprint
  • sampled SHA-256 blocks
  • file size
  • duration
  • codec/container profile
  1. Visual Fingerprint
  • keyframe perceptual hash (pHash/aHash)
  1. Origin Hints
  • first-seen path/name/timestamps

Match Policy

  • Exact match -> auto relink
  • High-confidence similarity -> Probable relink (auto + badge)
  • Low confidence -> create new media record

Metadata Embedding

  • Write media_uid into video metadata where possible (best-effort)
  • Never rely only on metadata (it can be stripped)
  • DB fingerprints remain source of truth

Data Model (Room SQLite)

media_asset

  • media_uid (UUID PK)
  • current_uri
  • display_name
  • category/subtype
  • size_bytes
  • duration_ms
  • codec_info
  • exact_fingerprint
  • visual_fingerprint
  • first_seen_at
  • last_seen_at
  • link_status (EXACT, PROBABLE, NEW)

ai_event

  • event_uid (UUID PK)
  • media_uid (FK)
  • event_type (PERSON, VEHICLE, PET, DANGEROUS_OBJECT)
  • start_ms
  • end_ms
  • confidence
  • bbox_norm
  • track_id
  • priority
  • thumbnail_ref

integrity_link_log

  • log_uid
  • media_uid
  • action (LINKED_EXACT, LINKED_PROBABLE, UNLINKED, MISMATCH)
  • score
  • timestamp

sync_queue (future)

  • op_uid
  • entity_type
  • entity_id
  • operation
  • payload_json
  • status
  • retry_count

Feature Toggles (Digital Forensics screen)

  • Enable Digital Forensics
  • Event classes (Person / Vehicle / Pet)
  • Dangerous object tags toggle (opt-in)
  • Show AI overlay in live preview
  • Daily summary (always-on analytics)
  • Heatmap insights (always-on analytics)
  • Discord alerts (Coming Soon placeholder)

Phase-by-Phase Execution

Phase 0 - UX Shell + Navigation

  • Create Digital Forensics card in Advanced
  • Add settings screen skeleton
  • Add Coming Soon Discord row
  • Add helper texts and safety disclaimers

Deliverable: Navigable Digital Forensics settings UI (no backend inference yet).

Phase 1 - DB & Identity Foundation

  • Add Room DB module, entities, DAOs, migrations (baseline v1 schema)
  • Implement exact fingerprint generator (sampled SHA-256 blocks)
  • Implement visual fingerprint generator (keyframe aHash baseline)
  • Build relink engine (exact + probable baseline scoring)
  • Integrate with existing Records scan pipeline (async background hook)
  • Add link status badges support in data layer (link_status stored)

Deliverable: Persistent asset identity with robust relink behavior.

Phase 2 - Event Capture Integration

  • Extend existing analysis pipeline to emit structured events (motion START/STOP transitions)
  • Persist events to ai_event
  • Store representative thumbnails (timeline reference via thumbnail_ref)
  • Keep dangerous-object tagging opt-in only
  • Ensure motion-trigger path remains stable (async, no control-flow changes)

Deliverable: Event metadata linked to media assets.

Phase 3 - Events UX

  • Add Events view in Records context
  • Add filters (type, date, confidence)
  • Tap event -> seek exact timestamp and open paused (no autoplay)
  • Show link quality badge (Exact/Probable)
  • Show proof frame thumbnail for each event row

Deliverable: Searchable evidence timeline experience.

Phase 4 - Insights + Overlay

  • Add daily summary cards
  • Add heatmap from bbox centers
  • Add optional live overlay rendering (auto labels baseline)
  • Ensure overlay-off mode has minimal overhead

Deliverable: Intelligence layer without manual drawing workflow.

Phase 5 - Hardening

  • Rename/move/import stress hardening (multi-signal relink guards + confidence threshold)
  • Re-encode similarity baseline validation path (visual hash probable relink)
  • Mid-range thermal/performance safeguards (async indexing/event writes, overlay toggle off by default)
  • Regression checks for Motion Lab + Records (compile + non-blocking hooks)
  • DB migration and recovery tests baseline (migration 1->2 added for new event metadata)

Deliverable: Production-ready reliability baseline.

Acceptance Criteria

  • Rename/move keeps data linked automatically
  • Paste-back relinks via exact/probable match
  • No false hijack of unrelated videos
  • Events open correct video timestamps
  • Motion Lab remains independent from Digital Forensics toggles
  • Mid-range realtime performance remains acceptable

Risks & Mitigations

  • Re-encoded files lose exact match
    Mitigation: visual fingerprint + probable-match badge

  • False probable relink
    Mitigation: confidence thresholds + link log + optional review queue later

  • Thermal increase with inference/overlay
    Mitigation: adaptive FPS and defaults OFF

Files to Maintain for Context

  • DIGITAL_FORENSICS_ROADMAP.md (master roadmap)
  • DIGITAL_FORENSICS_PHASE_LOG.md (ongoing decisions/findings)
  • MOTION_LAB_ROADMAP.md (separate, cross-link only)

Defaults

  • Digital Forensics: OFF by default
  • Dangerous object tags: OFF by default
  • Overlay: OFF by default
  • Match policy: Exact + Probable (with badge)
  • Discord alerts: Coming Soon (no active network flow)
Reference in New Issue View Git Blame Copy Permalink