Flame Detection Algorithm Research Articles

A Study on Flame and Smoke Detection Algorithm Using Convolutional Neural Network Based on Deep Learning

Recently, cases of large-scale fires, such as those at Jecheon Sports Center in 2017 and Miryang Sejong Hospital in 2018, have been increasing. We require more advanced techniques than the existing approaches to better detect fires and avoid these situations. In this study, a procedure for the detection of fire in a region of interest in an image is presented using image pre-processing and the application of a convolutional neural network based on deep-learning. Data training based on the haze dataset is included in the process so that the generation of indoor haze smoke, which is difficult to recognize using conventional methods, is also detected along with flames and smoke. The results indicated that fires in images can be identified with an accuracy of 92.3% and a precision of 93.5%.

基于图像的火焰检测算法

基于图像的火焰检测算法

On the Analysis and Detection of Flames With an Asynchronous Spiking Image Sensor

We have investigated the capabilities of a custom asynchronous spiking image sensor operating in the Near Infrared band to study flame radiation emissions, monitor their transient activity, and detect their presence. Asynchronous sensors have inherent capabilities, i.e., good temporal resolution, high dynamic range, and low data redundancy. This makes them competitive against infrared (IR) cameras and CMOS frame-based NIR imagers. In this paper, we analyze, discuss, and compare the experimental data measured with our sensor against results obtained with conventional devices. A set of measurements have been taken to study the flame emission levels and their transient variations. Moreover, a flame detection algorithm, adapted to our sensor asynchronous outputs, has been developed. Results show that asynchronous spiking sensors have an excellent potential for flame analysis and monitoring.

A convolutional neural network-based flame detection method in video sequence

Computer vision-based fire detection is one of the crucial tasks in modern surveillance system. In recent years, the convolutional neural network (CNN) has become an active topic because of its high accuracy recognition rate in a wide range of applications. How to reliably and effectively solve the problems of flame detection, however, has still been a challenging problem in practice. In this paper, we proposed a novel flame detection algorithm based on CNN in real time by processing the video data generated by an ordinary camera monitoring a scene. Firstly, to improve the efficiency of recognition, a candidate target area extraction algorithm is proposed for dealing with the suspected flame area. Secondly, the extracted feature maps of candidate areas are classified by the designed deep neural network model based on CNN. Finally, the corresponding alarm signal is obtained by the classification results. The experimental results show that the proposed method can effectively identify fire and achieve higher alarm rate in the homemade database. The proposed method can effectively realize the real-time performance of fire warning in practice.

Flame Detection Using Generic Color Model and Improved Block-Based PCA in Active Infrared Camera

In this paper, we proposed an all-weather flame detection algorithm which could make full use of active infrared cameras presently installed in many public places for surveillance purposes. Firstly, according to the different spectral imaging results in day and night, we propose a video type classification algorithm (VTCA) via imaging clues. VTCA could help us select different flame visual features in color image and infrared image. Secondly, we use a generic YCbCr-color-space-based chrominance model to extract regions of interest (ROI) of flame. Thirdly, two flame dynamic features are used to verify the candidate ROIs, which are common flame flicker feature and an improved block-based PCA in consecutive frames. The experimental results show that the proposed flame detection model has been successfully applied to various situations, including day and night, indoor and outdoor on our test video datasets, and it gives a better performance compared with other state-of-the-art methods.

Experimental study of corner fires—Part II: Flame spread over MDF panels

Having explained the characteristics of a corner fire in the configuration of Single Burning Item (SBI) test in Part I [1], the results of three flame spread experiments conducted with Medium Density Fiberboard (MDF) panels are discussed. The fire growth, in terms of flame heights and spread, is examined from two different angles visually and through Video Fire Analysis (VFA) with a flame detection algorithm. Total Heat Release Rates (HRR) and Smoke Production Rates (SPR), as well as total heat fluxes at several characteristic locations, are presented. Moreover, temperature evolutions are discussed for multiple locations and through the thickness of the panels. Also the backside temperatures, important as boundary condition for numerical simulations, are reported. The corner fire tests with MDF panels yield an average peak HRR of 151 kW and an average total heat flux exceeding 60 kW/m2 close to the burner, with average flame heights surpassing 1.5 m in about 60 s.

