Pablo Chacón

Elies Aragonès

[HackUPC2026] - HeatShield

Modulino™ Thermo

Arduino App Lab

Arduino® UNO™ Q 4GB

<html><body>HeatShield is an EdgeAI project designed to improve safety in climate shelters during extreme heatwaves. The urban problem it addresses is that public cooling spaces can quickly become difficult to manage when staff must simultaneously watch for medical emergencies, rising indoor heat stress, and overcrowding. In these conditions, vulnerable people may collapse without immediate notice, and admitting more people into an already stressed shelter can increase risk for everyone inside. HeatShield helps solve this by using local sensing and on-device intelligence to detect falls, trigger alerts if someone falls, monitor environmental conditions, track occupancy, and indicate whether it is still safe to admit more people. The hardware lineup includes an Arduino UNO Q as the main edge device, a USB webcam for on-device fall detection, a Modulino Thermo for temperature and humidity sensing, and the supporting USB hub, power supply, and USB-C cabling required to run the system. Together, these components allow the project to combine vision-based monitoring with real-time environmental data in a compact local setup. The AI model is an Edge Impulse object-detection model deployed locally as models/model.eim and used through the Arduino App Lab video_object_detection brick. The vision model was fine-tuned on the FallDataset IMVIA dataset and is responsible for identifying fall events from the webcam feed. To reduce false positives, the system does not react to a single instant detection alone; instead, it confirms a fall only when the detection is sustained for several seconds before triggering an emergency alert. The software architecture is fully local and built around Arduino App Lab. The app.yaml file defines the main application bricks, including video object detection and the web UI. The Arduino sketch reads temperature and humidity from the Modulino Thermo and sends those values into the Python runtime through Bridge.notify. In python/main.py, the application combines sensor readings, occupancy estimation from detections, and fall-confirmation logic to maintain the current shelter state. Additional modules handle sustained fall triggering and admission assessment, while the web dashboard in assets/ displays the live camera feed, environmental readings, occupancy, emergency state, and a simple green/yellow/red admission status for shelter operators. </body></html>

[HackUPC2026] - HeatShield

In climate shelters, our system detects heat-related falls and auto-calls for help, monitoring comfort and occupancy to ensure fast, life-saving response.