ARID ZONE JOURNAL OF ENGINEERING, TECHNOLOGY & ENVIRONMENT AZOJETE September 2022. Vol. 18(3):469-492 Published by the Faculty of Engineering, University of Maiduguri, Maiduguri, Nigeria. Print ISSN: 1596-2644, Electronic ISSN: 2545-5818 www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 453 ORIGINAL RESEARCH ARTICLE TRADITIONAL SENSOR-BASED AND COMPUTER VISION-BASED FIRE DETECTION SYSTEMS: A REVIEW A. Hassan and A. I. Audu* Department of Computer Engineering, University of Maiduguri, Maiduguri, Nigeria *Corresponding author’s email address: lm324fairchild@yahoo.com 1.0 Introduction Fire outbreak is one of the leading hazards endangering human lives, properties and the environment around the world today. It causes huge number of death, massive economic loss and serious ecological problems (Asih et al., 2018; Huang et al., 2017; Khan et al., 2019; Singh and Sharma, 2017). In recent years, a great number of fire incidents were reported across Nigeria. For instance, in Kano State, Fire Service records shows that 134 human lives and 635- million-naira assets were lost due to fire outbreaks in the state between January 2021 to December 2021(Rabiu, 2022). A typical incidence was that a fire event reported at Sabon- Gari Market in Kano City, Nigeria in which properties and shops worth millions of Naira were lost and took more than 10 hours to be put out completely by the Kano State Fire Service team. Not less than 64 people, including eight fire fighters, were reported to have gotten different degrees of injuries from a petroleum tanker explosion at a filling station in Sharada ARTICLE INFORMATION ABSTRACT Fire emergency is one of the dominant disasters which is threatening human society, infrastructure, and the environment in the world. Fire disaster can lead to a great number of human casualties, loss of vast natural resources and serious infrastructural damage. To minimize these such catastrophic menace, protect and save human lives and infrastructure, substantial research efforts have been devoted to developing fire detection systems using different approaches. This paper discusses fire detection systems and summarizes the state-of–the-art achievements. A total number of 60 publications related to fire detection methods are extracted from IEEE and Google Scholar database and analysed. Different fire detection methods are reviewed, and their merits and shortcomings are highlighted. The literature review presented in this paper illustrates the advantages of computer vision-based fire detection approaches over the traditional sensor-based approaches which are characterised by limited detection range, transport delay, and high false alarm. In addition, it shows that the high hardware requirement and computational complexity remain the key drawbacks of the computer vision-based methods and most of the research works made no attempt to implement such methods on embedded hardware platform such as FPGA. Further research efforts are required to design, develop, and implement computer vision-based systems for fire detection on embedded hardware targets in an efficient and computationally less intensive manner. © 2022 Faculty of Engineering, University of Maiduguri, Nigeria. All rights reserved. Submitted 25 January, 2022 Revised 9 April, 2022 Accepted 10 April, 2022 Keywords: Fire detection Computer vision Convolutional Neural Network Deep learning http://www.azojete.com.ng/ mailto:lm324fairchild@yahoo.com mailto:lm324fairchild@yahoo.com Arid Zone Journal of Engineering, Technology and Environment, September, 2022; Vol. 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818; www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 470 area of Kano State On 23 May 2021 (Rabiu, 2022). In order to mitigate against such huge losses and save valuable human lives, accurate and timely fire detection systems are inevitable (Islam et al., 2015). With advances in computing and electronics, many fire detection systems have been proposed in the last decades using different methodologies and approaches. Existing fire detection methods are: smoke detector, flame detector, heat detector, gas sensor (Cheong et al., 2011; Courbat et al., 2011; Fonollosa et al., 2016; Hoefer and Gutmacher, 2012; Krüll et al., 2012; Liu and Kim, 2003; Solórzano et al., 2022; Sowah et al., 2014; Wang et al., 2020) and computer vision (Asih et al., 2018; Chen et al., 2019; Filonenko and Danilo, 2018; Hossain et al., 2019; Huang et al., 2017; Jinlan et al., 2016; Kaabi et al., 2017; Monte et al., 2017; Rabiu, 2022; Russo et al., 2018; Santamaria-guerrero et al., 2018; Yuanbin, 2016). These methods have been classified into traditional sensor based and computer vision based methods (Chen et al., 2019). This paper aims at reviewing existing methods proposed for fire detection and highlighting future research direction. The remaining part of this paper is organized as follows. Section 2 presents the traditional sensor based methods. Section 3 presents the computer vision based methods. Finally, a summary and future direction is presented in section 4. 2. Traditional Sensor-Based Fire Detection Systems The conventional sensor based methods for fire detection measure ultraviolet or infrared radiation, heat, gas or particulate emissions generated by combustion (Wang et al., 2020). The operation principle of these methods is based on employing different physical parameters such as: light scattering, flame particles or air density fluctuation, ion capture by flame particles, smoke concentration to detect fire incidents (Chen et al., 2019; Vasiliev et al., 2017). However, the shortcomings of such methods include suffer from transport delay, conduction delay, and limited range detection. Additionally, these devices are not suitable for outdoor applications and can produce high false alarm rate (Khan et al., 2019; Rabiu, 2022). The operation principles, advantages and drawbacks of these methods are discussed in the subsequent section. 2.1 Thermal Detectors Fire detection systems based on heat sensing are designed to activate when the ambient temperature or rate of rise of temperature exceeds a set point. These sensing devices are not good for early warning due to the fact that they have to be in very close contact with the fire to be activated. However, they are so easy and inexpensive and highly reliable in areas where other sensors such as smoke detectors have a high level of false alarms like kitchens. They are also best suited in areas of home where systems based on smoke sensing cannot perform effectively due to extreme temperatures (Li, 2011; Perilla et al., 2018). Rate of rise (RoR) thermal sensing detectors, respond to emergency change or rise in ambient temperature from a normal baseline conditions. The set alarm criteria will be activated by any sudden increase in temperature. These kinds of thermal detectors can respond to a lower set conditions than if the set point were fixed (Li, 2011). file:///C:/user/Downloads/azojete143/www.azojete.com.ng mailto:lm324fairchild@yahoo.com Hassan and Audu: Traditional Sensor-Based and Computer Vision-Based Fire Detection Systems: A Review. AZOJETE, 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818, www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 471 One of the innovative thermal detectors for fire detection applications is the distributed fibre- optic temperature sensor. The optical fibre sensor cable has a faster response in detecting fluctuations in temperature compared to traditional thermal detectors due to its low mass. It is also completely immune to all sorts of nuisance emissions. In the late 1980’s, the distributed fibre optic temperature sensors using Rayleigh and Raman scattering were introduced. They have been employed in fire detection system for application areas with difficult ambient conditions, which include tunnels, underground railways and stations, conveyor lines, steal works and petroleum and chemical industries (Liu and Kim, 2003). Distributed optic fibre sensor using Rayleigh scattering measure the temperature by detecting changes in the quantity of reflected light, during the time the fibre is micro-bent because of heating effect. The fibre optic sensors based on Rayleigh scattering are majorly employed in applications, which include road tunnels and underground installations (Liu and Kim,2003). The distributed fibre optic sensing devices based on Raman scattering detects the change in temperature by measuring the ratio of stokes to anti-stokes back scattered intensity signals as a function of temperature (Liu and Kim, 2003). The Brillouin scattering based fibre optic device is one of the likely successors of both Rayleigh and Raman scattering systems for temperature sensing. The distributed optical fibre sensor using Brillouin scattering also has superior spatial and thermal resolutions in comparison with the Rayleigh and Raman scattering systems (Liu and Kim, 2003). 