Flame detection via Quaternion Discrete Cosine Transform based Spectral Saliency

To address the problem of detecting the specific object of flame, this paper proposes quaternion discrete cosine transform based video flame detection algorithm in conjunction with the bottom-up visual attention mechanisms and the flame features. According to the characteristics of the flame to be detected, the brightness, color, texture and motion information of the image are respectively defined and extracted to form a quaternion image, and then the flame saliency map is produced by Quaternion discrete cosine Transform with this quaternion image. Extensive experimental results verify the accuracy and robustness of the proposed method on flame detection.

Flame Spread Monitoring and Estimation of the Heat Release Rate from a Cable Tray Fire Using Video Fire Analysis (VFA)

The paper presents an application of video fire analysis (VFA) to a cable tray fire scenario, a scenario of interest to nuclear power plants where cables constitute a substantial part of the combustible load. Five horizontal trays, each 2.4 m long and 0.45 m wide, are positioned with a 0.3 m spacing (in the vertical direction) and set-up against an insulated wall. Each tray contains 49 power PVC cables of 13 mm outer diameter. Ignition is performed with an 80 kW propane burner centrally positioned at 0.2 m below the first (i.e. lowest) tray. Thanks to a flame detection algorithm, the VFA technique allows a reconstruction of the heat release rate (HRR) profile based on the estimated temporal evolution of flame and extinction fronts at the level of each tray. The obtained profile is generally in good agreement with the HRR measurements (from oxygen depletion calorimetry). However, the time to reach the peak HRR is overestimated by approximately 100 s and the peak HRR of 3 MW is underestimated by 17%. These results are nevertheless encouraging because they provide confidence in the flame spread measurements (which are estimated for the case at hand between 2 mm/s and 5 mm/s). Therefore, a systematic application of VFA to real-scale experiments is believed to generate a valuable set of data (e.g. flame spread rates) for the development of simplified modelling.

Global analysis of multi-compartment full-scale fire tests (‘Rabot2012’)

A global analysis is presented of a series of four multi-compartment full-scale fire tests carried out in an apartment located in a high-rise building (called the ‘Rabot’ tower in the city of Ghent, Belgium) in 2012. The data set, available at http://multimedialab.elis.ugent.be/rabot2012/, is called ‘Rabot2012’. These tests are believed to provide a valuable set of experimental data for fire modelling (CFD and two-zone) and monitoring purposes, for a number of reasons. First of all, the fuel package of intermediate complexity (i.e. two to three furniture items) and the ignition location were designed to focus on the initial flame spread and then the possible occurrence of secondary (and in one test tertiary) ignition. Also, very good repeatability is illustrated during the first minutes of the tests. Regarding the sensing aspect, in addition to the conventional tools such as thermocouples and velocity probes, a Video Fire Analysis (VFA) has been performed by applying smoke and flame detection algorithms to the footages obtained from a multi-view video network. Moreover, laboratory tests have been performed, putting the fuel packages as applied in the full-scale tests under a hood in free-burning conditions. The repeatability during the first minutes is confirmed and subsequent differences in fire development (free burning versus compartment set-up) are explained. This presentation of the ‘Rabot2012’ fire tests comprises: (1) a detailed description of the experimental set-up (i.e. geometry, fuel package and instrumentation), (2) an overview of the fire scenarios obtained in each of the four tests, and (3) a detailed account of the data processing carried out. Finally, elements of exploitation of the collected set of data are discussed.

Graphene based multifunctional flame sensor

Recently, graphene has attracted much attention due to its unique electrical and thermal properties along with its high surface area, and hence presents an ideal sensing material. We report a novel configuration of a graphene based flame sensor by exploiting the response of few layer graphene to a flame along two different directions, where flame detection results from a difference in heat transfer mechanisms. A complete sensor module was developed with a signal conditioning circuit that compensates for any drift in the baseline of the sensor, along with a flame detection algorithm implemented in a microcontroller to detect the flame. A pre-defined threshold for either of the sensors is tunable, which can be varied based on the nature of the flame, hence presenting a system that can be used for detection of any kind of flame. This finding also presents a scalable method that opens avenues to modify complicated sensing schemes.

