Multi-Sensor Photoelectric Fire Alarm Device Implementation for Early Fire Detection in Campsites

Abstract

With the growing popularity of leisure activities such as camping and glamping, the incidence of fires at camping sites has increased. This study focuses on improving the effectiveness of photoelectric fire alarm devices by incorporating temperature and humidity data for early fire detection in confined spaces, such as campsites. This study proposes a novel multi-sensor fire alarm system that dynamically adjusts fire detection threshold values based on temperature and humidity data collected by unmanned automatic weather observation systems. The prototype, which was implemented using Raspberry Pi and multiple sensors, demonstrated approximately 20% faster fire detection speed than existing photoelectric fire alarm systems, as verified through experiments in a simulated camping environment. The proposed approach is expected to advance fire alarm systems, enabling faster and more accurate fire detection in diverse environments, particularly at campsites.

1. Introduction

The devastating consequences of fires, including loss of life and property, inflict significant economic burdens worldwide, with annual damages reaching approximately 1% of the global gross domestic product (GDP) and claiming thousands of lives [1]. As leisure activities such as camping and glamping gain popularity, a worrying trend has emerged: a rise in fire incidents at camping sites. Data from the National Fire Data System corroborates this observation, revealing a 73% surge in camping participation from 3.01 million in 2017 to approximately 5.23 million in 2021. This increase in camping activities was correlated with a notable increase in fire incidents at campsites, from 43 to 71 over a three-year period in South Korea, as shown in Figure 1 [2].

The risk of fire outbreaks is particularly high in small and confined environments, such as camping sites. The predominant use of highly flammable materials, such as nylon and polyester, in tent construction exacerbates the rapid spread of fire, making extinguishment and evacuation more challenging, as shown in Figure 2. A fire that spreads radially and involves flaming typically follows a growth pattern proportional to the square of time and is referred to as a t-squared (t²) fire. These fires are categorized by their growth speed—ultra-fast, fast, medium, and slow—based on the time required for the fire to reach a heat release rate (HRR) of 1000 kW (1 MW). Lawrence [3] reported that fires in a nylon tent classified as a t² fire growth pattern spread extremely quickly, with the HRR potentially reaching 1 MW in just 20–42 s. This underscores the importance of even marginal reductions in fire detection time, which can provide vital moments for evacuation before flashover—the point at which a fire becomes uncontrollable. In the fire captured in Figure 2, the incident started at approximately 2:10:02, and the tent was fully engulfed by 2:14. Applying these observations to this study, a fire alarm system capable of detecting a fire within 2 min and 48 s could enable timely evacuation, preventing severe injuries by alerting occupants before the flames reach the tent entrance.

The proposed fire alarm system significantly enhances early fire detection, thereby reducing the risk of damage and injury. This is particularly vital in camping environments where tents can be completely consumed by fire within 5 min of ignition. The critical “golden time” for survival in such fires is the period before flashover, which typically occurs within approximately 5 min. Therefore, early detection is imperative to ensure survival and minimize injury.

Furthermore, incidents involving casualties due to toxic gas inhalation in confined spaces, such as tents, have been documented. For instance, in March 2015, a tragic incident at a glamping site on Ganghwa Island, South Korea, resulted in five fatalities [4]. More recently, on 11 November 2023, five campers in South Korea tragically died from carbon monoxide poisoning caused by a brazier inside their tent [5]. Fire-related accidents at campsites are not exclusive to South Korea [6]. This underscores the urgent need for the development and deployment of effective fire detection and alarm systems tailored for camping environments.

Figure 2. CCTV footage of campsite fires [7].

Figure 2. CCTV footage of campsite fires [7].

Fire detection systems are crucial for providing early warning systems and implementing prevention strategies by identifying critical indicators of fire, such as smoke, flames, and abnormal temperature increases. In a 2022 study, F. Khan [8] highlighted various sensors used in contemporary fire alarm systems, including heat, gas, flame, multi-composite, graphene oxide (GO) [9], and smoke sensors, as depicted in Figure 3.

