ENGLISH

Authors: Natàlia Cortés, Joan Marc Samó, Carolina Espin, Josep Mª Cases, Carlos Albaladejo

SafeHelmet - Collision Detection System

Project Name: SafeHelmet

🎯 Project Objectives

Develop a collision detection and alert system for static and dynamic objects, integrated into certified protective helmets as an accessory tool for road circulation
Detect moving objects simulating the reality of the road network
Address the risk posed by excessive speed and proximity of objects

🏗️ Web Architecture

The web interface has been primarily structured with:

Pages and Routes (SEO Optimized)

/ - Home page with a product introduction
/map - Visualization of locations, incidents and other data
/camera - Live monitoring with active detection
/about - Information about the project and team

Main Components

Enhanced Header

3 columns: Map Button | Logo | About Us + Menu
Intuitive navigation
Sticky header on scroll

Sidebar

Contextual menu (~25% width)
Only visible at /map and /camera
Smooth transitions
Overlay to close

Modern Design

Color gradients (Orange #ff3e00 → Yellow #ffaa00)
Transitions and hover effects
Responsive design (Desktop, Tablet, Mobile)

💻 Technologies Used

Frontend

Svelte 5.55.2 - Reactive framework
SvelteKit 2.57.0 - Full-stack framework with routing
TypeScript 6.0.2 - Static typing
Vite 8.0.7 - Fast build tool

Backend (Arduino)

Python - Main script
Arduino IDE - Hardware sketch
Collision detection via sensors

🚨 Alert Systems

Visual Alert

Two hertz flashing
Non-saturated red LED
Located on the handlebar

Sound Alert and Vibration

System integrated into the helmet
Immediate proximity notification

Intelligent Detection

Distance identification with OpenCV (implementable)
Object speed analysis
Real-time system

Web Notifications

Browser popup alerts shown on the camera page
Triggered when a person is detected near the critical zone
Environmental alerts for high noise levels or poor air quality
Alerts are rate-limited to avoid repeated notifications
Messages are short and actionable, for example:
- Person nearby: getting close to the critical zone.
- Noise is high. Consider moving to a quieter area.
- Air quality is poor.

📁 Project Structure

Interhack-2026-Qualcom/
├── src/
│   ├── routes/                    # SvelteKit routes
│   │   ├── +layout.svelte        # Main layout
│   │   ├── +page.svelte          # Landing
│   │   ├── map/+page.svelte      # Map
│   │   ├── about/+page.svelte    # Information
│   │   └── camera/+page.svelte   # Live monitoring
│   ├── lib/
│   │   ├── components/
│   │   │   └── molecules/templates/
│   │   │       ├── Header.svelte       # Enhanced header
│   │   │       ├── Sidebar.svelte      # Side menu
│   │   │       ├── About.svelte        # About page
│   │   │       └── MapLayout.svelte    # Map layout
│   │   └── styles/
│   │       └── global.css              # Global styles
│   └── static/icons/             # SVG icons
├── Arduino code/
│   ├── sketch/
│   │   └── sketch.ino            # Arduino code
│   └── python/
│       └── main.py               # Python script
├── package.json
└── svelte.config.js

🚀 How to Run

Development

npm install      # Install dependencies
npm run dev      # Start server (http://localhost:5173)

Production Build

npm run build    # Create optimized build
npm run preview  # Preview the build

Validation

npm run check          # Svelte check
npm run check:watch    # Check in watch mode

Arduino App Lab Structure

This project uses Arduino App Lab with:

sketch/ → Arduino .ino firmware running on the MCU
python/ → Python application running on Linux
Bridge communication between Arduino and Python

Arduino Project structure in Arduino App Lab:

project/
├── sketch/
│   ├── sketch.ino
│   ├── DistanceSensor.ino
│   ├── BuzzerModule.ino
│   ├── PixelsModule.ino
│   └── ThermoModule.ino
│
├── python/
│   └── main.py
│
└── app.yml

🎨 Color Palette

Color	Use	Value
Orange	Primary (Buttons, Links)	#ff3e00
Yellow	Secondary (Gradients)	#ffaa00
Light Gray	Background	#f4f4f4
Dark Gray	Text	#272727

🔄 Navigation Flow

Home (/)
    ↓
[Explore Map] [Learn More]
    ↓                ↓
  /map            /about
  ├─ Header       ├─ Header
  ├─ Menu ✓       ├─ Menu ✗
  └─ MapLayout    └─ Information

