AeroCoast Sentinel

• 

  Synthetic Scenario 1: Real-time INT8 bounding-box segmentation detecting washed-up marine plastic litter and ghost nets.

• 

  Synthetic Scenario 2: Severe dune scarp slope degradation alert triggered with geomorphic structural failure warnings.

• 

  Synthetic Scenario 3: Nominal baseline tracking for protected avian wildlife (Piping Plovers) across clean seagrass beds.

• 

  LiFePO4 4S power bus telemetry showing 96.4% SoC, 4.2W inference budget, and NOAA Station #8775870 tidal level baselines.

• 

  Industrial 24-hr analytics deck: Tidal surge cycles, 24-hr multi-class detections, and 98.32% Green AI carbon offset curves.

• 

  Complete AeroCoast Sentinel mission console: Edge-native SIMD vision, dual sensor feeds, and RFC 7946 GeoJSON export.

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for AeroCoast Sentinel## 🌊 Inspiration

Remote barrier islands along the South Texas shoreline (Mustang Island, Padre Island, and Corpus Christi Bay) represent fragile ecosystems vulnerable to marine plastic washup, storm-driven dune scarp erosion, and hypersaline degradation. Traditional coastal monitoring systems rely on continuous video streaming dispatched to power-hungry cloud server GPUs (250W–400W). In remote barrier islands, this paradigm fails completely: cellular coverage is unstable, power grids do not exist, and continuous cloud computation creates an unjustifiable carbon footprint.

We engineered AeroCoast Sentinel as an autonomous, zero-cloud edge defense and environmental conservation console that executes INT8 computer vision directly on edge hardware with 100% off-grid solar autonomy.

⚡ What it does

AeroCoast Sentinel serves as a field-deployable coastal observatory console:

• Sub-7ms INT8 Edge Vision: Executes bounding box segmentation on marine plastic debris, ghost nets, dune erosion scarping, and protected avian wildlife (Piping Plovers) at 60 FPS using client-side WebAssembly SIMD.
• NOAA & TCEQ Physical Calibration: Integrates real-time tidal baseline datum from NOAA Station #8775870 (Corpus Christi Bay) and hypersaline water metrics from TCEQ Mustang Island baselines.
• 4.2W LiFePO4 Energy Bus: Powered by an ultra-low-power solar budget with live MPPT telemetry, consuming 98.32% less energy than standard cloud GPU pipelines.
• Dual Optical Stream: Offers instant zero-latency switching between synthetic tactical scenarios and real local WebRTC edge webcams.
• RFC 7946 GeoJSON Audit Trail: Logs tamper-evident, geotagged hazard detections locally during network blackouts for batched LoRaWAN transmission.

🛠️ How we built it

• Edge Vision Engine: TensorFlow Lite Micro quantized INT8 inference accelerated with WebAssembly SIMD and WebGL fallback.
• Firmware & Embedded Target: ESP-IDF / PlatformIO C++ firmware configured for the Freenove ESP32-S3 WROOM CAM with Omnivision OV2640 sensor.
• Green AI Verification: Benchmarked using CodeCarbon standards against regional ERCOT grid emission factors.
• Telemetry UI: High-performance dark HUD engineered with vanilla ES6+ JavaScript, Canvas 2D, and CSS3 without bulky runtime frameworks.
• Deployment: Live production hosting running globally on Vercel Edge.

🧗 Challenges we ran into

• Thermal & Power Budgets: Confining multi-class computer vision within a strict 4.2W power envelope required aggressive INT8 quantization and memory alignment.
• Zero-Cloud Resilience: Designing deterministic offline buffering that safely archives critical ecological telemetry without depending on cellular networks during Gulf hurricanes.

🏆 Accomplishments that we're proud of

• Achieving deterministic 6.5ms inference latency in-browser and on microcontrollers.
• Validating 98.32% energy reduction (saving 2,153 kWh and 876 kg CO₂e annually compared to 250W cloud GPU

Built With

• c++
• codecarbon
• edge
• esp32-s3
• green-ai
• javascript
• noaa-api
• python
• tceq
• tensorflow-lite-micro
• webassembly