Multisensor intelligence for real time hazard detection
Team Members:
- Kavya Mohankumar
- Sai Pranavi Gogineni
Fire and flood damage cause billions of dollars in losses every year, yet most security systems only record incidents after they happen. Cameras miss events when guards aren’t watching, smoke detectors trigger too late, and false alarms cause people to ignore real emergencies.
SafeSense was built to answer one question:
What if cameras could recognize danger as it happens and confirm it before alerting?
Our goal was to create a real-time hazard detection system that prevents disasters instead of just documenting them — while eliminating false alarms that make existing systems unreliable.
SafeSense uses AI vision combined with physical sensors to detect fires, floods, and emergencies in real time.
- Detects flames using AI vision
- Confirms fires using temperature data
- Detects flooding through humidity spikes
- Detects people near danger zones
- Sends alerts only when multiple sensors agree
This sensor-fusion approach allows SafeSense to:
- Reject fake fires on TV screens
- Prevent alert fatigue
- Deliver verified alerts in under 2 seconds
Alerts are delivered via SMS, a web dashboard, and a local LCD display.
- Arduino Nano
- DHT22 Temperature & Humidity Sensor
- HC-SR04 Ultrasonic Distance Sensor
- Grove LCD Display
- USB Webcam
- Python
- Flask (Web Server)
- OpenCV (Video Processing)
- YOLOv8 (Computer Vision Model)
- PySerial (Arduino Communication)
- Twilio API (SMS Alerts)
- HTML / CSS / JavaScript (Dashboard)
Problem:
The system initially triggered alerts for YouTube videos, images of fire, and even orange clothing.
Solution:
We implemented a sensor fusion algorithm that requires temperature confirmation before classifying a fire as real.
Problem:
Arduino serial data was frequently corrupted or misread.
Solution:
We added structured serial formatting and validation on the Python side to ensure reliable sensor readings.
Problem:
Running AI detection and video streaming together caused lag.
Solution:
We switched to a faster YOLO model and used multithreading to separate video capture, inference, and sensor processing.
Problem:
We couldn’t safely test with real fire during the hackathon.
Solution:
We demonstrated SafeSense’s false-positive prevention, showing it correctly reject fake fires while still detecting real flooding conditions.
- Zero false positives during 6+ hours of testing
- Alert delivery in under 2 seconds
- Built for ~$30 using off-the-shelf hardware
- One system detects multiple hazard types
- YOLOv8 – Ultralytics open-source object detection model
- OpenCV – Open-source computer vision library
- Flask – Python web framework
- Twilio API – SMS alert delivery
- Arduino Framework – Microcontroller programming
- Add gas detection and sound analysis (glass breaking)
- Deploy on Raspberry Pi for standalone operation
- Build a mobile app for remote monitoring
- Pilot deployment in real buildings