Skip to content

Field Investigation — Physical Vessel Measurement, Evidence Capture & Secure Reporting

Implementation scope: This workflow is implemented in edgesentry-rs and edgesentry-app — not in this repository. This document is a requirements and design proposal to guide that implementation, and to complete the end-to-end picture for the Cap Vista proposal.

System Overview

Port Operations Centre
  ├── candidate_watchlist.parquet  ← from this repo (arktrace)
  ├── Live AIS feed
  └── VDES shore station
       │  task assignment + watchlist JSON (VDE-TDM ch 26/86)
       ▼
Patrol Vessel / UAV
  ├── Phase 1: Remote Surveillance
  │     ├── AIS live track of target (cross-check vs watchlist)
  │     ├── EO camera (long-range) → vessel identification at distance
  │     └── Decision: proceed to close approach Y/N
  │
  └── Phase 2: Close-Range Investigation
        ├── Sensor stack (tiered — see below)
        ├── Edge PC (edgesentry-app)
        │     ├── AI inference: hull marking OCR / 3D deviation
        │     ├── Evidence bundle (photos, point cloud, detections)
        │     └── edgesentry-audit: Ed25519 sign + BLAKE3 hash chain
        └── VDES report transmission → Port Operations Centre (ASM ch 24/84)

Phase 1: Remote Surveillance

Inputs

  • candidate_watchlist.parquet — MMSI, last known position, confidence score, top signals
  • Live AIS stream (aisstream.io WebSocket on patrol vessel)

Capabilities Required (edgesentry-app)

Capability Detail
Watchlist loader Ingest candidate_watchlist.parquet, display ranked candidates on chart
AIS cross-check Overlay live AIS track of target MMSI; flag if broadcast position diverges from predicted track
Long-range EO viewer Display feed from optical camera, capture screenshot with GPS + timestamp
Go/no-go decision log Duty officer records decision with reason; signed as an edgesentry-audit event

Hardware (remote surveillance tier)

Item Approx. cost Purpose
Long-range PTZ camera (e.g. Hikvision DS-2DE4425IWG) ~$400 Optical ID at 2–5km
Ruggedised tablet (Samsung Galaxy Tab Active) ~$600 edgesentry-app display
GPS receiver (u-blox M9N) ~$50 Position tagging

Phase 2: Close-Range Investigation

Measurement Tiers

All tiers feed into the same edgesentry-app pipeline and produce the same signed evidence bundle format.

Tier 1 — Camera + OCR (~$500 total hardware)
Hardware Hi-res camera (Sony RX100 or GoPro Hero) + GPS logger
Software OpenCV + Tesseract OCR
Detects Vessel identity: IMO number, vessel name, call sign, flag markings
Accuracy Sufficient for identity cross-check; no structural measurement
Use case First-pass ID; all deployments; minimum viable 
Camera frame → OpenCV preprocessing → Tesseract OCR
  → extracted: {imo_number, vessel_name, call_sign}
  → cross-check vs DuckDB vessel registry (Equasis + ITU MMSI)
  → match/mismatch flag → evidence bundle

Tier 2 — LiDAR + Point Cloud (~$3k–$8k hardware)
Hardware Livox Mid-360 (~$1k) or Ouster OS0-32 (~$3k) + NVIDIA Jetson Orin field PC
Software trilink-core (point cloud projection/deviation) + edgesentry-inspect
Detects Hull shape deviation vs registry silhouette; structural damage; waterline position (load estimation)
Accuracy 5–10mm deviation at 10m range
Use case Structural verification; draught/load cross-check; high-priority targets 
LiDAR scan → PointCloud
  → trilink-core::project_to_depth_map / height_map
  → AI inference (edgesentry-inspect): surface_void, deformation, hull_marking
  → trilink-core::unproject → world-space detections
  → deviation vs registry hull reference (PLY)
  → heatmap PNG + report JSON → evidence bundle

Tier 3 — Multi-spectral / Thermal (~$10k–$30k hardware)
Hardware FLIR Boson+ thermal camera (~$3k) + hyperspectral imager
Software Custom ONNX model; edgesentry-app HTTP inference endpoint
Detects Engine heat signature (running vs declared idle); cargo type proxy; night operation
Accuracy Qualitative; complements Tier 1/2
Use case High-value interdiction; night ops; UAV payload

Cost Summary

Tier Hardware cost Capability Recommended for
1 — Camera + OCR ~$500 Identity cross-check All deployments
2 — LiDAR ~$3k–$8k Identity + structural High-priority; patrol boats
3 — Thermal/multispectral ~$10k–$30k Identity + cargo + night High-value; UAV payload

Recommended starting point: Tier 1 + Tier 2 on a standard patrol vessel (~$8k total), Tier 3 as optional UAV escalation.

