Optical Flow Estimation with Gradient Detection for Identifying Flames
Detecting fire break out is absolutely necessary in order to prevent loss of life and property. The vision – based fire detection techniques with surveillance systems have become popular in the past decade. Video camera covers wide viewing range and from the data captured by video camera additional information can be extracted. Video sequences provide an insightful view about how the object and scenes in the video change over time. For indoor fire detection, traditional point sensors were used to detect heat or smoke particles. However, when it comes to open space it is not viable to employ these point sensors. Optical flow estimators transform the image sequence into estimated motion. Optical flow vector is created in the system and is used to depict the magnitude and direction of motion of an object as it moves from one frame to another. Based on the motion estimators, a set of motion features is presented that exploits the difference between dynamic fire motion and rigid fire motion. Two optical flow methods are designed for flame flow vector creation, namely, Non-Smooth Data (NSD) and Optimal Mass Transport (OMT). NSD and OMT methods are used for modelling flame with dynamic texture and saturated fire blob respectively. To further enhance the process of flame detection, gradient optical flow estimation methods and classification based on feature vectors have to be done.