2.2 Smoke Detectors Smoke is generated much earlier before other fire signatures during the phases of fire incident growth and development. The chance for successful fire suppression, escape and survivability can be maximized by initial rapid detection of smoke at very low levels by fire detection systems (Liu and Kim, 2003). Fire detection was performed mainly by visual inspection and confirmation by people. In an effort to develop a sensing device for toxic gases, a sensor that could detect smoke was accidently discovered by the Swiss chemist Walter Jaeger during the late 1930’s, there by paving way for development of modern smoke sensing devices (Sowah et al., 2014). The first automatic smoke detector developed was based on the sensing of the ionization current using americium-241 as a radioactive source in 1940 (Fonollosa et al., 2016). Fire detection systems based on smoke detection are much new innovation and commonly used in residential and life safety application during the 1970’s and 1980’s. Fundamentally, there are two sensing approaches for smoke detection that are employed for commercially available fire detection devices. These approaches are photoelectric detectors that use light scattering and ionization detectors. Smoke detectors mimic human sense of smell and are designed to recognize fire incidents while in their smouldering or early flame phase. The great benefit of smoke detectors is their capability to detect fire while it is yet in its incipient level. They are most preferred method for life safety and high content value applications due to the fact that they provide added space for emergency team to respond and control the developing fire incident before dangerous damages to happen. Fire detection systems using smoke sensing elements have greater immunity to inadvertent alarms and they are normally more expensive http://www.azojete.com.ng/ mailto:lm324fairchild@yahoo.com Arid Zone Journal of Engineering, Technology and Environment, September, 2022; Vol. 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818; www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 472 to get installed but much more reliable with an extremely low chance of false alarm if properly chosen and designed (Fonollosa et al., 2016; Perilla et al., 2018). Optical smoke sensing devices uses the light scattering nature of smoke particles to detect the presence of smoke during fire event. On the other hand, ionization smoke sensors are designed with an ionization chamber which contains a radioisotope, usually Americium. In the absence of smoke particles, the ionized air molecules within the chamber provide the passage of a small amount of electrical current between the charged electrodes in the chamber. The presence of smoke particles causes a drop in the current between the electrodes. The drop in current between the electrodes results in a fire warning being activated (Sowah et al., 2014). Standard fire detection systems rely on the sensing of backscattered light emitted by a Light Emitting Diode (LED) and reflected on smoke particles. However, such devices suffer from high response time, false alarm and have no capability to detect non-smoking fires (Courbat et al., 2011). 2.3 Flame Detectors Flame detectors mimic the human sense of sight and represent one of the major kinds of automated fire detection approach. These methods are line of sight systems which operate on either infrared, ultraviolet or combination principle. Radiant energy which is in the 4,000 to 7,700 angstroms range is an indicator of a flaming condition. The fire sensing device based on flame detection identifies the fire signature and transmits a signal to the fire notification module. Not like thermal and smoke based fire detection methods, line of sight must exist between the flame detection device and the fire source to recognize emitted fire signatures. But the application of these methods is severely limited. The benefit of these flame detectors is that they are very reliable in a noisy environment. They mostly deployed in high value energy and transportation applications where other fire detection devices based on traditional sensors would suffer from spurious activities. They are also commonly used in locomotive and aircraft maintenance facilities, refineries and fuel loading stations, and mines (Li, 2011; Perilla et al., 2018). Infrared detectors recognize fire incidents when a characteristic flame flicker generated by fire is received. The ultraviolet detectors identify fires when a flame emit an ultraviolet radiation and is detected. Fire detection devices based on flame recognition can be to deployed safeguard large areas and they have high response due to the fact they do not have to depend on smoke or heat generated by fire combustion. However, false alarm may be produced by radiation from other hostile sources which include welding, sunlight and tungsten lamps. Research and development effort have been made to mitigate against the nuisance alarms generated by infrared and ultraviolet detectors by introducing multi-wavelength radiation sensing and computational algorithm that determine object temperature, flame temperature, surface area and presence of a flame (Liu and Kim, 2003). Optical flame detectors provide higher reliability, great long-term stability and rapid response to fire emergency compared to fire detection systems using smoke detectors. Flame detectors based on optical sensing devices operate on specific spectral ranges to record the incoming file:///C:/user/Downloads/azojete143/www.azojete.com.ng mailto:lm324fairchild@yahoo.com Hassan and Audu: Traditional Sensor-Based and Computer Vision-Based Fire Detection Systems: A Review. AZOJETE, 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818, www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 473 electromagnetic energy at the specified wavelengths which include infrared, visible and ultraviolet. Ultraviolet-only flame detection devices operate on wavelength shorter than 400nm. They detect flames at a great speed of 3-4ms because of the ultraviolet high energy radiation generated by fires and explosion incidents at the instance of their ignition. Ultraviolet and infrared detection devices compare the threshold signal in two spectral ranges and their ratio to each other to determine the reliability of the fire signal to minimize false alarms. However, commercially available ultraviolet flame detection devices are based on Geiger- Muller counter which is a quartz tube filled with an inert gas which conducts electric current when a photon of wavelength between 185 and 260 nm of a flame makes the gas to conduct temporarily (Cheong et al.,2011). This kind of flame detection system is really huge in size and expensive, operates at high voltage, has short lifetime and electronically interferes with other devices in close range. Hence, the lower voltage and compact semiconductor ultraviolet sensing device is usually the interesting device (Cheong et al., 2011). Cheong et al. (2011) presented inexpensive and low-power wireless sensor node for ultraviolet detection of flame. The device employs the state-of –the-art nanotechnology based on semiconductor benefiting the advantage of low cost, compact size and light reliability. In addition, there is no external bias needed, minimizing power consumption and extending the device lifetime. There is also no interference with other devices as in ultraviolet flame sensing devices based on the Geiger-Muller counter. 2.4 Gas Sensors Volatile chemicals are released normally in fire combustions before smoke particles are generated. Therefore, fire detection devices based on chemical gas sensors may have a faster response time compared to the devices based on smoke detectors which now dominate the market. The key hypothesis underpinning the development of fire detection based on gas sensor is the fact that gases and volatile are generated in many fire types before smoke is produced. The early release of gas creates the possibility to develop fire detection systems with shorter response time compared to method based on smoke detection (Fonollosa et al., 2016). The first step in chemical gas sensor based method for fire detection starts with sensitivity analysis of different sensing technologies to relevant combustion products, usually carbon monoxide and carbon dioxide. However, sensor sensitivity to volatile gases produced in fire incidents determined by the feasibility of chemical sensing for fire identification, gas sensing devices also demonstrate high cross-sensitivity to water vapour and to great range of volatiles which are generated during many common daily activities. As a result, chemical gas-based systems for fire detection are vulnerable to false alarms (Fonollosa, et al., 2016). Therefore, sensor design for robust fire detection devices, have to extend beyond the mere analysis of sensor sensitivity to combustion products, and the examination of cross-sensitivity to interfering scenarios is necessary too (Fonollosa et al., 2016). In order to address the drawbacks of cross-sensitivity of chemical sensing devices and provide robust fire detection systems, sensor fusion algorithms have been investigated in the literature. http://www.azojete.com.ng/ mailto:lm324fairchild@yahoo.com Arid Zone Journal of Engineering, Technology and Environment, September, 2022; Vol. 