Fire Video Research Articles

Automated system for classifying images of video streams for timely detection of fires

ABSTRACT Early detection of a fire during its development stage is highly desired to minimize human and material losses. The improvement of the efficiency of these monitoring systems through the analysis of fire video data drawn from unmanned aerial vehicles is considered . Classification of video images when monitoring a fire event on the territory is used based on their segmentation into identical rectangular segments of a given size and then assigning them to one of three classes: Smoke, Flame, and Indifferent classes. The Walsh-Hadamard transform was used to form descriptors for the Weak classifiers. Descriptors were calculated for three Weak classifiers; first, the Walsh-Hadamard transform was calculated for the window of the entire segment, and its spectral coefficients were used for the first Weak classifier. Next, the descriptors were calculated in windows whose sizes were half and a quarter of the original window size. The classifier consisted of three independently-trained neural networks. A simple ensemble averaging block was used to combine the outputs of those networks. A special software code has been developed to classify video images. The code allows us to create a database of segments of the ‘Smoke’ and ‘Flame’ classes, determine the two-dimensional Walsh-Hadamard spectrum of video image segments, train fully connected neural networks, conduct exploratory analysis and evaluate the relevance of the calculated two-dimensional spectral coefficients. For validation we used real data from Сlosed Circuit Television cameras presented in open databases. The experimental studies on the classification of video data containing flames and smoke showed that the proposed method has an average smoke detection accuracy of 86% and an average flame detection accuracy of 89.5%. Errors II in smoke detection and flame detection had the averages of 13% and 4.5%, respectively.

Simulated fire video collection for advancing understanding of human behavior in building fires.

The goal of the present research was to develop a video collection of simulated fires to investigate how people perceive growing building fires. In fire safety science, a critical factor to occupant responses to building fires is the pre-movement period, determined by how long it takes an individual to initiate taking protective action with an incipient fire. Key to studying the psychological processes that contribute to the duration of the pre-movement period is presenting human subjects with building fires. One approach used in previous research is to present videos of building fires to individuals via scenarios. The numerical simulations used to model fire dynamics can be used to render videos for these scenarios. However, such simulations have predominantly been used in fire protection engineering to design buildings and are relatively inaccessible to social scientists. The present study documents a collection of videos, based on numerical simulations, which can be used by researchers to study human behavior in fire. These videos display developing fires in different types of rooms, growing at different rates, different smoke thickness, among other characteristics. As part of a validation study, participants were presented with subsets of the video clips and were asked to rate the perceived risk posed by the simulated fire. We observed that ratings varied by the intensity and growth rate of the fires, smoke opacity, type of room, and where the viewpoint was located from the fire. These effects aligned with those observed in previous fire science research, providing evidence that the videos could elicit perceived risk using fire simulations. The present research indicates that future studies can utilize the video library of fire simulations to study human perceptions of developing building fires as situational factors are systematically manipulated.

Unveiling Anomalies in Surveillance Videos

Abstract: The implementation of Convolutional Neural Network (CNN) algorithm for the detection of anomalies like fire and potholes in images or videos. Leveraging deep learning technique, the model aims to accurately identify fire, potholes, etc. within various contexts. The CNN architecture is trained on a dataset comprising diverse images to enhance its ability to generalize. This proposed project not only addresses safety concerns by promoting helmet usage and proper maintenance of roads but also contributes to the broader field of computer vision and object detection. This implementation involves training the CNN on annotated datasets, fine-tuning the model for optimal performance, and integrating it into a practical application for efficient anomaly identification. With a focus on real-time detection, the system utilizes image processing techniques and a trained CNN model to analyze visual data from cameras or video feeds. The system's versatility allows integration into surveillance systems, industrial sites, or any environment where fire detection is crucial for safety. By automating the detection process, the project contributes to minimizing human error and ensuring consistent monitoring. The proposed solution holds the potential to significantly impact safety protocols, particularly in industries where protective headgear is paramount. Potholes pose a significant threat to both drivers and pedestrians, leading to accidents and infrastructure damage. The proposed solution leverages CNN's ability to effectively process visual data, making it well-suited for image recognition tasks. Our approach involves training the CNN on a diverse dataset of road images to enable it to accurately identify potholes. The model will learn distinctive features and patterns associated with potholes, allowing for robust detection under various environmental conditions. Real-time implementation on embedded systems or cameras along roadways will enable instantaneous identification and alerting.