Furthermore, D. Song [10] demonstrated a significant correlation between the likelihood of fire occurrence and humidity and temperature. Based on this finding, we propose a method that integrates real-time meteorological information to highlight areas with high likelihood of fire occurrence on a map. However, the potential active use of humidity sensors in fire detection has not yet been widely discussed.

Various types of fire detectors, including photoelectric, fixed temperature, and rate-of-rise temperature detectors, are available on the market. According to the Korea Fire Industry Technology Institute [11], production figures for fire detectors between 10 May and 9 June 2023, included 22,000 rate-of-rise temperature detectors, 13,000 photoelectric detectors, and 7000 fixed temperature detectors. Photoelectric fire detectors rank second in production due to their rapid fire detection capability. Rapid fire detection is essential in confined spaces like camping sites, and we also considered cost to ensure wider adoption. Therefore, in this study, we focused on improving the cost-effectiveness of photoelectric fire detectors, which offer the benefit of fast fire detection. These detectors function by using a light-emitting element and a light-receiving sensor inside a dark chamber to detect smoke-induced light scattering. However, as Wu, L [12] reported, fires release not only smoke but also heat, flames, and gases. The characteristic parameters of a fire include temperature, smoke concentration, and carbon monoxide. Chemical sensors respond more rapidly than smoke alarms. Therefore, combining multiple sensors in a fire alarm system can enhance the detection speed and accuracy.

In this study, we analyzed regional temperature and humidity data obtained from the Fire Agency’s national fire system [13] and the Meteorological Agency’s disaster prevention system (Automatic Weather Station (AWS)) [14]. The objective was to explore the relationship between temperature, humidity, and the likelihood of fire occurrence, and to integrate these findings into the design and implementation of a photoelectric fire detector. A prototype fire alarm system was developed using a Raspberry Pi, DHT11, and a light sensor. Performance evaluations demonstrated that the prototype detected fires approximately 20% faster than existing photoelectric fire detectors on the market

The primary contributions of this study are as follows:

We propose an enhanced multi-sensor photoelectric fire detector that integrates temperature and humidity data into a traditional photoelectric fire alarm system.

The proposed multi-sensor photoelectric fire detector represents a significant advancement in fire detection technology, with potential applications in various environments where temperature and humidity are crucial factors for fire risk, such as camping sites, small rooms, and semi-basements.

The remainder of this paper is organized as follows: Section 2 reviews the current state of research on smoke detectors and distinguishes between visual and non-visual methods. This review establishes the direction of ongoing research and emphasizes the necessity and purpose of this study. Section 3 analyzes the influence of humidity and temperature on fire incidents and derives a multiple linear regression formula to develop an enhanced fire detection algorithm. The validity of the derived formula is verified by assessing its homoscedasticity, normality, and linearity. In addition, this section outlines the software framework of the fire detection system, describing the internal sensors and the prototype implementation process. In Section 4, experiments are conducted to empirically verify the performance of the fire detector in real-fire scenarios by analyzing the results of these tests and evaluating the variability of the detector’s alarm threshold in response to humidity and temperature changes. Finally, Section 5 summarizes the findings and discusses future development needs.

2. Related Studies

Currently, smoke detectors are categorized into two main types: visual and non-visual detectors. Recent studies have primarily focused on visual detectors, with efforts aimed at reducing error rates and enhancing detection accuracy. For example, J. Ryu [15] proposed a visual smoke detection algorithm that minimizes false detections by employing image preprocessing techniques, including HSV color transformation and Harris Corner Detection, to minimize the miss detection rate. In another notable study, S. Saponara [16] developed a fire and smoke detection system using video cameras, employing a deep learning model called YOLOv2 to reduce the computational load of complex models and decrease the false alarm rate. However, visual fire detectors present certain challenges despite their advanced capabilities. These systems require dedicated video feeds and extensive knowledge of the monitored areas, coupled with high installation costs.

