NaeonAIr - AI Eye for the Sky

NaeonAIr is a scalable smart-city safety platform that turns drones into intelligent airborne sensors, powered by edge AI, computer vision, and an IoT data backbone.

🌍 The Problem

Crowd disasters rarely happen without warning.

Before a crush incident occurs, there are measurable warning signs:

1. Rising crowd density
2. Conflicting movement flows
3. Increasing physical pressure between groups

Unfortunately, ground-based cameras and human monitoring often fail to detect these patterns early enough.

We asked:

What if a network of AI-powered drones could monitor crowd conditions from above and warn authorities before danger escalates?

That idea became NaeonAIr — AI Eye in the Sky.

✨ What Is NaeonAIr?

NaeonAIr is an end-to-end AIoT ecosystem that combines:

🛸 Smart drone payloads

🧠 Edge AI processing (Jetson Nano gateway)

🤖 Crowd risk analysis server

🌐 Smart city IoT integration (oneM2M Mobius)

🖥 Real-time TypeScript monitoring dashboard

Together, they form a digital twin of crowd safety.

This project consists of three main parts:

Drone AI Payload

Jetson Nano Edge Gateway

Server & Dashboard System

Drone AI Payload (Per Drone)

Drone frame + motors + propellers → UAV platform to carry the payload

Flight Controller (F405 or compatible) → Flight stabilization and power distribution

ESC (with 5V BEC output) → Motor control and 5V power supply

3S LiPo Battery → Main power source

XT60 Connector → Battery connection

ESP32-CAM (OV2640) → Captures aerial images/video

Raspberry Pi Pico 2W → Payload microcontroller (GPS + servo control)

NEO-6M GPS Module → Provides location data

Servo Motor → Optional signaling/actuation experiments

Jumper Wires → Electrical connections

Mounting Brackets / 3D Printed Mounts → To secure payload to drone

Vibration Damping Pads → Reduces camera shake

Edge Gateway (Ground Station)

NVIDIA Jetson Nano → AI gateway and inference server

Jetson Nano Power Supply (5V 4A) → Stable power for Jetson

MicroSD Card (32GB+) → OS and software storage

WiFi Router / Access Point → Network communication with drones and dashboard

Ethernet Cable (optional) → Wired network connection

AI & Backend Server

Jetson Nano (same device) → Runs AI inference and risk analysis

Python 3.x → Backend programming language

YOLO Model Weights → Person detection from aerial images

FastAPI → REST API server for analysis

Mobius oneM2M Server → Smart city IoT data platform

Monitoring Dashboard

Laptop or PC → Runs the dashboard interface

Modern Web Browser → Displays dashboard

Node.js → Runs TypeScript/React frontend

TypeScript + React → Dashboard framework

3D Map Library (MapboxJS) → 3D city visualization

Optional but Recommended

External WiFi Antennas → Better long-range drone communication

Portable Power Station → Field deployment of Jetson gateway

Tripod or Ground Mount → Stable placement for gateway equipment

This setup allows you to build a scalable AI drone monitoring system capable of analyzing real-time crowd safety and supporting up to 100 devices through the Jetson Nano gateway.

Each layer has a specific role in turning raw aerial footage into actionable safety intelligence.

Each drone carries a lightweight AIoT payload module that transforms it into a flying sensor node.

🔧 Hardware Components

ESP32-CAM (OV2640) → Captures aerial images/video

Raspberry Pi Pico 2W → Controls GPS, servo, and communication

NEO-6M GPS → Provides geolocation tagging

F405 Flight Controller → Integrates with UAV power & telemetry

Servo Motor → Enables physical signaling experiments

LiPo Battery + ESC → Power system

The drone does not perform heavy AI processing — it focuses on data capture and transmission.

📡 What Each Drone Sends

Each drone transmits:

1. Captured images
2. GPS coordinates
3. Device ID

These are sent to the ground gateway for centralized processing.

At the center of NaeonAIr is an NVIDIA Jetson Nano, which acts as the edge AI gateway for the entire drone network. Instead of running heavy AI models on every drone, we use the Jetson as a centralized processing hub. This makes the system lighter, more scalable, and easier to manage.

🚦 What the Jetson Nano Does

The Jetson works like a control tower for up to 100 drones.

Device Hub → Handles communication with multiple drones at the same time

Stream Router → Sends live video feeds to the monitoring dashboard

AI Processing Node → Runs the crowd detection and risk analysis services

IoT Bridge → Uploads results to the Mobius smart-city platform

Local Decision Layer → Can continue working even if internet is unstable

This design allows the drones to stay lightweight while the Jetson handles the heavy AI computation.

🔄 Hybrid Data Flow Architecture

To keep the system both fast and scalable, we use two types of communication:

⚡ Real-Time Monitoring (Low Latency)

1. For live viewing, the gateway communicates directly with the dashboard:

This provides instant video and status updates with minimal delay.

🌐 AI & IoT Data Pipeline

1. For analysis and long-term storage:

This separates real-time streaming from data storage, keeping both efficient.

🧠 Smart Media Handling (Why We Don’t Upload Images to IoT)

Images are large, and IoT platforms are not designed for heavy media storage.

So instead of uploading images directly to Mobius:

1. The AI server stores images locally
2. Only the image URL + analysis metadata are sent to Mobius

This keeps the IoT system lightweight while still allowing operators to view images when needed.

By using the Jetson Nano as a gateway, the system becomes:

✔ Scalable to large events

✔ Capable of managing many drones

✔ Efficient in both real-time and cloud-connected modes

This is what transforms NaeonAIr from a single drone demo into a city-scale AI monitoring system.