Deep Learning-Based Forest Fire Risk Research on Monitoring and Early Warning Algorithms

With the development of image processing technology and video analysis technology, forest fire monitoring technology based on video recognition is more and more important in the field of forest fire prevention and control. The objects currently applied to forest fire video image monitoring system monitoring are mainly flames and smoke. This paper proposes a forest fire risk monitoring and early warning algorithm, which integrates a deep learning model, infrared monitoring and early warning, and forest fire weather index. The algorithm first obtains the current visible image and infrared image of the same forest area, utilizing a smoke detection model based on deep learning to detect smoke in the visible image, and obtains the confidence level of the occurrence of fire in said visible image. Then, it determines whether the local temperature value of said infrared image exceeds a preset warning value, and obtains a judgment result based on the infrared image. It calculates again a current FWI based on environmental data, and determines a current fire danger level based on the current FWI. Finally, it determines whether or not to carry out a fire warning based on said fire danger level, said confidence level of the occurrence of fire in said visible image, and said judgment result based on the infrared image. The experimental results show that the accuracy of the algorithm in this paper reaches 94.12%, precision is 96.1%, recall is 93.67, and F1-score is 94.87. The algorithm in this paper can improve the accuracy of smoke identification at the early stage of forest fire danger occurrence, especially by excluding the interference caused by clouds, fog, dust, and so on, thus improving the fire danger warning accuracy.

FireMatch: A semi-supervised video fire detection network based on consistency and distribution alignment

Deep learning techniques have greatly enhanced the performance of fire detection in videos. However, video-based fire detection models heavily rely on labeled data, and the process of data labeling is particularly costly and time-consuming, especially when dealing with videos. Considering the limited quantity of labeled video data, we propose a semi-supervised fire detection model called FireMatch, which is based on consistency regularization and adversarial distribution alignment. Specifically, we first combine consistency regularization with pseudo-label. For unlabeled data, we design video data augmentation to obtain corresponding weakly augmented and strongly augmented samples. The proposed model predicts weakly augmented samples and retains pseudo-label above a threshold, while training on strongly augmented samples to predict these pseudo-labels for learning more robust feature representations. Secondly, we generate video cross-set augmented samples by adversarial distribution alignment to expand the training data and alleviate the decline in classification performance caused by insufficient labeled data. Finally, we introduce a fairness loss to help the model produce diverse predictions for input samples, thereby addressing the issue of high confidence with the non-fire class in fire classification scenarios. The FireMatch achieved an accuracy of 76.92% and 91.80% on two real-world fire datasets, respectively. The experimental results demonstrate that the proposed method outperforms the current state-of-the-art semi-supervised classification methods.

The Effect Of Fire Simulation For Student’s Knowledge About Fire In SMA X Depok 2022

Fire is one of the disasters that often occur and result in losses, both to people, assets, and productivity. Fires can occur due to the use of fire, gas explosions, electrical short circuits, HP exploding, natural factors, building conditions, such as the distance between buildings, materials, access to rescue, and weather factors. A school must meet the requirements of safety, health, comfort, and security. Schools must have a stable and sturdy building structure and a safe zone to withstand natural disasters. They should also have an active and/or passive fire protection system to prevent and deal with fire disasters. Knowledge and attitudes are the first parameters of preparedness, which form the basis of every school element’s behavior in disaster preparedness. The research objective is to determine the differences in knowledge scores before and after a fire simulation. This study uses a knowledge questionnaire that has been tested for validity. The sample size is 37 respondents, and the data is analyzed using a paired T-test. The results of the study showed differences in knowledge before and after the fire simulation (p-value = 0.002). Therefore, researchers suggest that schools routinely carry out simple fire simulations in collaboration with the Depok City Fire Department and incorporate topics regarding fire disasters in the curriculum. School can also conduct virtual fire drills or use other methods and media such as video fire simulations, fire safety posters, or safety inductions conducted before starting the first class of the day .

