Building a flood-resilience UAV for Cebu, Philippines using open-source components. Autonomous aerial intelligence at a fraction of commercial cost.
Commercial benchmarks that define what our DIY build must match or exceed.
Wingtra — $30,000–$50,000
| Type | Fixed-wing VTOL |
| Endurance | 59 min |
| Coverage | 72 km²/mission |
| Payload | 600 g |
| Accuracy (GSD) | 1.5 cm/px |
| PPK / RTK | Yes — survey-grade |
| Target market | Surveying, mapping, precision agriculture |
YellowScan — $22,000–$30,000
| Type | Solid-state LiDAR sensor |
| Range | 240 m |
| Accuracy | ±2 cm |
| Weight | 630 g |
| Points/m² | 100–300 |
| Returns | Up to 4 returns (first, last, inter) |
| Best for | Canopy-penetrating terrain surveys, flood modelling |
Full-frame 42MP compact — $3,000
| Sensor | 42.4 MP full-frame CMOS |
| Focal length | 35mm f/2 |
| ISO range | ISO 50–102400 |
| Video | Full HD 60fps |
| Weight | 507 g (body only) |
| GSD target | 2 cm/px @ 120m AGL |
| Hazard | No stabilisation; requires gimbal |
Open-source architecture selected for each subsystem. All components verified against Trinity Pro equivalent performance targets.
| Primary flight controller | Pixhawk 2.4.8 |
| Firmware | ArduPilot Copter 4.x (stable) |
| IMU redundancy | Dual IMU with temperature compensation |
| GPS / RTK | HereFlow RTK GPS + u-blox ZED-F9P |
| Ground station | Mission Planner / QGroundControl |
| Telemetry | SiK 915 MHz radio, 2 km range |
| Range | 190 m @ 80% reflectivity |
| Points/sec | 240,000 |
| FOV | 70.4° × 70.4° |
| Weight (body) | 480 g |
| Power draw | 12 W typical |
| Resistance | IP66 — rain/flood operations |
| Cost | ~$2,800 (vs. Qube 240 $22k+) |
NDVI and NDWI analysis for flood-plain moisture mapping. Bands selected for Cebu's vegetation index calibration cycle (Mar–Oct).
| Autopilot | ArduPilot Copter 4.x |
| Ground Control | Mission Planner / QGC |
| Survey Planning | Automated grid (ArduPilot spline) |
| Geotagging | ExifTool + GPS sync log |
| Imagery / SfM | OpenDroneMap (ODM), WebODM |
| Point clouds | CloudCompare, PDAL |
| Flood AI | Python + TensorFlow Lite (edge) |
| GIS output | QGIS, GeoJSON, PostGIS |
All-in Phase 3 build cost. DIY saves 80–90% versus commercial equivalent systems.
| # | Component | Category | Phase | Unit Cost (USD) | Qty | Subtotal | Source |
|---|---|---|---|---|---|---|---|
| 1 | DJI M600 Pro frame (refurbished) | Airframe | P1 | $1,850 | 1 | $1,850 | DJI refurb / China |
| 2 | Pixhawk 2.4.8 + IMU | Autopilot | P1 | $180 | 1 | $180 | Holybro / China |
| 3 | u-blox ZED-F9P RTK GPS | Autopilot | P1 | $240 | 1 | $240 | CRS Belgium |
| 4 | SiK 915 MHz telemetry radio | Autopilot | P1 | $55 | 2 | $110 | HK / China |
| 5 | Sony α6000 + Sigma 16mm f/1.4 | Camera | P1 | $900 | 1 | $900 | Japan / PH |
| 6 | Gremsy T3 gimbal | Camera | P1 | $480 | 1 | $480 | Gremsy / Taiwan |
| 7 | 6S 22,000mAh LiPo (2× + charger) | Power | P1 | $320 | 2 | $640 | CN / PH hobby |
| 8 | 40A ESC × 6 | Power | P1 | $45 | 6 | $270 | China |
| 9 | Raspberry Pi 4 (4GB) + case | Computing | P1 | $80 | 1 | $80 | RS / PH |
| 10 | ODM workstation (mini-ITX build) | Computing | P1 | $600 | 1 | $600 | DIY |
| 11 | Mission Planner laptop (used ThinkPad) | Computing | P1 | $250 | 1 | $250 | PH used market |
| 12 | Livox Avia LiDAR | LiDAR | P2 | $2,800 | 1 | $2,800 | Livox direct / HK |
| 13 | Livox Avia protective housing | LiDAR | P2 | $180 | 1 | $180 | DIY / 3D printed |
| 14 | Jetson Nano (4GB) + carrier board | Computing | P2 | $350 | 1 | $350 | NVIDIA dev kit |
| 15 | MicaSense RedEdge-M | Multispectral | P3 | $3,200 | 1 | $3,200 | MicaSense / US |
| 16 | Sunshine sensor + calibration panel | Multispectral | P3 | $480 | 1 | $480 | MicaSense |
| 17 | Spare parts & contingency (15%) | Other | All | $1,137 | — | $1,137 | — |
| TOTAL BUILD COST (Phase 1 + 2 + 3 + contingency) | $13,747 | ||||||
| Phase 1 MVP only | $5,600 | ||||||
Incremental capability building. Each phase produces deployable outputs, not dead-end prototypes.
Seven open problems that prevent this system from achieving commercial-grade performance. Each is a publishable research contribution.
No labelled UAV dataset exists for Cebu's flood topography. Training data must be generated from scratch, validated against PH-EWARS benchmarks, and open-sourced to support repeatability.
No published data on multirotor flight envelope during Cebu's typhoon seasons (Nov–Jan, 140+ km/h gusts). Autonomy fails without a published gust-recovery protocol and airframe stress modelling for sustained exposure.
No established methodology for fusing Livox Avia point clouds (phase-shifted, range-dependent density) with MicaSense 5-band imagery for flood-severity scoring. Requires novel co-registration and feature fusion pipeline.
Jetson Nano inference throughput for real-time flood segmentation (U-Net variant) is unverified on tropical urban imagery. Quantisation and model compression required; current literature is temperate-climate biased.
No validated community flood-risk scoring framework exists for Cebu's barangay structure. Metrics must be co-designed with local government units (LGUs) — model cannot be top-down imposed if it is to be trusted and sustained.
Philippine CAAP Civil Aviation Regulation Part 107 does not yet provide clear BVLOS pathways for community UAV operations. Research into the regulatory gap, international ICAO comparators, and a proposed framework for PH is needed.
No published cost-per-km² analysis comparing DIY UAV flood monitoring vs. satellite imagery (Sentinel-2) vs. manual survey for Cebu's urban-rural mix. Essential for securing LGU and donor funding for Phase 2–3 expansion.
End-to-end data flow from aerial capture to actionable community alerts. Built with open-source tooling throughout.
Recommended sourcing by region. All prices indicative and subject to market movement.