Research On Forest Smoke Detection Based On Video Analysis

Forest smoke detection system based on video analysis carries out machine vision related processing on forest surveillance video.These processing aims to detect whether there is smoke in the video and judge whether there is fire attacking in forest area.Four issues are focused in the system including de-haze processing,motion segmentation,feature extraction and classifier design.The main contents of the dissertation are as follows:1.Recent years,the deterioration of the environment becomes more and more serious.Haze weather often appears.Firstly,the haze processing of video images is studied,and typical defogging algorithms are analyzed experimentally.Considering the effect and time consumption,the algorithm of de-fogging combining with dark channel and histogram equalization is adopted.And the experiment of the no-fog image,the haze image,the mid-fog image and the dense fog image are respectively verified.2.Moving object segmentation is the prerequisite for the correct identification of smoke.In this paper,improved VIBE motion detected algorithm is used to analyze the forest surveillance video.Improved VIBE algorithm focused on two issues: the detection speed and the sensitivity of the scene switching.The switching variables are introduced to improve the detection speed based on the VIBE algorithm.The scene change detection and fast update are introduced to improve the sensitivity of the scene switching.Finally,the experimental verification of improving VIBE algorithm detection accuracy,detection speed and adoptability to scene change is carried out.3.Since motion areas in forest surveillance video may contain disturbing areas,analyzing the unique characteristics of smoke is crucial for subsequent smoke identification.The static and dynamic characteristics of the smoke are separately analyzed.The static characteristics mainly include the color,surface texture and edge contour.The dynamic characteristics mainly include the movement direction,the change of area and the periodic flutter intensity.And the divisibility of the six features is verified by cluster analysis.4.Three problems of applying BP neural network directly in forest fireworks recognition system are analyzed,and the structure of cascaded neural network is designed.Different networks are compared experimentally in forest monitoring video.The smoke recognition rate and efficiency are used to measure the detection of different networks and do some comparative analysis.Experimental results show that considering the recognition rate and efficiency,the secondary cascaded network designed in this paper has better effect.5.Using Visual Studio 2013 and MFC,the forest fire detection software based on video analysis was written,the forest fire detection algorithm was implemented using C++ language,and the software was tested.