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818; www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 474 The great concern with these algorithms is the good detection of fire while minimizing nuisance effects. Variety computational algorithms have been examined such as logic rules, neural network, probabilistic neural network, hierarchical linear discriminant analysis or k- nearest neighbours. However, yet further work is required to enhance detection time and nuisance rejection of fire alarms based on chemical sensing method (Fonollosa et al., 2016). Multi-sensor approach is motivated by the need for development of more robust fire detection system devoid of false alarms. This kind of approach uses more than one fire signature to detect fire events. Because of the robustness of multi-sensor based fire detection devices, recent research efforts on fire detection is hugely focused towards developing more robust algorithms and image processing methods based on the data obtained from sensors in order to minimize false alarms. These such kind of detectors which combine different type of sensors are seen as an effective method to minimize the shortcomings of systems employing individual sensors (Sowah et al., 2014; Wang et al., 2020). Fonollosa et al. (2016) presented a multi-sensor system based on chemical sensing for fire detection. The system was integrated in a sensing chamber in which different fire types and interference were performed. The result obtained showed that the system is able to identify fire from non-fire situations, and activate the alarm faster than conventional fire detection systems based on smoke detection for some fire types. However, there is need to achieve a shorter response time and enhance the robustness of fire detection devices using chemical sensing. In a development to increase the sensitivity of fire detection systems and reduce nuisance alarms, Sowah et al. (2014) presented the design and development of a multi-sensor fire detection and alarm system based on fuzzy logic. A microcontroller was used to process data obtained from a smoke sensor, temperature sensor and flame sensor using fuzzy logic algorithm to confirm fire status. Díaz-Ramírez et al. (2012) proposed and evaluated two algorithms for forest fire detection. The algorithms are designed based on information fusion techniques. The first algorithm employs threshold method and nodes attached with temperature, humidity and light sensors. The second algorithm employs the Dempster-Shafer theory that has a basic assumption that the nodes use temperature and humidity sensors. Fire detection systems based on gas sensing are well-suited for fast fire detection since they have high sensitivity to the early release of volatiles. On the other hand, these systems have low specificity because they respond to volatile gases from non-fire sources. The use of pattern recognition algorithms can enhance the effectiveness of such gas sensor-based fire detection methods. All works reviewed on traditional sensor-based fire detection methods are provided in Table 1. file:///C:/user/Downloads/azojete143/www.azojete.com.ng mailto:lm324fairchild@yahoo.com Hassan and Audu: Traditional Sensor-Based and Computer Vision-Based Fire Detection Systems: A Review. AZOJETE, 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818, www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 475 Table 1: Summary of Traditional Sensor based Fire Detection Methods Author(s) and Year Technique Used Contribution Research Gap Li, 2011; Perilla et al., 2018 Thermal sensing Easy and inexpensive Slow response, limited detection range Liu & Kim, 2003 Thermal sensing Insensitive to noise Slow response, limited detection range Fonollosa et al., 2016; Perilla et al., 2018 Smoke sensing Ability to detect fire in its early stage Slow response, high false alarm, limited detection range Courbat et al., 2011 Smoke sensing Capability to detect fire in its early stage High response time, high false alarm, expensive, limited detection range Cheong et al., 2011 Flame sensing Inexpensive and insensitive to external interference Limited detection range Fonollosa et al., 2016 Gas sensing Faster response compared to smoke detectors Still the response time and robustness can be enhanced 3. Computer Vision Based Fire Detection Systems Advances in computing and electronics in recent years have led to the rapid development of fire detection systems based on image processing and computer vision techniques. The computer vision based method employs surveillance camera to capture video image in real time and then input it into computing system to extract the fire image features and then apply a suitable machine learning or deep learning algorithm to detect fire in order to activate the alarm device. These Computer vision based methods for fire detection can address the limitations of the conventional sensor based methods and have the advantages of faster response, accurate detection and rich descriptive information (Chen et al., 2019). These computer vision-based systems for fire detection can be classified into hand-engineered feature-based methods or deep learning based methods (Chen et al., 2019). 3.1 Hand Crafted Features Based Methods The hand-crafted feature based methods can be mainly distinguished into two classes based on the analysed features: colour based and motion based. The colour feature based methods are designed around the consideration that a fire flame which is assumed to be produced by common combustibles such as wood, plastic, paper, or others, can be uniquely characterized by its distinctive colour feature in such a way that the evaluation of the colour components in RGB (Red, Green, Blue), YUV (Luminance, Chrominance) or any other colour space is http://www.azojete.com.ng/ mailto:lm324fairchild@yahoo.com Arid Zone Journal of Engineering, Technology and Environment, September, 2022; Vol. 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818; www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 476 strongly robust to recognize the presence of flame. This basic idea motivates a number of recent methods (Foggia et al., 2015). The introduction of the HSI (Hue, Saturation, and Intensity) colour space in great ways simplifies the definition of the rule for the designer by being better for providing a user focused approach way of describing colour. The major drawback of such colour based methods is that they are in particular sensitive to the fluctuation in scene brightness, hence leading to a greater level of false positive because of the shadows present or different tonalities of red. This limitation can be overcome by switching to a YUV colour space. To sum up, it is evidently clear that the methods based on colour feature information, despite the fact that being naturally simple to design, can be best suited for sterile environment only, where there is no moving object generally inside. The major shortcomings of such methods are about the number of false positives when deployed in usually populated areas. People wearing red clothes or red cars may be wrongly recognized as fire feature due to their dominant colour. Several studies have been proposed in the literature to address this problem (Foggia, et al., 2015). All these studies begin with the assumption that a fire flame changes its shapes in continuous manner and the unstructured motion of red regions can make it possible to discriminate it from moving rigid object in the scene. For example, the physical features of the fire are employed to design a feature vector based on improved optical flow algorithm, which have capability to analyse in variety of ways both the temporal features of fire and its saturated flame. Dynamic features have been employed in a study, where a two-phase texture detection process have been proposed in the literature to enhance the speed of the segmentation phase, extremely useful to extract a broad class of shape-based features, making the detection of the fire in a good time possible. The wavelet transform has also been employed in some studies to segment accurately the dynamic behaviour of flame regions. It is observed that the methods using wavelet transform, compared to colour based methods, cannot be used on still images and generally require extremely high frame per second to ensure reasonable results, hence limiting their applicability. However, the major limitation of methods based on motion is that the performance enhancement is most of the times paid from different perspective. Several sensitive parameters need to be rightly configured for the application at hand in most cases. Secondly, the flame motion and shape depend to some extent on the burning material as well as on the geographic conditions (Foggia et al., 2015). In recent years, several methods have been proposed based on computer vision algorithms employing hand engineered features to achieving robust fire detection using video images captured by surveillance cameras by combining the static and dynamic features of fire. For instance, (Foggia et al., 2015) proposed a fire detection using employing a combination of expert based on colour, shape and flame motion information. The method has been tested on a broad database to evaluate its performance in terms of sensitivity and specificity. Experiment results obtained show that the effectiveness of the approach. file:///C:/user/Downloads/azojete143/www.azojete.com.ng mailto:lm324fairchild@yahoo.com Hassan and Audu: Traditional Sensor-Based and Computer Vision-Based Fire Detection Systems: A Review. AZOJETE, 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818, www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 477 An algorithm for forest fire smoke detection is proposed by Cai et al. (2016). The algorithm consists of three aspects: moving target segmentation, feature extraction and classification. Firstly, Visual Background Extractor (ViBE) is used for motion segmentation. Morphological operation is performed to connect concentrated but not linked regions. Then, the static and dynamic features of smoke are obtained based on colour histogram, energies of wavelet sub images, compactness and direction of moving target. Finally, the feature vector is classified using support vector machine. The algorithm has an accuracy rate of 92.7% but still the accuracy needs to be improved and tested on sufficient number of smoke video dataset to enhance its performance. Shuai et al. (2016) proposed an algorithm for smoke detection in remote surveillance video based on fast self-tuning background subtraction method and judgement of smoke static and dynamic features. The proposed