Now that drone images reach the gateway, we need a system that can understand what’s happening in the crowd.

This is the role of the AI Crowd Risk Analysis Server.

It transforms raw aerial images into actionable safety intelligence.

🔍 Part 1: Detect People Using AI (YOLO)

The first step is identifying people in each drone image.

We use a fine-tuned YOLO (You Only Look Once) model trained for aerial/top-down views.

Pipeline:

Output from this step:

1. Total crowd count
2. Locations of people in the scene

This gives us the “where” and “how many”, but not yet the danger level.

🌊 Part 2: Analyze Crowd Flow & Pressure

Crowd disasters often happen not just because of density, but because:

Moving groups push into already dense areas, creating dangerous compression.

So we go beyond counting people.

We analyze motion between frames to estimate:

1. Direction of movement of crowd clusters
2. Differences in speed between neighboring groups
3. Areas where movement slows but incoming flow continues

These interactions produce a crowd pressure metric, which is strongly linked to crush risk.

This is what separates NaeonAIr from simple people-counting systems.

⚠️ Part 3: Calculate the Risk Score

Finally, we combine multiple indicators into one easy-to-understand safety signal.

The system considers:

1. 👥 Total crowd count
2. 📈 Density trends over time
3. 🌊 Movement flow patterns
4. 🧱 Pressure indicators

These are fused into:

Instead of just showing numbers, the system provides an early warning level that can help responders act before a situation becomes critical.

Once the AI server calculates crowd risk, the data shouldn’t just stay on one computer.

To make the system useful at a city scale, we connect it to Mobius, a oneM2M-based IoT platform.

Mobius acts as a central data hub where crowd analytics, drone status, and alerts can be stored and accessed by other systems.

📦 How Data Is Organized in Mobius

Inside Mobius, we create an application entity for the drone system:

Each part has a role:

1. crowd_analysis → Stores AI results (crowd count, density level, risk score)
2. drone_status → Stores drone telemetry (GPS, altitude, battery, device ID)
3. alerts → Stores warnings when risk levels become high

This structure keeps data organized and easy to access for dashboards or other city systems.

🌆 Why Smart City Integration Matters

By sending data to Mobius, the system becomes more than a local experiment.

It enables:

1. 📊 Historical analysis of crowd conditions over time
2. 🔗 Integration with other smart city services (traffic, emergency systems, etc.)
3. 🖥 City-wide monitoring dashboards that combine data from many sources

Instead of one drone feeding one screen, NaeonAIr becomes part of a connected urban safety network.

AI results are only useful if people can understand them quickly.

That’s why NaeonAIr includes a web-based monitoring dashboard built with TypeScript.

This dashboard acts as the live control and visualization center for the entire drone network.

It connects to:

1. The Jetson Nano gateway for live streams
2. The Mobius IoT platform for AI analysis results

🎛 Key Dashboard Features

The primary components include:

📦 Device Management

Operators can manage the entire drone fleet from one screen:

1. Register multiple drone devices
2. View which drones are active or offline
3. Monitor incoming data per device

This makes the system scalable beyond a single drone.

🧠 AI Mode Selection

Different situations require different views. The dashboard allows operators to switch between analysis modes:

Density → Visualizes how crowded an area is

Danger → Displays the AI-estimated crowd risk level

This helps users focus on either crowd size or potential danger buildup.

📷 Live Camera Streams

Operators can view real-time video feeds from drone cameras.

This provides:

1. Visual confirmation of AI results
2. Context for alerts
3. Better situational awareness during events

🌍 3D Map-Based Risk Visualization

Instead of showing only numbers, the system overlays risk data on a 3D city map.

This allows operators to:

1. Understand where risk is increasing
2. Track crowd movement patterns across areas
3. Make faster, location-based decisions in emergencies

It’s not just data — it’s spatial risk awareness.

Because the system is gateway-based, it can support:

1. Up to 100 drone devices
2. CCTV cameras
3. Fixed smart sensors
4. Mobile robots

All follow the same pipeline:

NaeonAIr is a research and safety-support system.

1. Must comply with UAV flight regulations
2. Must respect privacy laws
3. Designed for crowd safety monitoring, not surveillance misuse
4. Supports human decision-makers, not autonomous enforcement

NaeonAIr transforms drones from flying cameras into:

AI-powered, city-scale crowd safety sensors

By combining edge AI, computer vision, IoT integration, and real-time visualization, the system provides a digital twin of crowd risk — enabling faster, smarter, and safer event management.

To contribute and explore more, please visit our repositories:

1. https://github.com/SUNSET-Sejong-University/NaeonAir (For hardware code and gateway)
2. https://github.com/SUNSET-Sejong-University/wisdrone-interface (For interface and system integration)
3. https://github.com/SUNSET-Sejong-University/wisdrone-ai-drone (For AI server and algorithm)

NaeonAIr — AI Eye in the Sky was developed as a collaborative research and engineering project by:

👨‍💻 Choi Hyong Chan

AI Systems • Dashboard Architecture

1. Real-time monitoring interface (TypeScript dashboard)
2. System integration and deployment
3. Mobius (oneM2M) IoT platform integration

👨‍💻 Kang Naeon

AI & Data Intelligence • Smart City Integration

1. Crowd risk analysis algorithms
2. AI inference pipeline

👨‍💻 Das Prithwis

Embedded Systems • Drone Hardware • Communication

1. Drone payload electronics
2. ESP32-CAM imaging system
3. Wireless communication pipeline
4. Jetson Nano gateway design