Fire Video Recognition Based on Channel Feature Enhancement

Fire Video Recognition Based on Channel Feature Enhancement

Image-Based Fire Detection Using Dynamic Threshold Grayscale Segmentation and Residual Network Transfer Learning

Research on fire detection based on deep learning has been increasing lately, and current models differ in accuracy and computational complexity. To reduce the loss of semantic information and improve prediction accuracy, this paper combines dynamic threshold grayscale segmentation and residual network transfer learning. An effective training data extension strategy integrates the experimental data of a real fire platform and a forest-image-based fire database, where the experimental data of the real fire platform are shot based on different lighting and backgrounds. It has been proven that this is a good solution that can effectively solve the problem of training robust deep models relying on datasets with large diversity, which can improve the overall performance of the training model. Then, we introduce a network architecture based on dynamic threshold grayscale segmentation and residual net transfer learning, which reaches residual net level accuracy on image-based fire detection tasks with less computation. We train the model with a self-built image dataset, test it on a general fire image dataset, and evaluate it on a self-built fire video dataset. In numerous experiments, our method produced a training and testing accuracy of 98.26% and 99.90%, respectively, with corresponding losses of 0.036 and 0.04. The evaluation accuracy was 90% under the condition of a complex sample environment. Compared with other models, our method has advantages in accuracy, training loss, and cost time. Comprehensive comparisons show effectiveness and improved overall performance.

Video Fire Detection Methods Based on Deep Learning: Datasets, Methods, and Future Directions

Among various calamities, conflagrations stand out as one of the most-prevalent and -menacing adversities, posing significant perils to public safety and societal progress. Traditional fire-detection systems primarily rely on sensor-based detection techniques, which have inherent limitations in accurately and promptly detecting fires, especially in complex environments. In recent years, with the advancement of computer vision technology, video-oriented fire detection techniques, owing to their non-contact sensing, adaptability to diverse environments, and comprehensive information acquisition, have progressively emerged as a novel solution. However, approaches based on handcrafted feature extraction struggle to cope with variations in smoke or flame caused by different combustibles, lighting conditions, and other factors. As a powerful and flexible machine learning framework, deep learning has demonstrated significant advantages in video fire detection. This paper summarizes deep-learning-based video-fire-detection methods, focusing on recent advances in deep learning approaches and commonly used datasets for fire recognition, fire object detection, and fire segmentation. Furthermore, this paper provides a review and outlook on the development prospects of this field.

Automatic real-time fire distance, size and power measurement driven by stereo camera and deep learning

Automatic real-time fire characterization is a crucial requirement of future smart firefighting. This work proposes a novel computer vision method to automatically measure the fire heat release rate, even when the camera is moving in real-time. Firstly, a portable binocular stereo camera is used to capture the real-time fire video stream that is fed into a pre-trained computer-vision model frame-by-frame to detect the fire region. By identifying the fire location inside the image, the real-time distance between the camera and the fire source is determined. This fire distance helps re-scale the images to match the input scale of the AI-image Fire Calorimetry. Then, the deep learning model can automatically output the transient fire power in real time. Results show that the stereo vision system is capable of accurately measuring the distance between the camera and the fire source, flame height, and power, with a relative error of less than 20%. This work provides an automatic and flexible way to measure the distance, power and hazard of fire in real-time, and such a method has broad applications in firefighting operations and decision-making.

Analysis of false alarm causes in video fire detection systems

Video-based fire detection systems represent an innovative path in fire signalling. Thanks to a suitably designed algorithm, a system of this kind can enable the detection of a flame based on its characteristics such as colour or shape, which were not previously used in classical fire detection systems. Video-based detection systems, due to their early stage of development in the fire protection market, are not yet a certified, fully tested method for early fire detection. This paper focuses on the analysis of possible causes of false alarms occurring in video-based fire detection systems in relation to classical Fire Alarm Systems (FAS). For this purpose, a video-based flame detection algorithm is designed and implemented to further analyse the phenomena occurring in such systems.

Fire Warning Algorithm based on Combining R-S Fusion Layer Images with Convolutional Neural Networks

In order to overcome the problems of low detection rate and high false alarm rate of fire detection system, a video fire detection algorithm based on R-S weighted model and convolutional neural network is proposed. First, the R-S fusion layer image is obtained by RGB color space and HSI color space; Then, the convolutional neural network is used to construct the fire detection algorithm. During the training process, the extracted R-S fusion layer image is sent to the convolutional neural network for training. During the detection process, the test image is sent to the model for detection. The experimental results show that the method has more effective flame detection effect.