Given these limitations, non-visual methods play a crucial role in providing cost-effective and widely accessible fire detection solutions. The most widely used non-visual smoke detector is the photoelectric detector. This type of detector operates by detecting light scattering caused by smoke particles using a light-emitting element and a photodetector. When smoke is present, light scattering reduces the amount of light reaching the photodetector, indicating a fire. The widespread adoption of photoelectric smoke detectors is largely due to their relatively simple design and affordability compared to other types. However, these detectors are prone to higher error rates and slower response times, which has led to ongoing research efforts aimed at improving their performance. Recent advances in non-visual detection methods are exemplified by the work of Hayashi, Y [17], who proposed a method for detecting smoldering fires using a capacitive MEMS hydrogen sensor equipped with a microheater, enabling fire detection with minimal smoke production. This structure successfully detected 66-ppm hydrogen generated by fire. In addition, Qiu, X [18] proposed a method for detecting fires using a carbon monoxide (CO) sensor for early fire detection. The system was designed to detect CO with high sensitivity by applying wavelength modulation spectroscopy (WMS) to a 32-bit microcontroller. However, these non-visual detectors do not lack challenges. Despite their effectiveness in detecting hydrogen or CO, they do not significantly improve the error rate. As highlighted by R. Bogue [19], many detection technologies are available for fire detection; however, no single solution is universally applicable. This underscores the need for an ongoing discussion and exploration of various methods tailored to different situations and locations.

In contrast, a significant body of research on fire occurrences has leveraged time-series meteorological data, including variables such as temperature, humidity, wind speed, cloud cover, atmospheric pressure, and precipitation, along with wildfire occurrence data. These datasets are often used in logistic regression models to estimate wildfire probabilities. For instance, Jain, P. [20] recently identified a correlation between increasing regional wildfires and climate change, analyzing the trends between each factor, and highlighting the impact of decreasing relative humidity and rising temperatures. Similarly, to quantitatively assess the causal relationship between meteorological factors and wildfires, J. Yoo [21] investigated the factors influencing seasonal wildfires in Gangwon Province, South Korea, using PLS-SEM. In addition, to forecast the risk of large wildfire occurrences, S. Kang [22] developed criteria for predicting the risk of large wildfires through a case analysis spanning approximately 20 years using effective humidity and average wind speed. Beyond wildfire contexts, D. Song [10] investigated the relationship between meteorological data and fire incidents by calculating the hourly frequency of fire occurrences based on humidity and temperature and by statistically analyzing the regional characteristics for each range.

In previous studies focusing on fires at camping sites, H. Choi [23] analyzed the primary causes of fires in campervans, while E. Lee [24] explored the actual conditions and regulatory framework for fires in camping vehicles, reviewing case studies and prevention strategies. M. Almeida [25] highlighted the heightened fire risk in camping sites located in dense forest environments and analyzed fire trends in such areas. In addition, J.F. Fraser [26] identified that children often suffer burns from campfires at camping sites and conducted investigations on the evolution of fires originating from previous campfires.

Previous studies have highlighted a strong correlation between various meteorological conditions and the occurrence of actual fires. This underscores the critical need for research into fire alarm systems capable of rapidly detecting fires in small, enclosed spaces, such as camping tents, by leveraging these meteorological conditions.

3. Multi-Sensor Photoelectric Fire Detection System

The relationship between meteorological data and fire data was initially analyzed, leading to the implementation of a multi-sensor photoelectric fire detector that leverages this relationship.

3.1. Relationship Between Weather and Fire Occurrence

3.1.1. Data for Analysis

This study’s dataset comprises hourly meteorological and fire data recorded in 2021 from various regions, including Seoul, Daegu, North Gyeongsang, Gyeonggi, and Daejeon, in South Korea [13,14]. Meteorological data, including temperature, humidity, wind direction, wind speed, and precipitation, were collected at minute intervals from Automatic Weather Stations (AWS) situated at key observation stations in areas corresponding to the 17 fire headquarters. This dataset encompasses a total of 1,185,864 individual cases, with this study focusing primarily on temperature and humidity data, as illustrated in

Table 1. Sample data extracted from AWS dataset.

