what is it?
A project made for Hacklytics 2024, Flame AI is an advanced fire detection system alerting first responders & admins quickly, providing key data for optimal resource allocation & swift rescues. Flame AI revolutionizes fire safety with its advanced detection technology, offering a wide range of benefits:
- Live Fire Detection: Real-time analysis of camera/video streams to detect fires
- Population Monitoring: Detects and reports the number of individuals in affected areas
- Multi-Camera Support: Monitors multiple camera views within a building for comprehensive coverage
- Instant Notifications: Sends alerts via Gmail and phone to designated administrators and first responders
- Intelligent Forecasting: Identifies regions at risk of fire using climate data
- Interactive Image Map: Offers real-time environmental metadata through an intuitive interface
cool things/ what i learned
The machine learning in this project utilizes a hybrid approach, pairing a yolov5 computer vision model to detect people and a logistic regression model trained on fire sensory data to reduce the number of false positives
We fine-tuned an existing yolov5 model for image classification on a dataset of fire images, tailored to our purpose
We used openCV to capture and preprocess the video stream data from our laptop, converting each frame to a format sutiable for analysis
To combat the idea of false positives, we used a logistic regression model trained on atmospheric data with features in most commercial sensors to learn more about the smoke
Used SMTP to send direct alerts immediately