algorithm has two steps. First, extracting the moving smoke area using the fast self-tuning background subtraction algorithm proposed, which achieves better results compared to colour based and motion based segmentation approaches. Then, colour, shape and variance feature of smoke are employed to eliminate smoke-like features to guarantee the accuracy of the detection. The results obtained show that the algorithm could detect smoke regions in remote video surveillance in real-time, precise and reliable manner and it is robust to environmental change. However, the algorithm is not tested on large video dataset. Shrivastava and Matlani (2016) proposed an efficient computer vision based algorithm for smoke detection using background subtraction and K-means segmentation. The algorithm is comprised of five steps: (i) Smoke video image is read (ii) The RGB colour space is transformed to L*a*b colour space to classify pixels as smoke or not. Spatial clustering and analysis is carried out on smoke colour pixels. (iii) Frame differencing method is employed to extract background from current frame to detect moving pixels. (iv). K-means method is used to segment colour features into three clusters based on Euclidean distance metric. (v) The results obtained from background subtraction and K-means segmentation are combined for final judgement. The result obtained demonstrates that the algorithm has a better performance compared to fuzzy c-means method. Nevertheless, the performance of the algorithm can be improved. Wang et al. (2016) proposed an algorithm for fire smoke detection using optical flow approach and texture feature. The proposed algorithm consists of the following steps. (i) Firstly, bilateral filtering is employed to extract noises from poor video quality in image frames. Then, foreground regions are segmented by using Gauss mixture model and background subtraction method to remove static interference. (ii) Colour model used to segment region of interest. Median filter employed to remove isolated points and morphological processing to remove small isolated region to obtain a continuous target smoke area. (iii) Smoke texture features segmented using local binary pattern (LBP) and local binary pattern variance (LBPV) in region of interest. (iv) Motion feature in region of interest contour is extracted using optical flow approach based on image pyramid. (v) Finally, support vector machine is used to identify smoke based on the optical flow motion feature and LBP and LBPV texture feature. Simulation http://www.azojete.com.ng/ mailto:lm324fairchild@yahoo.com Arid Zone Journal of Engineering, Technology and Environment, September, 2022; Vol. 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818; www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 478 results reported illustrate that the algorithm is timely and accurate in fire smoke detection and robust to external environmental changes. The algorithm has also less computational complexity, fast computational speed but it is tested only five videos which are not sufficient enough. A computer vision based smoke detection approach for intelligent surveillance system is proposed by Yuanbin (2016). The algorithm consists of the following steps. (i) Fuzzy logic is used to enhance image. (ii)Target smoke area is segmented using Gaussian Mixture Model (GMM). (iii) Static and dynamic features of smoke are extracted and fused. (vi) The segmented features are normalized and input into support vector machine algorithm for smoke classification. The algorithm is simple and achieves high detection rate and real-time performance. However, the algorithm could generate false alarm in a complex video environment and the processing speed need to be improved for deployment on embedded systems. Gunawaardena et al. (2016) proposed a computer vision based algorithm for fire detection that can be compatible with the existing traditional sensor based devices is proposed to enhance the performance and robustness of the conventional system. The proposed video based fire alarming system consists of five steps. The system employs adaptive background subtraction to segment foreground moving region. The rule-based fire colour model is utilized to confirm whether the extracted foreground object is a fire colour feature. The fire pixel recognition is modelled using YCbCr colour space. Morphological operation is performed to remove isolated blocks from the segmented fire target regions and enhance the features by growing the missing pixels. The dynamic feature of the extracted target fire region is analysed in the temporal domain to extract the flame flicker to remove spurious fire-like colour feature or non-fire like objects. All these features are fused to form the fire detection system. An algorithm is developed to segment flame from fire video images. The proposed algorithm is computationally inexpensive, simple and can be integrated into existing security monitoring systems. It detects fire and flame successfully; however, it is affected by the scene illumination and tested on small dataset. Jinlan et al. (2016) presented a method of fire and smoke detection using Surendra background and grey bitmap plane algorithm. Firstly, the dynamic background is modelled using Otsu adaptive threshold and Surendra background. Then, the moving target region is extracted by the continuous cumulative moving average of moving target region and the morphological processing algorithm. The smoke region is extracted using colour feature. Finally, the fire region is segmented using OHTA colour space and grey bitmap plane algorithm. The results obtained show that the proposed method can successfully recognize fire in moving vehicle scene, red plastic bag waving scene, red balloon fluttering scene, moving vehicle similar to fire colour. However, the algorithm is not compared with other recent methods and also not tested on sufficient dataset. file:///C:/user/Downloads/azojete143/www.azojete.com.ng mailto:lm324fairchild@yahoo.com Hassan and Audu: Traditional Sensor-Based and Computer Vision-Based Fire Detection Systems: A Review. AZOJETE, 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818, www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 479 An algorithm for real-time fire detection based on AdaBoost, local binary pattern(LBP) and convolutional neural network in video sequence is proposed by Maksymiv et al. (2017). The proposed method has two main parts. First, region of interest (ROI) is extracted using AdaBoost and LBP algorithms. Combination of AdaBoost and LBP is fast but generates high false positive number. Convolutional neural network (CNN) architecture is employed to address this problem so as to make the algorithm better and reliable. Lastly, two support vector machines (SVM) classifiers provide a judgment on presence of fire or smoke in video sequences. The method has a reasonable execution time which means that it can perform in real-time at video rate. The experimental results obtained demonstrate that the method can achieve more than 95% detection rate. Nevertheless, this method has high false positive number in smoke and fog detection. Wu et al. (2017) proposed a fire and smoke detection method based on adaptive threshold of feature and deep convolutional neural network in video images. The ViBE approach is employed to segment the background from the video image sequence and update the exact moving regions using frame-by-frame differences method. Dynamic and static features extracted are fused to detect fire and smoke region of interest. Deep learning is employed to detect static fire and smoke features based on Caffe model. Experiments illustrate that the proposed method achieves efficient fire and smoke detection. But the method is not tested on large dataset and deep learning models are generally computationally expensive. An algorithm for smoke detection based on simplified descriptors of video information is proposed by Monte et al. (2017). Firstly, the descriptors of the scene on every frame are obtained. The inference processes of smoke presence are input by the descriptors and background estimation, which is updated with every frame. A group of classifiers perform the smoke inference process. The fusion outcomes of these classifiers provide a reliable detection. Simple Boolean logic combines the process outcomes, with minimum and maximum threshold values. The proposed approach is simple and efficient for deployment on computationally low embedded applications for real-time surveillance. Nevertheless, the algorithm does not take into consideration the variability of illumination and uncertainty of the environment. Pritam and Dewan (2017) proposed fire detection system based on LUV colour space and hybrid transforms. Video captured are read and frame are segmented. All frames are analysed to confirm the presence of fire. Fire is detected by the presence of fire pixels in the image. Fire pixels are recognized using the features of LUV colour space and hybrid transforms. Edge detection algorithm is employed to extract the edges of the image when flame colour detects existence of fire. Image segmentation methods are employed to extract the fire region of interest from the background. The results reported show that the proposed system can achieve better performance. However, the system is not tested and compared with other approaches. Vijayalakshmi and Muruganand (2017) proposed an effective smoke detection approach in video images for early fire emergency systems. The proposed algorithm uses fuzzy c-means clustering and background subtraction to segment smoke region of interest from moving http://www.azojete.com.ng/ mailto:lm324fairchild@yahoo.com Arid Zone Journal of Engineering, Technology and Environment, September, 2022; Vol. 