Dynamic Fire Detection and Alerting Using Image Processing for Drone Applications

In this paper, a technique has been proposed for a dynamic fire detection system based on Python opencv image processing. On the one hand, consider the system consists of a quadcopter/drone equipped with a mobile phone having the IP webcam app for live video streaming, and the drone is controlled autonomously using a PC, dedicated software like mission planner. On the other hand, another PC with a Python algorithm is used for the processing of the video image to detect fire. The dynamic approach consists in capturing the live stream webcam and detecting any fire presence in the video frame with the HAAR cascade classifier dataset and all this in real-time. After takeoff of the drone, the area mapping is done through waypoints on the Mission planner. A second PC (fire control) is used to start the live stream, detect fire and make decisions. The webcam IP address is fed into the algorithm to extract the video frame. The frame is in RGB format and with the help of the pre-trained model along with the threshold parameters used; the algorithm can detect fire in the live streaming coming from the drone’s camera. Once the fire is detected, the frame becomes stable for some time so the operator sitting in the control room can keep track of the event happening. Functionality is added to the code to track some specific pixels (red, blue, and white) in the fire-detected region and display approximate temperature values based on preset thresholds of these colors. Since it is very difficult to measure the temperature on the image without using a thermal camera and this logic is a novel one. Also, logic is used to capture a fire incident into a frame .jpg image and send it with an email as an attachment to the concerned fire department after a 6s delay. The attachment image is a very important and unique feature as it reveals to the fire department the type of incidents that occurred and even the area if the camera is good enough.