This project showcases a full proof-of-concept for an intelligent, drone compatible fire and smoke detection system powered by a Raspberry Pi and a custom trained YOLO (You Only Look Once) model. The system captures live video feed and uses the YOLO object detection algorithm to identify visible signs of fire and smoke in real time, eliminating the need for constant cloud connectivity.
Built entirely in Python, the pipeline combines computer vision, geolocation, and telemetry. After calculating the fire's angle and distance based on the bounding box and camera field of view, the system uses trigonometry and geospatial conversion formulas to compute the fire's actual coordinates. These coordinate, along with confidence scores and timestamps, are then sent through multiple alerting channels.
The system supports webhook-based API notifications, SMS alerts using Twilio, and detailed email messages with GPS tagged information. It also logs every detection locally and uploads the records to a remote server using SSH/SFTP. A network check ensures network reliability by switching between Wi-Fi and cellular connections when necessary.
This proof of concept demonstrates the viability of edge-based wildfire detection using AI, GPS, and real-time communications, all running on a Raspberry Pi. Future development will include hardware-integrated LiDAR support, drone based field-testing, and visualization dashboards for environmental monitoring and rapid emergency response.