Border Probe Mesh | Solar-Powered Early Warning System

Securing Europe's Eastern Border

Autonomous detection where traditional surveillance fails

The Challenge

Lithuania's extensive forested border presents unique surveillance challenges: dense vegetation blocks line-of-sight, wildlife triggers cause high false alarm rates, extreme seasonal temperatures strain equipment, and traditional infrastructure is costly to deploy and maintain in remote areas.

The Solution

A mesh of small, autonomous "border probes" deployed at strategic intervals along the border. Each probe combines solar power for year-round energy autonomy, thermal + visual AI detection for intelligent threat classification, compass-based directional awareness for precision triangulation, and LoRa mesh networking for reliable long-range communication without cellular infrastructure.

The two-stage AI pipeline filters 70-90% of wildlife false alarms before alerting operators, dramatically reducing operator fatigue while maintaining detection reliability.

$150–200 Per Unit at Scale

30–60m Probe Spacing

365 Days Energy Autonomy

<25m Triangulation Accuracy

Two-Stage Detection Pipeline

Power-efficient AI classification that sleeps until it matters

Silent Watch

Deep sleep mode
~150 µA / 0.5 mW

99%+ of time

Thermal Trigger

MLX90640 detects
heat signature

Wake event

Stage 1: Thermal Triage

Lightweight classifier
~50 ms / 0.001 Wh

Fast decision

70–90% Wildlife → Sleep

Stage 2: Visual AI

OV2640 + TFLite Micro
~1–2 min / 0.05–0.10 Wh

Deep analysis

Alert Broadcast

LoRa mesh alert +
detection ray + image

ALERT

0.001 Wh Thermal triage per event

0.05–0.10 Wh Full analysis per event

70–90% Triggers resolved at Stage 1

Hardware Components

Purpose-selected components for autonomous border monitoring

LILYGO T-Beam Supreme

$30–40

ESP32-S3 SX1262 LoRa L76K GNSS 8MB PSRAM

Main compute & communication board with integrated LoRa radio and GPS.

MLX90640 Thermal Sensor

$40–60

32×24 px 55° FOV 1.72°/px

Far-infrared thermal array for day/night human detection at 15–30 m range.

LSM303AGR

~$4

Magnetometer Accelerometer

Compass heading for detection ray orientation and tamper detection via accelerometer.

OV2640 Camera

$5–15

2 MP 96×96 inference Full-res save

Visual camera for Stage 2 AI classification, downsampled for edge inference.

10W Solar Panel

$20–35

Monocrystalline 70° mount Hydrophobic

Primary power source with optimized tilt angle for Lithuanian latitude.

LiFePO4 Battery Pack

$40–70

8× 18650 80–100 Wh LiFePO4

High-capacity battery for multi-day autonomy during winter low-solar periods.

LoRa Antenna

$5–15

868 MHz EU SMA

External antenna for extended LoRa mesh range through forest canopy.

IP67 Enclosure

$15–40

Weatherproof Tamper detect

Sealed ABS housing with cable glands, rated for -40°C to +60°C.

Industrial MicroSD

$10–15

Crash-resistant Industrial grade

Local evidence storage for captured images and event logs.

MPPT Charge Controller

$5–10

Solar MPPT LiFePO4

Maximum power point tracking for optimal solar energy harvesting.

Total BOM per probe unit

$190 – $348

At scale (100+ units): $150 – $200

Technical Specifications

Power budget, solar production, and sensor capabilities

Power Budget

Mode	Current	Energy
Sleep (99%+ time)	~150 µA	0.5 mW
Stage 1: Thermal triage	50 mA, 100 ms	0.001 Wh
Stage 2: Full analysis	200–300 mA, 1–2 min	0.05–0.10 Wh

Solar vs Consumption

Summer

20–28 Wh/day

~2.1

10–13x

Spring / Autumn

8–14 Wh/day

~1.0

8–14x

Winter

1.5–4 Wh/day

~0.3

5–13x

10W panel, Lithuania latitude, with forest canopy factor

Sensor Capabilities

Thermal resolution	32 × 24 pixels
Thermal FOV	55°
Angular accuracy	~0.86° (sub-pixel)
Detection range	15–30 m (human)
Camera resolution	2 MP (OV2640)
AI inference input	96 × 96 px
LoRa frequency	868 MHz (EU)
LoRa spreading	SF10
GPS accuracy	~2.5 m (averaged)
Compass heading	±2° (calibrated)

Base Station / Hub (per 10–20 probes)

T-Beam Supreme + high-gain antenna$50–70

Raspberry Pi 4 or mini PC$50–80

4G/LTE modem$30–50

Solar + battery$80–150

Enclosure$30–50

Total$250–400

LoRa Mesh Network Architecture

Resilient communication without cellular infrastructure

Direct-to-Base

Preferred mode. Probes send alerts straight to the nearest border post receiver for lowest latency.

Mesh Relay

When direct range is limited, alerts hop through neighboring probes using Meshtastic mesh protocol.

Health Monitoring

Heartbeat every 15–30 min reports battery, solar rate, GPS quality, heading, temperature, and event count.

Detection Ray Triangulation

Pinpointing intrusions with sub-degree angular precision

1

GPS + Compass Heading

Each probe knows its exact position and the direction it faces, establishing a geographic reference frame.

2

Thermal Centroid Bearing

The MLX90640 determines the angular position of a heat source within its 55° FOV to ~0.86° accuracy via sub-pixel centroid fitting.

3

Detection Ray

Combining GPS position + compass heading + thermal bearing produces a "detection ray" — a precise direction line in absolute geographic coordinates.

4

Multi-Probe Intersection

Two or more probes observing the same target produce intersecting rays, pinpointing the target's location to an ~80–100 m² zone. More probes = higher confidence.

Consortium Partners

Sub-Project 6 consortium within the national R&D programme

VDK

Vilniaus Dizaino Kolegija

Design for Safety & Security

MRU

Mykolo Riomerio Universitetas

Regulatory & Safety-Cybersecurity Compliance

MN

UAB “Masternode”

Sensor Hardware Technology Development

LX

UAB “Lexita”

Software Technology Development

ET

Egle Termine

Consortium Partner

Regulatory Compliance

Designed for full EU regulatory alignment

EU Fundamental Rights Agency

Non-intrusive surveillance approach. No facial recognition. Thermal-only classification preserves individual privacy while detecting threats.

AI Act

Transparent classification pipeline. All AI decisions logged with confidence scores. Human-in-the-loop for critical alerts.

NIS2

AES-256 encrypted LoRa communications. Secure boot on ESP32-S3. Tamper detection with accelerometer alerts.

MVP Targets & Deliverables

36-month project milestones and measurable outcomes

TRL 6–9 Technology Readiness Level

70–90% Wildlife False Alarm Filtering

<25m Triangulation Position Error