From /map:
  ↓ (Menu Button)
[Sidebar] - 25% width
  - Settings
  - Documentation
  - Support
  - Contact

✨ Notable Features

✅ SEO Optimized - Meta tags and titles on each page ✅ Responsive - Works on all devices ✅ Smooth Transitions - Fade animations between pages ✅ Conditional Menu - Only visible where appropriate ✅ Sticky Header - Adapts to scrolling ✅ Interactivity - Hover effects and animations

📝 Upcoming Improvements

[ ] Integrate real map (Leaflet/Mapbox)
[ ] Websocket for live data
[ ] Statistics dashboard
[ ] Authentication system
[ ] Product gallery
[ ] 3D helmet animation

🤝 Contributors

Interhack 2026 Team
Qualcomm Innovation Challenge

Last updated: May 10, 2026

Evolution:

The next stages of the product will focus mainly on scalability and the incorporation of infrared light systems to improve illumination or even allow nighttime circulation.

SOFTWARE

AI Training

For real-time collision detection with low latency, we have implemented a TinyML model using the Edge Impulse platform. This allows us to perform Edge Computing. Image processing is done locally on the device without needing to rely on the cloud, leveraging a local model.

Training Process:

Data Acquisition: Creation of a custom dataset with more than 100 captures of rubber ducks, taken with the Logitech Brio 105 Business Webcam at different angles. Considering factors such as distances and lighting conditions to simulate varied obstacles.
Impulse Design: Image pre-processing to optimize memory consumption.
Feature extraction through image processing blocks.
Detection Model: Training of a neural network optimized for mobile devices (Object Detection).
Optimization: Conversion of the model to a lightweight format (TensorFlow Lite equivalent) to run on a high-performance, low-power chip.
Local Implementation (On-Device AI): Unlike other systems that send video to a server, our AI runs directly on a dedicated microchip integrated into the helmet.

Model characteristics

Minimum latency: The alert is instantaneous (critical to avoid collisions).
Privacy: No images of public roads are stored or sent; the system only generates alerts and anonymous numerical data.
Model Result. Since our product requires measuring distances to objects, we will use the ratio of the object's size relative to the full image as a reference. Since this ratio must vary in proportion to the distance, the training data available to us (oriented to work with centroids) is inefficient and therefore the model has not given us the desired results.

As a consequence, we used a pre-trained model provided by Qualcomm Innovation Challenge to carry out our demo.

Hardware

This prototype incorporates multiple hardware sensors that allow us to test the real latency of the different inputs of our design.

Main Components

Module	Units	Description	Reference
[Arduino Modulino Vibro]	1	Sensor to alert the user with vibration	(https://docs.arduino.cc/hardware/modulino-vibro/)
[Arduino Modulino Pixels]	1	Customizable display	(https://docs.arduino.cc/hardware/modulino-pixels/)
[Arduino Modulino Thermo]	1	Ambient temperature sensor	(https://docs.arduino.cc/hardware/modulino-thermo/)
[Arduino Modulino Distance]	1	Sensor that detects the distance of a hazard	(https://docs.arduino.cc/hardware/modulino-distance/)
[Arduino Modulino Buzzer]	1	Sound alert for the user	(https://docs.arduino.cc/hardware/modulino-buzzer/)
[Arduino Modulino Movement]	1	Accelerometer for decision making	(https://docs.arduino.cc/hardware/modulino-movement/)

FUTURE PERSPECTIVE

To transform this prototype into a market product, our roadmap includes:

High Performance AI and 360° Vision:
Multi-object Recognition: Training the model to identify vehicles, pedestrians, traffic signs and animals with a near-zero error rate.
Night Vision: Implementation of infrared sensors to maintain safety during night driving or in tunnels.
Advanced Environmental Monitoring
Air Quality Sensors (NOx, PM2.5): Integration of specific sensors to measure suspended particles and nitrogen dioxide. This would allow the City Council to have a street-by-street pollution map in real time.
Asphalt Analysis: Using the accelerometer to automatically detect potholes or irregularities in the road surface and report them directly to urban maintenance services.
Connectivity and Ecosystem
V2X (Vehicle-to-Everything): Direct connection with traffic lights and other vehicles to anticipate collisions before the object is visible to the camera.
User App: A mobile device-compatible interface where the cyclist can check their safety statistics and receive rewards for the environmental data contributed.