Intelligent Fire Detector Based On Smoke And Flame Video Recognition

Fires are frequent natural disaster.Detecting and reporting fire early is the key to reducing the serious consequences caused by the fire.Therefore,fire detection technology has important research and application value.Compared with traditional fire detection technology,video recognition-based fire detection technology has the advantages of wide detection range,strong anti-interference ability,and high intelligence.It is especially suitable for outdoor and high-space building environments where traditional fire detection means are not ideal.Embedded video-based fire detectors have the advantages of good real-time performance and flexible system layout.They are the main development direction of video fire detection products.At present,domestic video fire detection products can only detect flame targets in video,and use analog video output,which is not compatible with existing video surveillance networks.Therefore they have not been widely used.The current research on video fire detection technology focuses on theory and algorithm innovation on the PC platform,without considering the application in embedded devices.In response to this situation,This paper proposes a design scheme for intelligent fire detectors based on flame and smoke recognition.This scheme is based on an embedded ARM platform and Linux system,and uses a web camera as a video acquisition device.On this basis,a flame and smoke target recognition algorithm is designed and The software was implemented,and the video transmission function of the web camera was also realized,thereby realizing real-time fire video detection.The main research work of this article is as follows:First of all,based on the research of the previous fire foreground target detection algorithm,this article uses a combination of vibe background model-based moving target detection algorithm and color filtering algorithm to obtain flame and smoke target images.Compared with the traditional single processing method,it can filter out More irrelevant interference to the target,and the edge of the target is clearer,so as to extract the target features more accurately.Secondly,based on the common algorithms of video fire detection technology and the projection of the video sequence on the time axis,this paper proposes a flame and smoke feature extraction method that combines the static and dynamic features of the image.The method first uses Moving object detection and color criteria to screen out suspect targets,and then superimposes suspicious target images of a continuous frame to construct a projection of the target image on the time axis.Finally,static and dynamic features are extracted based on the projected image.The projected image of the video sequence on the time axis can not only reflect the color and shape of the target image,but also reflect the motion changes of the target image.Compared with the traditional image dynamic feature extraction method,this method can reduce the interference caused by the irregular movement of flames and smoke,and improve the robustness of the feature.And Then,This paper collects fire and interference videos to form a sample set,and extracts the flame and smoke targets to form a feature library,designs an SVM multi-classifier and uses the feature library test set to classify the features to train the classification model,so as to achieve the simultaneous flame and smoke targets probe.The classification results of the validation set show that the classifier has a higher accuracy for flame and smoke target recognition.Finally,in this paper the Open CV computer vision library is used to implement and test the software on the Visual Studio platform,and then the source files of the software are crosscompiled on the Linux platform and transplanted to the embedded ARM platform for operation.Software test and hardware operation results show that the intelligent fire detector based on flame and smoke video recognition proposed in this paper can detect fires in the video and has broad application prospects.