The Design And Implement Of Fire Smoke Detection On Video

Fire is a kind of uncontrolled burning,which poses a great threat to the safety of people's production and life.Smoke is one of the important image features that appeared in the early stage of a fire.Therefore,accurate and real-time smoke detection of surveillance video in various environments is an important means to reduce fire hazards.In this paper,the smoke event detection technology in surveillance video is researched.Firstly,the moving target is detected by continuous video frames,and the smoke candidate region is determined by certain screening conditions.Then the candidate region is used as the prior knowledge of subsequent target recognition and classification.Through the feature extraction and feature fusion of this region and the support vector machine to classify the obtained feature vectors,the video classification results are obtained,which not only meets the real-time detection of video smoke detection,but also has high detection accuracy.Furthermore,a video smoke detection algorithm based on Fast R-CNN network is proposed,which uses the extracted smoke candidate region information to filter the region suggestion frame extracted by Fast R-CNN,which reduces the time complexity and improves the detection for subsequent training.Speed and accuracy.The main work of this paper is as follows:(1)An algorithm for smoke candidate region extraction based on background modeling and smoke characteristics is proposed.First,the image filtering and color conversion models are used to process the noise and illumination imbalance in the image,eliminating the effects of noise and illumination imbalance in smoke video.Then,the background image of the video is established by using the ViBe background modeling method to obtain the moving target area.Finally,the RGB spatial color information of the target area is used to determine whether the area is a candidate smoke area.Provides the basis for subsequent machine learning to detect video smoke targets.(2)A video smoke detection method based on smoke candidate region and multi-features is proposed.By extracting the candidate smoke regions for multi-feature extraction and extracting the extracted features separately,the SVM classifier with smoke detection capability can be obtained.By detecting the contour irregularity feature and the moving direction statistical feature of the smoke candidate area and discriminating,the texture feature extraction is performed for the smoke candidate area which simultaneously conforms to the smoke contour feature and the moving direction statistical feature,and then the SVM support vector machine is used.The extracted texture features are discriminated and the smoke detection results are obtained.The effectiveness of the proposed algorithm is verified by experiments.Accuracy of video smoke target detection can be improved by the combination of dynamic and static features(3)A video smoke detection algorithm based on smoke candidate area and Fast R-CNN network is designed.The algorithm successfully applies the smoke motion information in the video and the feature extraction ability of Fast R-CNN network to fire smoke detection,and solves the video Smoke has long relied on the limitations of hand-designed extraction features.This chapter first introduces the architecture of the model,establishes a smoke detection model,and filters the regional suggestion box generated by the search algorithm through the smoke candidate area.Then the design idea and algorithm implementation process of the algorithm are introduced.Finally,the smoke detection effect of different video frames in the same video is compared by experiments.The detection effect of the algorithm in different scenarios is also verified,and the robustness of the proposed algorithm is demonstrated.By comparing the response speed of the algorithm and the Fast R-CNN algorithm to smoke,the timeliness of the smoke detection is demonstrated by the algorithm,and the algorithm can timely warn the fire.