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818; www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 480 regions. Result obtained show that the algorithm outperforms other algorithms and increase the accuracy smoke detection. However, the algorithm is not tested on large variety of dataset and the performance still can be improved because it is only based on motion segmentation. Zhang et al. (2017) developed a method for smoke detection in video based dissipation function and ViBE. The proposed algorithm employs two key features of smoke to detect it. Visual background extractor (ViBE) is employed to segment moving objects in static scene. The atmospheric model is used to detect semi-transparent feature of smoke. The similarity between extracted regions and original background regions in the same position is computed using normalized cross correlation and the value of the original background is chosen from the frame that is nearest to the current frame. The blocks with high similarity are identified as smoke parts. Experiment results demonstrate that the algorithm can recognize fire and integrated into surveillance system. But the algorithm is not tested on sufficient dataset. Huang et al. (2017) proposed an algorithm for video based smoke detection in low illumination indoor environment using support vector machine. The smoke region is extracted and then the area diffusion and main direction of movement of smoke region are selected as the dynamic features and texture is extracted as the static feature to recognize fire smoke in low- light indoor scene. Support vector machine is employed to design the classifier to realize smoke detection. Finally, the algorithm is tested under variety of conditions. The results obtained show the algorithm achieves high adaptability, high detection rate and good noise immunity. But the algorithm performs in only indoor environment and it is not tested on sufficient dataset. An effective method for real-time fire detection based on video processing is proposed by Khan et al. (2018). The proposed method employs flame colour and spatial-temporal features to detect fire regions. Firstly, fire colour pixels are segmented using a set of improved rules based on RGB colour space. The foreground is segmented based on change in blue component trained using neural network. Lastly, the region of interest is extracted using growing area analysis. Experimental results illustrate that the proposed method outperforms other state of the art approaches and yields 97.7% detection accuracy. Nevertheless, the method is sensitive to variation in illumination and based on shallow features. Yang et al. (2018) proposed a technique for video based smoke detection by combining Gaussian Mixture Model (GMM) and HSV colour model with the deep convolutional neural network. First, each frame is divided into 10-by-10 sub-images by the method. Secondly, Gaussian Mixture Model is employed to extract the motion region and later the sub-images block which contains the moving region is segmented. Then, the motion blocks segmented by the GMM is used to extract target smoke blocks successively using HSV colour feature analysis. Finally, the method confirms whether the bounding box detected by trained model is combined with target smoke block, or the bounding box does not intersect with one of the target smoke blocks, the bounding box will be segmented. The method has improved the detection accuracy and significantly reduced the false detection rate. However, the method is not tested on large dataset, and still the detection accuracy can be enhanced. file:///C:/user/Downloads/azojete143/www.azojete.com.ng mailto:lm324fairchild@yahoo.com Hassan and Audu: Traditional Sensor-Based and Computer Vision-Based Fire Detection Systems: A Review. AZOJETE, 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818, www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 481 Zhao et al. (2018) proposed an effective method for smoke detection based on automatic vision system. The method employs an adaptive background subtraction technique to segment moving regions. Non-smoke regions are extracted out by employing colour feature and fuzzification. Regions that satisfy both the colour analysis and fuzzification criteria are classified as candidate smoke regions. The detected smoke regions are combined with symbolic objects around to confirm the nature of the smoke source using contextual object detection. The method precisely connects specific scenes with spatial and temporal domain statistical approaches to identify target objects. However, the method cannot perform well in complex video scenes. Kaabi et al. (2018) proposed a novel approach for video based smoke detection based on machine learning approach to overcome forest wildfires. Firstly, the video is divided into frames for training with deep belief network. In addition, Gaussian Mixture Model is employed to segment candidate regions from the target smoke video to be trained by using Deep Belief Network. Logistic regression is used to classify on the output data. The approach is reported to achieve 95% detection rate. However, larger frame size can make the method to diverge and it is computationally expensive which makes it not feasible to be deployed on embedded systems. A robust algorithm based on information on colour and shape was proposed by Rabiu (2022). The algorithm starts with loading video image database developed to detect presence or absence of fire in video frames. Background subtraction is used by this method to compare the current frame with the reference frame. When the outcome of the background subtraction is less than the set point, the frame is discarded and the next frame is checked. But if the difference is equal or greater than the set point, the frame is then subjected to colour and shape test. This is achieved by employing combined RGB colour space model and shape feature. The proposed algorithm was extremely robust in detecting fire compared to those algorithms based on colour or motion features. 3.2 Deep Learning Based Methods The traditional hand engineered feature based methods do look to be great but do not seem to be the ideal choice to employ on image based tasks. Hence, these days deep learning based methods have achieved superior performance on computer vision tasks. In addition, deep learning based methods have a variety of applications such as object recognition and classification in videos, images and these days is also employed for real-time detection of activities, speech recognition and natural language processing. Current research on fire detection based on computer vision have proposed many approaches using deep neural networks such as CNN have achieved great results. Therefore, studies on fire detection using CNNs have been proposed by some researchers in order to improve the performance of the existing fire detection systems (Majid et al., 2022). For example, in Vinicius et al. (2021) a robust fire detection system based on two-dimensional convolutional neural network where an object tracker is integrated into an object detector with the aim of confirming if detected events show fire or smoke behaviour over time. http://www.azojete.com.ng/ mailto:lm324fairchild@yahoo.com Arid Zone Journal of Engineering, Technology and Environment, September, 2022; Vol. 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818; www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 482 In the work of Saeed et al. (2020), a hybrid approach was presented that consists of an Adaboost MLP neural network model to forecast fire events. Subsequently, an Adaboost LBP model was developed to extract the region of interest (ROIs) and finally a convolutional neural network for fire detection with videos and images captured from installed surveillance camera. Majid et al. (2022) presented an attention based convolutional neural network model for fire detection using real life images. An attention mechanism was also added to the model which achieved great improvement in the performance of data. The introduced neural network has demonstrated significantly great performance on the testing dataset having minimal false positive level. In Aktas et al.