Detection of flame based on computational vision has visible attention in the past years. Many selective features such as shape, colour, texture, etc., have been used to detect flames. The drawback to detect flames with these features are that they uses Lucas-Kanade method which is an optical flow method and it adopt flow constancy same for the adjacent pixels. This will reduce the reliability to detect flames from the videos. In this paper point-wise approach is used in which the conditions are applied to every pixel instead of continuous regions. And also this paper is not uses thermal heat fire detection method as used in classical approach. In this paper we have used binarization algorithm which provides accuracy of detecting fire flames as it very useful for detecting suspicious regions of flames in video files and also helps to eliminate background nosiness from the videos. With the first section, moving pixel's region is calculated and differentiated from the rigid object region. Then features are extracted as frame separation, flicker colour detection, source matching, radius of fire object region, gradient area and stream of frames are then stack into a video file. Then with the output of feature extraction, the flames are recognized whether they are present or not. If the flames are detected then fire alarm is raised to take fast action on fire and similarly send the result to the server for security purpose. The proposed procedure can be used on a huge video record. This paper allows a well-ordered atmosphere to inspect frame rate and provide robustness. The major advantage of the proposed algorithm is that it works well with CCTV cameras and low resolution video files that can detect the flames located at the far distance and can be used for commercial purposes. © 2006-2015 Asian Research Publishing Network (ARPN).