Flame detection algorithm based on a saliency detection technique and the uniform local binary pattern in the YCbCr color space

Computer vision-based fire detection involves flame detection and smoke detection. This paper proposes a new flame detection algorithm that is based on a saliency detection technique and on the uniform local binary pattern (ULBP). In still images and video sequences, an area that contains an open flame is always noticeable because fire is an exceptional event. Thus, to utilize the color information of flame pixels, the probability density function (pdf) of the flame pixel color can be obtained using Parzen window nonparametric estimation. This a priori pdf is then fused with the saliency detection phase as top-down information so that the flame candidate area can be extracted. To reduce the number of false alarms, the image texture of the candidate area is analyzed by ULBP, and an exponential function with two parameters is utilized to model the texture of the flame area. According to the experimental results, our proposed method can reduce the number of false alarms greatly compared with an alternative algorithm, while ensuring the accurate classification of positive samples. The classification performance of our proposed method is proven to be better than that of alternative algorithms.

Aid to Detect Fire and Smoke in Open Space Surveillance

As we know artificial neural network is use for analysing and training the sensed data which gathered by different channels. In this paper we use different combinations of techniques to detect smoke and flame detection algorithms in a video. The past sensed data cannot respond quickly and fire and smoke may not capture quickly. The region partitioning technique is proposed, which will increase the accuracy and also reduce test data so that rather than using a whole frame in a video it uses on part of that frame. The flame characteristics are used for normalization data we are processing. The use of neural network in combination with image processing can improve the accuracy and also help to predict the data. In improvement the wrong alarm problem is decreased. The double band method is use to detect fire. The region which we are analysing for detecting fire and smoke is calculated directly so that it can reduce computational time. Keyword: Artificial Neural Network, Flame Detection, Image Processing., Smoke, Surveeillance

DSP Embedded Early Fire Detection Method Using IR Thermal Video

Here we present a simple flame detection method for an infrared (IR) thermal camera based real-time fire surveillance digital signal processor (DSP) system. Infrared thermal cameras are especially advantageous for unattended fire surveillance. All-weather monitoring is possible, regardless of illumination and climate conditions, and the data quantity to be processed is one-third that of color videos. Conventional IR camera-based fire detection methods used mainly pixel-based temporal correlation functions. In the temporal correlation function-based methods, temporal changes in pixel intensity generated by the irregular motion and spreading of the flame pixels are measured using correlation functions. The correlation values of non-flame regions are uniform, but the flame regions have irregular temporal correlation values. To satisfy the requirement of early detection, all fire detection techniques should be practically applied within a very short period of time. The conventional pixel-based correlation function is computationally intensive. In this paper, we propose an IR camera-based simple flame detection algorithm optimized with a compact embedded DSP system to achieve early detection. To reduce the computational load, block-based calculations are used to select the candidate flame region and measure the temporal motion of flames. These functions are used together to obtain the early flame detection algorithm. The proposed simple algorithm was tested to verify the required function and performance in real-time using IR test videos and a real-time DSP system. The findings indicated that the system detected the flames within 5 to 20 seconds, and had a correct flame detection ratio of 100% with an acceptable false detection ratio in video sequence level.

Comparison of intensity flickering features for video based flame detection algorithms

Flickering is the most explicit visual characteristic of flames. Flames flicker in height, size and in brightness. Video based flame detection algorithms often analyze flickering of pixel intensities over time to detect flames. In this study we investigate five different pixel intensity flickering features based on methods presented in previous work. We compare the classification rates that features achieve on a large video database containing flames and non-flame objects. We depict differences of the features to each other and explain how these differences affect the classification rates. We point out components that should be considered for the design of flickering features.

Early Flame Detection in Video Sequences based on D-S Evidence Theory

A multi-feature fusion early flame detection algorithm based on D-S evidence theory is proposed. In this algorithm, first the method based on YcbCr and RGB color spaces is used for extracting the flame region of interest. Then flame classifiers based on flame flicker frequency and flame image correlation between frames are selected as two features of D-S feature fusion, the basic probability assignment functions of two features are defined. Finally, the combination discipline of D-S evidence theory is used to determine the final result of all the feature classifiers. Experiments show that the proposed detection algorithm gives 89.5% correct flame rate with a 4.5% false alarm rate and the method is efficient and fast with wide application prospects in fire detection.

