Fire is among the major threats to the safety of human life, homes and the property in both industrial and domestic areas. In order to resist fire threats efficiently, depends mainly on how early the detection of fire is done. Fire produces smoke, light and heat that help in identifying the fire. Fire can be disastrous and dangerous causing loss of lives and properties. To avoid this disastrous and dangerous fire situation, there is necessity of detecting fire at the early stages and act upon immediately to handle the situation. This paper works for the development of fire detection and alert system for early recognition and surveillance of fire with the help of CNN and machine vision techniques. The system comprises of surveillance cameras, to record the video for running the fire detection algorithm based on image processing to detect real time fire. This system is standalone and capable of transmitting recorded videos for fire detection anywhere in the world. Therefore, here we use CNN and computer vision technology to alert the user and fire station when fire is detected at the incident site. With the alert notification, a video of incident and the location of site are transmitted to user and fire station. As the area is constantly under surveillance, this video will help to perceive, how many people are trapped inside the affected area, so that the fire station can dispatch the team of competent rescuers on the basis of the video to rescue the people. The superiority of this model is that, we prevent using smoke and flame based sensors that might generate false alarms. In case, if this system triggers a false alarm then it can be verified by examining the recorded video. Hence the proposed system is used for synchronized detection of fire to properly detect fire incidents and send an alert, along with short video of fire to the remote fire alarm control unit. We correlated the execution of our method with that of recently reported fire detection approaches, using widely executed performance matrices for testing the achieved results through fire classification. This proposed system is successfully capable of detecting and notifying about the disastrous fire incidence at high speed and accurately. Future advancement in this field will include fire detection automatically and fire alarm generation. This system can be extended for multiple and widely scattered transmission nodes.

Tensorflow, OpenCV, Fire-detection, Machine Learning, CNN, python, fire alarm