Fire Detection Using Unmanned Aerial Vehicle

Fire is one of the most dangerous types of accidents that can occur and may cause huge harm to people's lives and threaten environmental resources by spreading quickly, which makes it out of control. This critical aspect requires an intelligent system to control it correctly and as soon as possible. To achieve this, a platform has been proposed for a fully automated fire detection system to overcome fire spreading, this goal has been achieved by utilizing the UAVs (Unmanned Aerial Vehicles) as an IoT service role equipped with Pixhawk4 as a flight controller and an onboard fire detection system using Raspberry-Pi4 equipped with a camera to detect the fires using Convolutional Neural Network (CNN) algorithm. The drone systems' particular properties of high speed, flexible movement, and easy control, as well as the embedded sensors in flight controller like gyroscope, accelerometer, and compass, make it the best choice for the real-time monitoring system. The whole system was tested and the real-time fire video was detected using the MobileNetv2 as a CNN pre-trained model which has trained and achieved an accuracy 99.79% .

Transformer-Based Fire Detection in Videos

Fire detection in videos forms a valuable feature in surveillance systems, as its utilization can prevent hazardous situations. The combination of an accurate and fast model is necessary for the effective confrontation of this significant task. In this work, a transformer-based network for the detection of fire in videos is proposed. It is an encoder-decoder architecture that consumes the current frame that is under examination, in order to compute attention scores. These scores denote which parts of the input frame are more relevant for the expected fire detection output. The model is capable of recognizing fire in video frames and specifying its exact location in the image plane in real-time, as can be seen in the experimental results, in the form of segmentation mask. The proposed methodology has been trained and evaluated for two computer vision tasks, the full-frame classification task (fire/no fire in frames) and the fire localization task. In comparison with the state-of-the-art models, the proposed method achieves outstanding results in both tasks, with 97% accuracy, 20.4 fps processing time, 0.02 false positive rate for fire localization, and 97% for f-score and recall metrics in the full-frame classification task.

Real-Time Forest Fire Detection by Ensemble Lightweight YOLOX-L and Defogging Method.

Forest fires can destroy forest and inflict great damage to the ecosystem. Fortunately, forest fire detection with video has achieved remarkable results in enabling timely and accurate fire warnings. However, the traditional forest fire detection method relies heavily on artificially designed features; CNN-based methods require a large number of parameters. In addition, forest fire detection is easily disturbed by fog. To solve these issues, a lightweight YOLOX-L and defogging algorithm-based forest fire detection method, GXLD, is proposed. GXLD uses the dark channel prior to defog the image to obtain a fog-free image. After the lightweight improvement of YOLOX-L by GhostNet, depth separable convolution, and SENet, we obtain the YOLOX-L-Light and use it to detect the forest fire in the fog-free image. To evaluate the performance of YOLOX-L-Light and GXLD, mean average precision (mAP) was used to evaluate the detection accuracy, and network parameters were used to evaluate the lightweight effect. Experiments on our forest fire dataset show that the number of the parameters of YOLOX-L-Light decreased by 92.6%, and the mAP increased by 1.96%. The mAP of GXLD is 87.47%, which is 2.46% higher than that of YOLOX-L; and the average fps of GXLD is 26.33 when the input image size is 1280 × 720. Even in a foggy environment, the GXLD can detect a forest fire in real time with a high accuracy, target confidence, and target integrity. This research proposes a lightweight forest fire detection method (GXLD) with fog removal. Therefore, GXLD can detect a forest fire with a high accuracy in real time. The proposed GXLD has the advantages of defogging, a high target confidence, and a high target integrity, which makes it more suitable for the development of a modern forest fire video detection system.

Development of early fire detection model for buildings using computer vision-based CCTV

A fire in a building directly affects the lives of its occupants. Therefore, a safer environmental system that can minimize the damage caused by a fire occurring indoors needs to be developed. Recent studies on computer vision-based rapid fire detection methods aim to overcome the limitations of general fire detectors and prevent false alarms using advanced deep learning technology. However, studies considering the development of a video fire detection model for indoor usage and its implementation in an actual test room are lacking. We developed a computer vision-based early fire detection model (EFDM) using an indoor closed-circuit television (CCTV) surveillance. The proposed EFDM derives the fire detection time through actual fire tests. The possibility and necessity of using fire detectors indoors is confirmed by comparing the results with the fire detection times of a general fire detector. The developed model achieves a recall, precision, and mAP0.5 performance of 0.97, 0.91, and 0.96, respectively. The fire recorded in the indoor fire video test dataset was detected within 8 s. The possibility of fire detection from three combustibles according to Underwriters Laboratories (UL) 268 B is confirmed via experimentation. A difference of up to 307 s is observed when the fire detection times of the EFDM and general fire detectors are compared. The useable range is confirmed by detecting a fire within 1 s of the maximum visible range of the CCTV. The proposed method can help contribute to the reduction of unfortunate property damages or casualties caused by a fire.