(2019), the existing CNN based methods for fire detection in video sequences was extended by integrating Multiple Instance Learning (MIL). MIL ease the need for having accurate locations of fire patches in video frames, which are needed for patch level CNN training. Only frame level labels showing the presence of fire somewhere in a video frame are required instead. Thus, reducing substantially the annotation and training efforts. The proposed approach was tested on a new fire datasets developed by extending some of the recently used fire datasets with video sequences obtained from the web. Result obtained illustrated that the presented method enhances fire detection performance. Muhammad et al. (2018) proposed a cost-effective CNN architecture for fire detection in surveillance videos. The model is inspired from Google Net architecture which is reasonably less computationally complex and suited for the intended task compared to other networks that have high computational cost such as Alex Net. Taking into consideration the nature of the target task and fire data, the model is fine tuned to achieve balance between efficiency and accuracy. Results obtained using benchmark fire datasets showed that the proposed model was effective and suited for fire detection based video images captured by cameras from surveillance systems in comparison to the state-of–the art methods. In Xie et al. (2020), a method was proposed that employs the motion-flicker based dynamic features and deep static features for video based fire detection. First, the dynamic features are segmented by, analysing the differences in motion and flicker features between fire and non- fire objects in video scenes. Second, an adaptive lightweight convolutional neural network is proposed to segment the deep static features of fire. Finally, the dynamic and static features of fire are cascaded to develop a video based fire detection system with enhanced efficiency in terms of accuracy and run time. Zhang et al. (2021) developed an effective asymmetric encoder-decoder U-shape architecture based on Squeeze Net, Attention U-Net and Squeeze Net performs most of the time as an extractor and a discriminator of forest fire. The model uses attention mechanism to extract useful features and suppress non-related features by embedding Attention Gate (AG) units in the skip connection of U-shape structure. Key features are highlighted through this way so that the proposed method may be great at forest fire extraction problems with a minimum set of parameters. file:///C:/user/Downloads/azojete143/www.azojete.com.ng mailto:lm324fairchild@yahoo.com Hassan and Audu: Traditional Sensor-Based and Computer Vision-Based Fire Detection Systems: A Review. AZOJETE, 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818, www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 483 In the work of Mukhopadhyay et al. (2019), a model was developed that has the capability of forecasting fire event with a reasonably great accuracy. The model is based on the convolutional neural network architecture, Mobile Net. The developed model can be deployed on embedded computing devices in an easy manner due to the fact that it has small size and hence can also be employed as a robust stand-alone fire detection system. Hao et al. (2019) proposed an algorithm based on convolutional neural network for video based flame detection. First, the video image is extracted RGB-HIS colour synthesis model and enhanced adaptive mixture Gaussian model. Then, the classifier is designed by learning a great number of fire dataset using the CNN designed. Then, the flame region of interest is extracted by the classifier criteria. The result obtained revealed that the proposed algorithm has high detection accuracy. Tao et al. (2016) proposed an innovative method based convolutional neural network that can be trained end to end from raw image pixel values to classifier outputs and segment automatically features in a way to avoid the tedious image pre-processing stage of the hand- engineered feature based methods. Experiment results obtained showed the proposed method has high detection accuracy rates with low false alarm rates on the small dataset that clearly achieves greater performance compared to the existing methods. In Yang Wang et al. (2019), presented a method for video based smoke detection that combines conventional smoke detection and a lightweight convolutional neural network. First, the smoke YUV colour space information is fused to minimize the smoke region of interest based on the VIBE algorithm. Secondly, the lightweight convolutional neural network is designed to segment the information on smoke image in an automatic manner. The training speed of the model is enhanced by using the transfer learning approach for deep learning. The proposed method achieves the real-time performance of smoke detection, guarantees the accuracy of smoke detection and minimizes the false positive level. Son et al. (2018) proposed a method for fire detection based on deep learning. The deep learning networks employed are Alex Net, Google Net and VGG-16 in three manners. Image input that were captured by surveillance camera is classified into three different categories: normal, smoke and flame. Then, the network is trained to detect each corresponding category. The results obtained illustrate that all the three network models were able to recognize fire at over ninety percent detection accuracy. In the work of Zeng et al. (2018), an enhanced object detection method based on deep convolutional neural network for fire detection was proposed. Firstly, the feature extractor is substituted in various neural network object detectors for faster R-CNN, Single Shot MultiBox Detector (SSD), Region based fully Convolutional Networks (R-FCN). Secondly, the MSCOCO dataset was used to optimize the parameters of the object algorithm. Lastly, the smoke detection dataset was used to perform the experiments. The proposed algorithm has http://www.azojete.com.ng/ mailto:lm324fairchild@yahoo.com Arid Zone Journal of Engineering, Technology and Environment, September, 2022; Vol. 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818; www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 484 achieved great results in terms of accuracy and computational speed compared with the existing smoke detection methods. Shi et al. (2019) developed a method that combines dark channel image input and a relative precise CNN. The difference between the smoke and background could be greatly enhanced by the dark channel of an image. The comparatively concise CNN could be efficiently trained on an insufficient dataset. Results from the experiments performed show extensively that the proposed method achieved great performance when compared to other smoke detection methods. In the work of Sadewa et al. (2019), an image base fire detection system was developed using CNN. The proposed system achieved an accuracy level of over ninety percent. Table 2 provides a summary of works reviewed on computer vision based fire detection. Table 2. Summary of Works reviewed on computer vision based fire detections. Author(s) and Year Algorithm Dataset Contribution Research gap Foggia et al. (2015). Hand crafted features based on motion, colour and shape fire video Robust detection High false alarm Cai et al. (2016) Static and dynamic features based on colour histogram, energies of wavelet sub images, compactness and direction of moving target Smoke images 92.7% detection rate High false alarm Shuai et al. (2016) Static and dynamic features based on motion, colour and shape Smoke video Real-time and reliable detection Not tested on sufficient dataset Shrivastava and Matlani (2016) Background subtraction and K-means clustering Smoke video Good detection rate High false alarm file:///C:/user/Downloads/azojete143/www.azojete.com.ng mailto:lm324fairchild@yahoo.com Hassan and Audu: Traditional Sensor-Based and Computer Vision-Based Fire Detection Systems: A Review. AZOJETE, 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818, www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 485 Wang et al. (2016) Optical flow motion and texture feature Smoke video Timely and accurate detection Tested on small dataset Yuanbin, (2016) Static and dynamic features Smoke video Simple and real- time detection approach High false alarm Gunawaardena et al. (2016) Motion and colour features, adaptive thresholding Fire video Simple and computationally inexpensive Not robust to environmental changes Jinlan et al. (2016) Colour and motion features Smoke video Successful detection Not tested on large dataset Maksymiv et al. (2017) Local binary pattern and convolutional neural network Fire video Real time fire detection High false alarm Wu et al. (2017) Adaptive thresholding and deep CNN Smoke and fire video Effective detection Computationally expensive Monte et al. (2017) Background estimation Smoke video Simple and efficient Not robust enough Pritam and Dewan (2017) LUV colour space and hybrid transform Fire video Good detection rate High false alarm Vijayalakshmi and Muruganand (2017) Fuzzy c-means clustering and background subtraction Smoke video Accurate detection High false alarm Zhang et al. (2017) Visual background extractor and dissipation function Smoke video Good detection Not tested on large dataset Huang et al. (2017) Support vector machine Smoke video High detection rate Performs only in door environment http://www.azojete.com.ng/ mailto:lm324fairchild@yahoo.com Arid Zone Journal of Engineering, Technology and Environment, September, 2022; Vol. 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818; www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 486 Khan et al. (2018) Colour and spatial temporal features Fire video Good detection rate Sensitive to illumination variation Yang et al. (2018) Gaussian mixture model, HSV colour model and deep CNN Smoke video Reasonable detection rate Not tested on large dataset Zhao et al. (2018) Adaptive background subtraction smoke Good detection rate Not robust to environmental changes Kaabi et al. (2018) Machine learning Smoke video good detection rate Computationally expensive Rabiu (2022) Colour and shape information Fire video Robust and reliable High false alarm Saeed et al. (2020) Adaboost MLP neural network model Forest fire Good detection Computationally expensive Majid et al. (2022) Convolutional neural network Fire video Good detection Computationally expensive Aktas et al. (2019) Convolutional neural network Fire video Good detection Computationally expensive Muhammad et al. (2018) Convolutional neural network Fire video Effective detection Computationally expensive Xie et al. (2020) Convolutional neural network Fire video Effective detection High computational cost Zhang et al. (2021) Convolutional neural network Forest fire Effective detection Not tested on large dataset Mukhopadhyay et al. (2019) Convolutional neural network Fire video Good detection Not tested on large dataset Hao et al. (2019) Convolutional neural network Fire video High detection rate High computational cost file:///C:/user/Downloads/azojete143/www.azojete.com.ng mailto:lm324fairchild@yahoo.com Hassan and Audu: Traditional Sensor-Based and Computer Vision-Based Fire Detection Systems: A Review. AZOJETE, 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818, www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 487 Tao et al. (2016) Convolutional neural network Smoke video High detection rate High computational cost Yang Wang et al.