Design And Realization Of Image-Based Flame Detection And Alarm System

Fire has always been one of the main disasters that endanger public safety and social property.Early detection of fire occurrence and reduction of fire losses are the direction of continuous in-depth research in the field of fire protection.With the rapid development of my country’s economic construction,the number of super-large space buildings continues to increase,and such buildings have a large fire load and are densely populated.Once a fire occurs,it will cause great damage.Because traditional fire detectors usually use contact judgment,this method is difficult to quickly and accurately complete fire warning in such large-scale spaces.Therefore,this paper studies and designs an image-based flame detection and alarm system to realize the detection and alarm of early fire flames in super-large spaces and outdoor scenes.The main research contents are as follows:(1)The design of the system flame detector is completed with the special monitoring chip of Huawei Hisilicon,and the collection and encoding of the visible light and near-infrared dualband video images are completed through its internal media processing platform,and the spectral characteristics of the flame are used in the near-infrared band to add The filter filters out visible light,which overcomes the problem of false negatives caused by the weak flames of early fires and the inability to collect clear images with visible light in harsh environments and weather.(2)For the collected video images,the method of combining moving target detection and color space segmentation is used to extract the suspected flame area in the visible light image,and the method of combining moving target detection and threshold segmentation is used to extract the suspected flame of the near-infrared image.region,and finally fill the extracted region completely by using the region growing method.Extract the features of the extracted suspected flame area,extract the color moment and local binary pattern in the visible light area;extract the circularity,contour roughness,area change rate,and centroid change rate in the nearinfrared area,and extract the extracted The features are combined into a sequence and input into the support vector machine to train the flame recognition model.Finally,the experiments show that the accuracy of the algorithm studied in this paper for flame detection reaches 98%.The algorithm is transplanted to the detector hardware to run,and the running speed of the algorithm is improved by scaling the image size,and the real-time performance of the algorithm is improved on the premise of ensuring the stability of the recognition accuracy,so that the system does not need a high-performance host to participate in the operation and can alarm offline.(3)Design and implement the monitoring system software of the system,including video browsing and playback,equipment search and control,alarm reception display and log recording,etc.,to remind managers in time when a fire occurs and provide fire scene conditions.The experimental test of the system shows that the system can detect the occurrence of early fire in time and give an alarm,and has a low false alarm rate.