| In this paper, we take the early fire detection indoor based on the video as the goal, researching the fire smoke detection algorithm. This paper presents a method of combining multiple features to reduce false alarms caused by environmental factors interfere.The article use codebook for background modeling to extract motion scene area, and then combined inter-frame motion pixel ratio with color characteristics for the initial judgment suspicious fire smoke area extracting. Through this process, reducing the size of the subsequent processing of the picture area, improving program efficiency. In the detection process, an average gradient information, area and boundary irregularities are being used for precise positioning, greatly improve the accuracy of detection.Finally according to the actual monitoring of the scene,we design a background update policy for the three common interferences in actual monitoring scene. By those processes we can make sure the model in the program maximum close to the actual surveillance,which for long-term stability essential part of video surveillance.This article using support vector machine to classification. Firstly through the manual processing, Motion pixel extraction ratios between frames, The area change rate, boundary irregularity are Extracted, and then combined with the color features of static images to trains the classifier. In the detection process, we set a time window, to make the judgment with the average gradient.In the detection of practical application, This method in indoor environment, have a good identification result on the pedestrian light changing and wearing with smoke-like clothes. And through the cigarette test, This method can also identify effectively fire smoke.Therefore,the recognition results of this method with strong reliability. |