( 2019) Convolutional neural network Smoke video Good detection Not tested on large dataset Son et al. (2018) Convolutional neural network Fire video High detection rate Computationally expensive Zeng et al. (2018), Convolutional neural network Fire video High detection rate High computational cost Shi et al. (2019) Convolutional neural network Smoke image High detection rate Computationally inexpensive Sadewa et al. (2019) Convolutional neural network Fire image High detection rate Computationally inexpensive 4. Conclusion In this review paper, a survey of various available and proposed fire detection systems was performed. The two categories of fire detection systems were reviewed. The first category is the traditional sensor based methods which involve measuring the ultraviolet or infrared radiation, heat, gas or particulate emissions produced by fire combustion. The methods that belong to this category include thermal detectors, smoke detectors, flame detectors and gas sensors. These methods have a number of shortcomings which include transport delay, conduction delay, limited detection range, high false positive and do not provide rich information. In the second category, the computer vision based methods employ image processing and artificial intelligence techniques to detect fire in video images captured by surveillance systems. These systems can be broadly classified into handcrafted feature based and deep learning based methods. The hand crafted feature based methods tend to be time consuming and affected by environmental conditions while the deep learning based methods learn and extract dep features automatically which eliminate the need for hand feature engineering but they are limited by their high computational cost. Based on this analysis, it can be concluded that each method has some merits and shortcomings. For example, the sensor based methods are suitable for indoor application but not appropriate for outdoor environment. It also suffer from transport delay and high false alarm. The computer vision based methods using hand crafted features are affected by the complexity of scene and environmental factors. This problem is addressed by the deep learning based methods but such methods are, however, computationally expensive which restricts their applicability for deployment on real-time embedded computing systems. We observe that the recent advances in vision-based fire detection are driven by CNN. This development can be due to the fact CNN achieves superior performance in various computer vision tasks. But the trade-off is between accuracy, model size and computational speed. http://www.azojete.com.ng/ mailto:lm324fairchild@yahoo.com Arid Zone Journal of Engineering, Technology and Environment, September, 2022; Vol. 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818; www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 488 Finally, it is evidently clear that great amount of academic research efforts has been invested in computer vision-based approaches for fire detection to address the limitations of the sensor-based systems. But many of these methods have not been implemented on hardware platforms such as FPGA. Computational complexity and high hardware requirement are the major shortcomings of the computer-based approaches. Therefore, more research work is required to design, develop, and implement such systems on hardware targets in a computationally efficient manner. References Aktas, M., Bayramcavus, A. and Akgun, T. 2019. Multiple instance learning for CNN based fire detection and localization.16th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS) held in Taipei, Taiwan, 18 - 21 September 2019. pp 1 - 8. Asih, LC., Sthevanie, F. and Ramadhani, K N. 2018. Visual Based Fire Detection System using Speeded Up Robust Feature and Support Vector Machine. 6th IEEE International Conference on Information and Communication Technology (ICoICT) held in Bandung, Indonesia, 03 - 05 May 2018.pp 485 - 488. Cai, M., Lu, X., Wu, X. and Feng, Y. 2016. Intelligent Video Analysis-based Forest Fires Smoke Detection Algorithms. 12th IEEE International Conference on Natural Computation, Fuzzy Systems and Knowledge Discovery (ICNC-FSKD) held in Changsha, China, 13 - 15 August 2016. pp 1504 - 1508. Chen, K., Cheng, Y., Bai, H., Mou, C. and Zhang, Y. 2019. Research on Image Fire Detection Based on Support Vector Machine. 9th International Conference on Fire Science and Fire Protection Engineering (ICFSFPE) held in Chengdu, China, 18 - 20 October 2019. pp 1 -7. Cheong, P., Chang, KF., Lai, YH., Ho, SK., Sou, IK. and Tam, KW. 2011. A ZigBee-based wireless sensor network node for ultraviolet detection of flame. IEEE Transactions on Industrial Electronics, 58(11): 5271–5277. Courbat, J., Pascu, M., Gutmacher, D., Briand, D., Wöllenstein, J., Hoefer, U., Severin, K. and De Rooij, NF. 2011. A colorimetric CO sensor for fire detection. Procedia Engineering, 25:1329–1332. Díaz-Ramírez, A., Tafoya, LA., Atempa, JA. and Mejía-Alvarez, P. 2012. Wireless Sensor Networks and Fusion Information Methods for Forest Fire Detection. Procedia Technology, 3(4322): 69–79. Filonenko, A. and Danilo, C. 2018. Fast Smoke Detection for video Surveillance Using CUDA. IEEE Transactions on Industrial Informatics, 14(2): 725–733. Foggia, P., Saggese, A. and Vento, M. 2015. Real-Time Fire Detection for Video-Surveillance Applications Using a Combination of Experts Based on Colour, Shape, and Motion. IEEE transactions on Circuits and Systems for Video Technology, 25(9): 1545–1556. Fonollosa, J., Solórzano, A., Jiménez-Soto, JM., Oller-Moreno, S. and Marco, S. 2016. Smart Chemical System for Reliable Fire Detection. Procedia Engineering, 168: 444–447. Gunawaardena, AE., Ruwanthika, RMM. and Jayasekara, AGBP. 2016. Computer vision based file:///C:/user/Downloads/azojete143/www.azojete.com.ng mailto:lm324fairchild@yahoo.com Hassan and Audu: Traditional Sensor-Based and Computer Vision-Based Fire Detection Systems: A Review. AZOJETE, 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818, www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 489 fire alarming system. IEEE Moratuwa Engineering Research Conference (MERCon) held in Moratuwa, Sri Lanka, 05 - 06 April 2016. pp. 325–330. Hao, X., Linhu, C. and Weixin, H. 2019. Video Flame Detection Using Convolutional Neural Networks. 4th IEE International Conference on Image, Vision, and Computing (ICIVC) held in Xiamen, China, 05 - 07 July 2019. pp 539–543. Hoefer, U. and Guttmacher, D. 2012. Fire gas detection. Procedia Engineering, 47:1446–1459. Hossain, FMA., Zhang, Y., Yuan, C. and Su, CY. 2019. Wildfire flame and smoke detection using static image features and artificial neural network. 1st International Conference on Industrial Artificial Intelligence (IAI) held in Shenyang, China, 25 - 27 July 2019. pp 1–6. Huang, M., Wang, Y. and Hu, Y. 2017. Smoke Identification of Low-light Indoor Video Based on Support Vector Machine. 3rd IEEE International Conference on Computer and Communications (ICCC) held in Chengdu, China, 13 - 16 December 2017. pp 2045 - 2049. Islam, T., Rahman, HA. and Syrus, MA. 2015. Fire detection system with indoor localization using ZigBee based wireless sensor network. 4th IEEE International Conference on Informatics, Electronics and Vision (ICIEV) held in Fukuoka, Japan, 15 - 18 June 2015. pp 1 - 6. Jinlan, L., Lin, W., Ruliang, Z., Chengquan H. and Yan, R. 2016. A Method of Fire and Smoke Detection Based on Surendra Background and Gray Bitmap Plane Algorithm. 8th IEEE International Conference on Information Technology in Medicine and Education (ITME) held in Fuzhou, China, 23 - 25 December 2016. pp. 370 - 374. Kaabi, R., Bouchouicha, M., Fnaiech, F. and Moreau, E. 2017. State of the art of smoke detection in visible and IR range. IEEE International Conference on Smart, Monitored and Controlled Cities (SM2C) held in Sfax, Tunisia, 17- 19 February 2017. pp. 81– 86. Kaabi, R., Sayadi, M., Bouchouicha, M., Fnaiech, F. and Moreau, E. 2018. Early Smoke Detection of forest wildfire video using Deep Belief Network. 4th IEEE International Conference on Advanced Technologies for Signal and Image Processing (ATSIP) held in Sousse, Tunisia, 21 - 24 March 2018. pp. 1–6. Khan, MNA., Tanveer, T., Khurshid, K., Zaki, H. and Zaidi, SSI. 2019. Fire detection system using raspberry pi. IEEE International Conference on Information Science and Communication Technology (ICISCT) held in Karachi, Pakistan, 09 - 10 March 2019. pp. 1–6. Khan, R., Uddin, J. and Corraya, S. 2018. Real-Time Fire Detection Using Enhanced Colour Segmentation and Novel Foreground Extraction. Proceedings of 4th International Conference on Advances in Electrical Engineering (ICAEE) held in Dhaka, Bangladesh, 28 - 30 September 2017. Krüll, W., Tobera, R., Willms, I., Essen, H. and Von Wahl, N. 2012. Early forest fire detection and verification using optical smoke, gas and microwave sensors. Procedia Engineering, 45:584–594. Li, Q. 2011. Estimation of fire detection time. Procedia Engineering, 11: 233–241. Liu, Z. and Kim, AK. 2003. Review of recent developments in fire detection technologies. http://www.azojete.com.ng/ mailto:lm324fairchild@yahoo.com Arid Zone Journal of Engineering, Technology and Environment, September, 2022; Vol. 