Vision Based Fire Flame Detection System Using Optical flow Features and Artificial Neural Network

: Detecting the break-out of a fire rapidly is vital for prevention of material damage and human casualties. VFD technology is becoming the focal point of research with its advantages of high intuitive, speed and anti-jamming capability. But most of current methods for video fire detection have high rates of false alarms. Here a novel video fire flame detection method based on 3 flame colour model, optical flow features and neural network is presented. . The method is proposed as followed, first, candidate fire regions are determined by 3 flame colour model they are RGB, HSV and YCbCr instead of single colour space RGB. Even though RGB colour space can be used for pixel classification, it has disadvantages of illumination dependence. Then,. a pyramidal implementation of Lucas Kanade feature tracker is used to analyzing dynamic optical flow features such as average and variation of optical flow velocity and optical flow orientation in the candidate region. Then a back-propagation neural network is used to learn and classify the statistic of the flames from nuisances based on the optical flow features. This method can distinguish flame videos from disturbances which having the same colour distribution as flame, such as car lights, and have a remarkable accuracy. . Also, In this method by calculating ratio of the area of the fire pixels in the video, and whole pixel area it give alert message danger when this value increase from a threshold during testing the video, also by using this value we can understand the rate growth of fire