A Real-time Video Fire Flame and Smoke Detection Algorithm

Real-time video fire flame and smoke detection method based on foreground image accumulation and optical flow technique is presented. Accumulation images are calculated of the foreground images which are extracted using frame differential method. Two parameters are used for the foreground image accumulation to differentiate flame candidate areas from that of smoke. The flame regions are recognized by a statistical model build by foreground accumulation image, while the optical flow is calculated and a motion feature discriminating model to recognize smoke regions is used. The algorithm could realize the real-time fire detection of the following three detection cases: fire with flame and none smoke, fire with smoke and none flame, and fire with both flame and smoke.

카메라 영상을 이용한 연기 및 화염의 조기 감지 최신 연구 동향

시대가 발전함에 따라 초고층 건물들이 도처에 세워지고 밀집되어 있다. 이러한 건물에 화재가 발생되면 발화지점 근처로 불이번지면서 대형화재의 위험성이 높아지고 이에 따른 인명 및 재산 피해가 증가한다. 따라서 이런 대형화재를 예방하고 피해를 최소화하기 위해서 화재를 미연에 감지하는 화재감지 기술에 대한 필요성이 높아지고 있다. 화재를 감지하기 위해 열감지기, 연기감지기, 불꽃감지기 등을 사용하는 방법이 있으나 본 논문에서는 감시 카메라에서 들어오는 입력 영상을 분석하여 화염과 연기를 초기에 감지하는 화재감지 시스템의 최근 연구 동향을 알아보고자 한다. 또한 이러한 화염과 연기 감지 알고리즘들을 다양한 형태의 동영상을 이용하여 구현 및 성능을 평가하였다. With the rapid development of technology, skyscrapers are widely spread and they are tightly coupled. If fire occurs in a building, it is easily spread to neighboring buildings, resulting in the large number of victims and property damages. To remove fire disasters, the need for early fire detection techniques is increasing. To detect fire, detecting devices for heat, smoke, and flame have been used widely. However, this paper surveys and presents the latest research which focuses on early smoke and flame detection algorithms and systems with camera's input images. In addition, this paper implements and evaluates the performance of these flame and smoke detection algorithms with several types of movies.

A new synthesis method for signals for testing of flame-detection algorithms

In this contribution, the development of a method for generating input data, for testing and optimising fire recognition algorithms, is presented. This work concentrates on the analysis and evaluation of sensor signals for detecting electromagnetic radiation in the infrared spectrum (IR). A current problem is the differentiation between the signal of a flame and the effect of non-fire IR sources in the same IR wavelength area that cause disturbance signals. The objective is to achieve a highly sensitive, early detection of fire and at the same time suppress false alarms. For testing of signal-processing algorithms, a great number of input data must be available reflecting a great variety of combinations of flame and non-fire IR source scenarios. Generally, these data records are created by test fires and physical simulation of simultaneous non-fire IR sources. The aim of this work is to develop a method for generating such input data based on a reduced number of experimental data. Our approach is to gain model functions based on measured sensor signals that are used to create parametric models. Assuming the linearity of the sensor transfer function, the synthesised signal components are superposed linearly and simulation signals are generated as new input signals. A uniform approach was developed for the simulation of signals caused by flames and non-IR sources. Three signal components were defined by their frequency range, the base function with f<1 Hz , the flickering component with 1 Hz⩽f<10 Hz and the noise component with 10 Hz⩽f<256 Hz . The modelling of the base function was carried out by cubic spline interpolation. Assuming that the flickering and noise components are stationary and non-deterministic, autoregressive models are used for the analysis of signals and for generating new random sequences. Comparison of the measured signal and the respective generated simulation data show a good consistency. Hence, considerably more effective input data records can be created for testing and optimisation of signal-processing algorithms. Generally, the method presented can also be applied for the simulation of signals generated by sensors utilising other physical detection principles.