Evidence Bundle Format

All tiers produce a signed JSON bundle consumed by edgesentry-audit:

{
  "mmsi": "123456789",
  "imo": "9876543",
  "vessel_name_observed": "STAR PHOENIX",
  "vessel_name_ais": "OCEAN GLORY",
  "flag_observed": "PA",
  "flag_ais": "KH",
  "position": {"lat": 1.2631, "lon": 103.8200},
  "timestamp_utc": "2026-04-15T08:34:00Z",
  "tier": 2,
  "detections": [
    {"class": "imo_number", "value": "9876543", "confidence": 0.94},
    {"class": "name_mismatch", "value": true}
  ],
  "heatmap_png": "<S3 key>",
  "photos": ["<S3 key 1>", "<S3 key 2>"],
  "inspector_id": "officer-007",
  "watchlist_confidence": 0.87,
  "watchlist_top_signals": [
    {"feature": "ais_gap_count_30d", "value": 12, "contribution": 0.31}
  ]
}

This JSON is wrapped in an edgesentry-audit AuditRecord: - Ed25519 signature — device key in hardware secure element; proves record origin - BLAKE3 hash chain (prev_record_hash) — deletion or reordering is cryptographically detectable - Sequence number — strictly monotonic; replay detectable

Secure Reporting: VDES

Why VDES

VHF Data Exchange System (ITU-R M.2092) operates in the maritime VHF band (156–174 MHz): - Data rate: ~307 kbps on ASM channels - Range: ~40nm ship-to-shore; ~200nm via VDES-SAT - No internet dependency — critical for grey-zone operations - Integrates with existing AIS infrastructure

Channel Plan

Channel Use
ASM 1/2 (ch 24/84) Evidence bundle transmission (signed JSON + heatmap) — patrol → shore
VDE-TDM (ch 26/86) Watchlist task download — shore → patrol vessel
AIS (ch 87/88) Vessel tracking continuity (standard AIS)

Message Flow

Patrol Vessel (edgesentry-app)
  1. Complete evidence bundle → edgesentry-audit AuditRecord (Ed25519 signed)
  2. Compress: zstd (JSON) + JPEG (heatmap)
  3. Fragment into VDES ASM frames (max 1023 bytes/frame)
  4. Transmit on ASM ch 24/84 with ARQ (automatic repeat request)
       │
       │  VHF (~40nm) or VDES-SAT (global)
       ▼
Shore Station (VDES transceiver + REST gateway)
  5. Reassemble frames → AuditRecord
  6. edgesentry-audit IngestService::ingest()
       ├── verify Ed25519 signature
       ├── verify BLAKE3 hash chain
       └── reject if out-of-sequence or tampered
  7. Store to S3 Object Lock (WORM) — immutable evidence
  8. Push alert to Port Operations Centre dashboard

Fallback Channels

Channel Bandwidth Range Use case
VDES ASM ~307 kbps ~40nm / global (SAT) Primary — full evidence bundle
Iridium SBD 340 bytes/msg Global Alert only — MMSI + confidence score
LoRaWAN ~250 bps ~5–15km Harbour area — minimal alert
4G/5G LTE Broadband Port area Fallback when within cellular range

edgesentry-rs / edgesentry-app Implementation Requirements

edgesentry-rs additions needed

  • AuditRecord serialisation to VDES ASM frame format (fragmentation + reassembly)
  • IngestService VDES transport adapter (alongside existing MQTT/HTTP)
  • Shore-side VDES gateway process (receives ASM frames → REST → ingest)

edgesentry-app additions needed

  • Watchlist tab: load candidate_watchlist.parquet, display map + ranked list
  • AIS live overlay: WebSocket AIS client; overlay target track on chart
  • Evidence capture wizard: guided flow — photo → OCR → LiDAR (if available) → sign → transmit
  • VDES transmit queue: queue signed bundles, transmit when channel available, retry on failure
  • Tier selection: config flag for Tier 1 / 2 / 3 sensor stack
  • Hull silhouette reference format (PLY from vessel registry CAD)
  • scan_delta (CP3, already in trilink-core roadmap) for T0/T1 change detection between inspections