High-Precision Real-Time Forest Fire Video Detection Using One-Class Model

Machine learning-base fire detection methods play a vital role in the current forest fire monitoring systems. In the literature, fire detection is usually viewed as a two-class (or multi-class) classification task. However, this violates the most foundational hypothesis in machine learning, e.g., independent and identical distribution (i.i.d.), especially for the non-fire samples drawn from a complex forest background. Moreover, for omni-directional video -monitoring, the background is also always changing, which leads this violation to a worse situation. In this work, by relaxing the i.i.d. of non-fire samples, we aim to learn a one-class model that just relies on the fire samples. Considering the requirements of high-precision and real-time detection, training samples are directly constructed on the fire pixels without a complex feature transformation. Additionally, we also provide a batch decision-making strategy to speed up fire detection. This work also includes an extensive experimental comparison on the public forest fire videos, obtained by ground- or unmanned aerial vehicle (UAV)-monitoring cameras. Compared with the state-of-the-art methods, the results show the superiority of our proposal in terms of a high-fire detection rate, low-error warning rate, accurate fire location positioning, and real-time detection.

EdgeFireSmoke: A Novel Lightweight CNN Model for Real-Time Video Fire–Smoke Detection

The planet Earth is being affected by a series of wildfires, which have been steadily increasing over the last two decades. Forests have undergone deforestation due to natural forest fires and forest fires caused by man. These events are occurring on a global scale, and in Brazil, these wildfires are having an extreme impact on the Amazon forest as well as other forest biomes. This article proposes a novel lightweight convolutional neural network (CNN) model for wildfire detection through RGB images. This new method presents more advantages than the other methods used for the same task. Our CNN architecture can be used with aerial images from unmanned aerial vehicles and from video surveillance systems, combined with edge computing devices for image processing with a CNN. The proposed system is able to send wildfire alerts. The images do not have to be sent to a cloud computer as they can be processed in an edge device. However, it sends a string of alerts whenever a wildfire is detected. The evaluation of our proposed method showed that it required about 30 ms for the classification time, per image, and achieved an accuracy of 98.97% and an <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$F1$</tex-math></inline-formula> -score of 95.77%, which is a very promising result.

Application of colorimetry in neural network methods of fire detection in woodlands

Currently, neural network methods of fire detection have been used to monitor fires in forests. To date, such systems have been developed as the Prometheus project, the Fire Video Detector, and the Artificial Intelligence University method. These methods make it possible to determine a fire with an accuracy of more than 90 %, for which a combination of recurrent and light-line neural networks is used. The article proposes a method for improving the efficiency of neural network methods for determining fires in forests based on computer colorimetry. The use of this method makes it possible to increase the efficiency of neural network methods for detecting fire when using multiple cameras of a video surveillance system. Thus, the article presents a comparison of the performance of a convolutional neural network with and without the use of a colorimetric module. According to the results, the efficiency of work has increased by more than 20 %.

Real-Time Fire Detection Method for Electric Vehicle Charging Stations Based on Machine Vision

During the charging process of electric vehicles (EV), the circuit inside the charger plug is connected in series, the charger input voltage does not match the rated input voltage, the temperature caused by the severe heating of the charging time is too high for too long, and other factors are very likely to trigger a fire in the vehicle charging pile. In this paper, an improved You Only Look Once v4 (YOLOv4) real-time target detection algorithm based on machine vision is proposed to monitor the site based on existing monitoring equipment, transmit live video information in real-time, expand the monitoring range, and significantly reduce the cost of use. During the experiment, the improved neural network model was trained by a homemade fire video image dataset, and a K-means clustering algorithm iwasintroduced to recalculate the anchor frame size for the specific object of flame; the existing dataset was used to perform multiple divisions by using a tenfold cross-validation algorithm, thus avoiding the selection of chance hyperparameters and models that do not have generalization ability because of special divisions. The experimental results show that the improved algorithm is fast and accurate in detecting large-size flames in real-time and small-size flames at the beginning of a fire, with a detection speed of 43 fps/s, mAP value of 91.53%, and F1 value of 0.91. Compared with YOLOv3 and YOLOv4 models, the improved model is sensitive to detecting different sizes of flames. It can suppress false alarms well in a variety of complex lighting environments. The prediction frame size fits the area where the target is located, the detection accuracy remains stable, and the comprehensive performance of the network model is significantly improved to meet the demand of real-time monitoring. It is significant for developing the EV industry and enhancing emergency response capability.