| Most of the previous fire detection methods are only contact smoke detection methods based on the physical and chemical properties of smoke.These contact smoke detectors can only be installed in small indoor environments such as hotels and train cars,for many large indoor spaces and outdoor spaces.For many large indoor spaces and outdoor spaces such as forests,warehouses and other areas prone to fire,this type of contact detector can not effectively cover the entire area.And because of the different speeds of smoke propagation in different scenarios,the timeliness of such contact detectors has also been questioned.With the wide application of photographic equipment in daily life,the popularity of long-distance data transmission and the increase of d ata transmission speed,people can obtain video images taken by cameras at any time and any place at any time and any place.This provides conditions for fire safety testing in special locations.The image-based smoke detection method is mainly to find a v ideo image-based automatic smoke detection technology to provide full-time video monitoring for some special areas,the image-based smoke detection method has the advantages of real-time,low environmental constraints and the ability to provide visualized image information,and is currently receiving more and more attention.From the perspective of time,these video smoke detection methods can be divided into two categories,namely,a static image feature analysis method based on a single image and a smoke motion feature analysis method based on a video sequence.From the current literature on smoke detection technology,we can see that almost all of the smoke image acquisition methods are obtained through the photographic equipment,and the dynamic informat ion of these video data is often ignored.Whether it is the overall movement of smoke or its internal motion,the rational use of these dynamic information may greatly improve the detection accuracy of smoke.In view of this problem,this paper proposes a smoke detection method based on video sequence-based smoke timing changes.In the smoke detection process of this paper,some simple and efficient detection methods are used to detect the suspect ed smoke area on each image,the background difference method is used to detect the change area on each frame of the video sequence,and connecting the changed pixel points through the connected domain analysis to form a single moving object,finally,RGB color feature detection is performed on these moving objects to determine an object suspected of being smoked in each image.With the above simple detection method,the smoke image can be distinguished from most objects,but some objects similar to the smoke color cannot be distingui shed.Therefore,after the completion of the detection of the single image,it enters the timing detection process.In order to statistically change the characteristics of the suspected area,all moving objects in the continuous video image need to be tracked in real time,and the area and boundary information of the suspected area are recorded and stored,the area of the area and the trend of the boundary are used to determine whether the area is a smoke area.The selection of the variation characteristics is mainly based on the characteristics of smoke changes obtained by experimental observation.During the process of smoke propagation,a solid combustible material is burned as a process of solid p articles propagating in the air,the area of the propagation process is constantly expanding.And because smoke is a kind of fluid in the air,it does not have a fixed contour,that is,its contour changes at any time during the movement,so the boundary of the smoke will be different from the general rigid objec t,and it will always change constantly.In this paper,the Adaboost classification algorithm is used to train the appropriate smoke detection classifier.The Adaboost classification algorithm is a classification algorithm that seeks the best value through continuous iteration.The Adaboost classification algorithm can be used to find the optimal value of smoke area and boundary change rate from multiple smoke samples,and then apply it to the main program of smoke detection.Finally,the effectiveness of the smoke detection method is verified by several experiments.The selected experimental data is part of the training samples and the video segments taken in the field,these video segments are input into the smoke detection classifier and the data of the classifier are recorded.In order to understand the performance of the algorithm more intuitively,several different smoke detection algorithms are selected in the experimental phase for comparative analysis,including smoke feature detection algorithm based on single frame image and smoke motion direction detection algorithm based on optical flow method.The experimental results show that in the indoor environment,the positive detection rate of the algorithm is 94%,the false detection rate is less than 5%,the missed detection rate is lower than %2,and the positive detection rate of the smoke detection algorithm in the outdoor environment is above 90%.The false detection rate is lower than %7,and the missed detection rate is lower than %2.Compared wit h the smoke feature detection algorithm based on single frame image,the detection accuracy of the algorithm is greatly improved.At the same time,due to the use of the Adaboost classification algorithm for training the classifier,the algorithm is greatl y improved in terms of detection efficiency.The experimental results show that the average detection time of each image based on the optical flow direction detection algorithm is about 4.3s,and the detection method designed in this paper has an average d etection time of about 2.1s per image.The detection time is greatly reduced.It can be seen that the smoke detection algorithm of this paper has certain practicability. |
PDF Full Text Request