Research On Video Flame Detection Algorithm Based On Feature Optimization

The damage caused by fire is enormous.Therefore,the study of fire detection is of great significance.At present,fire detectors on the market are mainly divided into two types: traditional type and image type.Traditional fire detectors determine whether there is a fire through the detection of temperature,smoke,etc.,the use of the range is limited,and can not be intuitive observation of the scene;Image-based fire detectors detect the fire through the identification of video images,and have a wide range of use.Through video observation,the real-time fire can be judged more accurately.Most of the existing image-based fire detectors use infrared images for detection.However,in the “Visual Image Early Fire Alarm System Technical Regulations”(CECS 448:2016)published by the China Engineering Construction Standardization Association,visible image early fires are The alarm system refers to a fire alarm system that uses only computer video image analysis technology for fire detection,and does not include a combination of image fire detection and infrared or ultraviolet fire detection.Therefore,this dissertation studies the flame detection algorithm based on ordinary color video.This paper presents a flame identification algorithm combined with video quality evaluation to achieve the detection and positioning of flame in video.The algorithm combines traditional fire detection and recognition methods based on video images with image quality assessment,and realizes dynamic screening of flame features based on video quality and feature-weighted flame discrimination based on video quality in the recognition process,thereby improving recognition speed and accuracy,and make it able to adapt to a variety of different quality video;Then use MATLAB to simulate the algorithm.Finally,using the existing flame video library,this paper compares the algorithm with the existing algorithms.The main innovations of this paper include: combining traditional video flame detection with video quality;proposing dynamic screening of flame features based on video quality;proposing feature-weighted flame discrimination based on video quality;and carrying out the algorithm for evaluating non-reference image sharpness Improvements,and fitting analysis with two classical algorithms;proposed a colorless image detection algorithm based on HSV color model,and compared with an existing algorithm.