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818; www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 490 Journal of Fire Protection Engineering, 13(2): 129–151. Majid, S., Alenezi, F., Masood, S., Ahmad, M., Gündüz, ES. and Polat, K. 2022. Attention based CNN model for fire detection and localization in real-world images. Expert Systems with Applications, 189: 116114. Maksymiv, O., Rak, T. and Peleshko, D. 2017. Real-time Fire Detection Method Combining AdaBoost, LBP and Convolutional Neural Network in Video Sequence. 14th IEEE International Conference on The Experience of Designing and Application of CAD Systems in Microelectronics (CADSM) held in Lviv, Ukraine, 21 - 25 February. pp. 351–353. Monte, G., Pastore, JI., Ballarin, V. and Liscovsky, P. 2017. Smoke Detection Using Simplified Descriptors of Video Information. IEEE International Conference on Industrial Technology (ICIT) held in Toronto, Canada, 22 - 25 March 2017. pp. 1070 - 1075. Muhammad, K., Ahmad, J., Mehmood, I., Rho, S. and Baik, SW. 2018. Convolutional Neural Networks Based Fire Detection in Surveillance Videos. IEEE Access, 6: 18174–18183. Mukhopadhyay, D., Iyer, R., Kadam, S. and Koli, R. 2019. PGA Deployable Fire Detection Model for Real- Time Video Surveillance Systems Using Convolutional Neural Networks. IEEE Global Conference for Advancement in Technology (GCAT) held in Bangalore, India, 18 - 20 October 2019. pp. 1 -7. Perilla, FS., Villanueva, GR., Cacanindin, NM. and Palaoag, TD. 2018. Fire safety and alert system using Arduino sensors with IoT integration. Proceedings of the 7th ACM International Conference on Software and Computer Applications held in Kuantan, Malaysia, 8 - 10 February. pp. 199–203. Pritam, D. and Dewan, JH. 2017. Detection of fire using image processing techniques with LUV colour space. 2nd IEEE International Conference for Convergence in Technology (I2CT) held in Mumbai, India, 07 - 09 April 2017.pp. 1158–1162. Rabiu, H. 2022. Robust Fire Detection Algorithm using Colour and Shape Information. Bayero Journal of Engineering and Technology (BJET), 17(1):22–29. Russo, AU., Deb, K., Tista, SC. and Islam, A. 2018. Smoke Detection Method Based on LBP and SVM from Surveillance Camera. IEEE international Conference on Computer, Communication, Chemical, Material, and Electronic Engineering (IC4ME2) held in Rajshahi, Bangladesh, 08 - 09 February 2018. pp. 1 - 4. Sadewa, RP., Irawan, B. and Setianingsih, C. 2019. Fire Detection Using Image Processing Techniques with Convolutional Neural Networks. IEEE International Seminar on Research of Information Technology and Intelligent Systems (ISRITI) held in Yogyakarta, Indonesia, 05 - 06 December 2019. pp.290–295. Saeed, F., Paul, A., Karthigaikumar, P. and Nayyar, A. 2020. Convolutional neural network based early fire detection. Multimedia Tools and Applications, 79(13–14): 9083–9099. Santamaria-guerrero, D., Mejia-flores, K., Castro-madrid, LC., Toscano-medina, R., Prudente- tixteco, L., Olivares-mercado, J., Toscano-medina, K. and Sanchez-perez, G. 2018. Smoke file:///C:/user/Downloads/azojete143/www.azojete.com.ng mailto:lm324fairchild@yahoo.com Hassan and Audu: Traditional Sensor-Based and Computer Vision-Based Fire Detection Systems: A Review. AZOJETE, 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818, www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 491 Detection in Video using a Raspberry Pi. IEEE International Conference on Engineering Veracruz (ICEV) held in Boca del Rio, Mexico, 14 - 17 October 2019.pp. 1 - 6. Shi, X., Lu, N. and Cui, Z. 2019. Smoke Detection Based on Dark Channel and Convolutional Neural Networks. 5th IEEE International Conference on Big Data and Information Analytics (BigDIA) held in Kunming, China, 08 - 10 July 2019. pp. 23–28. Shrivastava, M. and Matlani, P. 2016. A Smoke Detection Algorithm based on K-means Segmentation. IEEE International Conference on Audio, Language, and Image Processing (ICALIP) held in Shanghai, China, 11 - 12 July 2016. pp. 301 - 305. Shuai, L., Bo, W., Ranran, D., Zhiqiang, Z. and Sun, L. 2016. A novel smoke detection algorithm based on Fast Self-tuning background subtraction. IEEE Chinese Control and Decision Conference (CCDC) held in Yinchuan, China, 28 - 30 May 2016. pp. 3539–3543. Singh, PK. and Sharma, A. 2017. An insight to forest fire detection techniques using wireless sensor networks. 4th IEEE International Conference on Signal Processing, Computing and Control (ISPCC) held in Solan, India, 21- 23 September 2017. pp. 647–653. Solórzano, A., Eichmann, J., Fernández, L., Ziems, B., Jiménez-Soto, JM., Marco, S. and Fonollosa, J. 2022. Early fire detection based on gas sensor arrays: Multivariate calibration and validation. Sensors and Actuators B: Chemical, 352(1). Son, G-Y., Jang-Sik, P., Yoon, B-W. and Song, J-G. 2018. Video Based Smoke and Flame Detection Using Convolutional Neural Network. 4th IEEE International Conference on Signal- Image Technology & Internet-Based Systems (SITIS) held in Las Palmas de Gran Canaria, Spain, 26 - 29 November 2018. pp.365–368. Sowah, R., Ofoli, AR., Krakani, S. and Fiawoo, S. 2014. Hardware module design of a real-time multi-sensor fire detection and notification system using fuzzy logic. IEEE Industry Application Society Annual Meeting (IAS) held in Vancouver, BC, Canada, 05 - 09 October 2014. pp. 1- 6. Tao, C., Wang, P. and Zhang, J. 2016. Smoke Detection Based on Deep Convolutional Neural Networks. IEEE International Conference on Industrial Informatics - Computing Technology, Intelligent, Technology, Industrial Information Integration (ICIICII) held in Wuhan, China, 03 - 04 December 2016.pp. 150 - 153. Vasiliev, AA., Grigoriev, GY., Lagutin, AS., Nabiev, SS., Pisliakov, AV., Samotaev, NN. and Sokolov, AV. 2017. Contemporary technologies of early detection of fire in space vehicles. Acta Astronautica, 135(11): 76–82. Vijayalakshmi, SR. and Muruganand, S. 2017. Smoke Detection in Video Images Using background method for early fire alarm system.2nd IEEE International Conference on Communication and Electronics Systems (ICCES) held in Coimbatore, India, 19 - 20 October 2017. pp.167 - 171. Vinicius, P., Venancioˆ, ABDe., Rezende, MT., Lisboa, CA. and Barbosa, VA. 2021. Fire Detection based on a Two-Dimensional Convolutional Neural Network and Temporal Analysis. IEEE Latin American Conference on Computational Intelligence (LA-CCI) held in Temuco, Chile, 02 - 04 November 2021. pp. 1 - 6. http://www.azojete.com.ng/ mailto:lm324fairchild@yahoo.com Arid Zone Journal of Engineering, Technology and Environment, September, 2022; Vol. 18(3):469-492. ISSN 1596-2644; e-ISSN 2545-5818; www.azojete.com.ng Corresponding author’s e-mail address: lm324fairchild@yahoo.com 492 Wang, X., Zhou, H., Arnott, WP., Meyer, ME., Taylor, S., Firouzkouhi, H., Moosmüller, H., Chow, JC. and Watson, JG. 2020. Evaluation of gas and particle sensors for detecting spacecraft-relevant fire emissions. Fire Safety Journal, 113(03):102977. Wang, Y., Wu, A., Zhang, J., Zhao, M., Li, W. and Dong, N. 2016. Fire smoke detection based on texture features and optical flow vector of contour.12th IEEE World Congress on Intelligent Control and Automation (WCICA) held in Guilin, China, 12 - 15 June 2016. pp. 2879–2883. Wang, Y., Cheng, J. and Liu, T. 2019. Smoke video detection based on lightweight convolutional neural network. 3rd IEEE International Conference on Electronic Information Technology and Computer Engineering (EITCE) held in Xiamen, China, 18 - 20 October 2019. pp.1462–1466. Wu, X., Lu, X. and Leung, H. 2017. An Adaptive Threshold Deep Learning Method for Fire and Smoke Detection. IEEE International Conference on Systems, Man, and Cybernetics (SMC) held in Banff, AB, Canada, 05 - 08 October 2017.pp. 1954 - 1959. Xie, Y., Zhu, J., Cao, Y., Zhang, Y., Feng, D., Zhang, Y. and Chen, M. 2020. Efficient video fire detection exploiting motion-flicker-based dynamic features and deep static features. IEEE Access, 8:81904–81917. Yang, Z., Shi, W., Huang, Z., Yin, Z., Yang, F. and Wang, M. 2018. Combining Gaussian Mixture Model and HSV Model with Deep Convolution Neural Network for Detecting Smoke in Videos. 18th IEEE International Conference on Communication Technology (ICCT) held in Chongqing, China, 08 -11 October 2018. pp. 1266 –1270. Yuanbin, W. 2016. Smoke Recognition Based on Machine Vision. IEEE International Symposium on Computer, Consumer, and Control (IS3C) held in Xi’an, China, 04 - 06 July 2016.pp. 668–671. Zeng, J., Lin, Z., Qi, C., Zhao, X. and Wang, FAN. 2018. An Improved Object Detection Method based on Deep Convolution Neural Network for Smoke Detection. IEEE International Conference on Machine Learning and Cybernetics (ICMLC) held in Chengdu, China, 15 - 18 July 2018. pp. 184 - 189. Zhang, J., Zhu, H., Wang, P. and Ling, X. 2021. ATT Squeeze U-Net: A Lightweight Network for Forest Fire Detection and Recognition. IEEE Access, 9:10858–10870. Zhang, Q., Liu, F., Li, X. and Li, B. 2017. Dissipation Function and ViBe Based Smoke Detection in Video. 2nd IEEE International Conference on Multimedia and Image Processing (ICMIP) held in Wuhan, China, 17 - 19 March 2017.pp. 325 - 329. Zhao, X., Ji, H., Zhang, D. and Bao, H. 2018. Fire Smoke Detection Based on Contextual Object Detection. 3rd IEEE International Conference on Image, Vision, and Computing (ICIVC) held in Chongqing, China, 27 - 29 June 2018. pp. 473–476. file:///C:/user/Downloads/azojete143/www.azojete.com.ng mailto:lm324fairchild@yahoo.com