Location	Location Name	Day	Temperature (°C)	Humidity (%)
116	Gwanak-gu	2021-01-01 1:00	−12.3	74
116	Gwanak-gu	2021-01-01 2:00	−11.6	75
116	Gwanak-gu	2021-01-01 3:00	−12	74
116	Gwanak-gu	2021-01-01 4:00	−11.6	76
116	Gwanak-gu	2021-01-01 5:00	−10.8	60
116	Gwanak-gu	2021-01-01 6:00	−10.5	54
116	Gwanak-gu	2021-01-01 7:00	−9.8	37
116	Gwanak-gu	2021-01-01 8:00	−9.9	52
116	Gwanak-gu	2021-01-01 9:00	−9.4	62
116	Gwanak-gu	2021-01-01 10:00	−7.9	67

, considering their influence on fire ignition at campsite tents.

Fire occurrence data, as presented in

Table 2. Fire occurrence data from the National Fire Information System.

Date of Fire Outbreak	City	District	Ignition Factor
2020-01-01 0:00	Seoul	Guro-gu	Careless
2020-01-01 0:05	Gwangju	Gwangsan-gu	Careless
2020-01-01 0:06	Gwangju	Gwangsan-gu	Unknown
2020-01-01 0:07	Gyeonggi	Yeoju-si	Electrical factor
2020-01-01 0:12	Gyeonggi	Yangpyeong-gun	Electrical factor
2020-01-01 0:21	Incheon	Michuol-gu	Unknown
2020-01-01 0:43	Gwangju	Buk-gu	Careless
2020-01-01 0:57	Incheon	Ganghwa-gun	Careless
2020-01-01 1:12	Gyeongsangdo	Changnyeong-gun	Electrical factor
2020-01-01 1:15	Seoul	Gannam-gu	Unknown

, were obtained from the National Fire Information System [13], which contains fire investigation reports documented by 183 Fire stations across South Korea. This dataset contains 184 distinct attributes that were categorized under the National Fire Information System and comprised a total of 36,268 instances of fire occurrences.

In this study, we merged data from these two datasets based on the fire occurrence date and the corresponding province, district, or city within the regions of Seoul, Daegu, North Gyeongsang, Gyeonggi, and Daejeon in South Korea. Initially, as indicated in

Table 3. Sample of merged data.

Date of Fire Outbreak	City	District	Ignition Factor	Location	Temperature (°C)	Humidity (%)
2021-01-02 9:00	Daegu	Sinam	Electrical factor	860	−0.9	65
2021-01-02 11:00	Daegu	Sinam	Mechanical factor	860	0.3	42
2021-01-02 14:00	Daegu	Dalseon	Unknown	828	3.5	29
2021-01-02 14:00	Daegu	Sinam	Electrical factor	860	1.6	30
2021-01-02 18:00	Daegu	Sinam	arson	860	−0.7	30

, a total of 24,412 cases were extracted from the data. After excluding records with ambiguous regional names or missing humidity data, the final dataset used for the analysis consisted of 15,946 cases.

3.1.2. The Relationship Model Between Fire Occurrence and Temperature and Humidity

In this study, the extracted and merged data were categorized into intervals based on temperature and humidity. The temperature was segmented into approximately 5° intervals ranging from −40 °C to +40 °C, and the humidity into approximately 5% intervals ranging from 15% to 90%. The number of fire occurrences in each interval was then counted. Similarly, the AWS data were categorized based on the frequency of occurrence within the same temperature and humidity range. The AWS dataset covers 109 regions, and temperature and humidity were simultaneously recorded. To account for the differences among the datasets, the number of fire occurrences was multiplied by 109, and this adjusted figure was then divided by the frequency of occurrences in the AWS data for each temperature and humidity range to calculate the fire occurrence rate. The data preprocessing included several steps. First, data points with abnormally low fire occurrence rates at specific temperatures and humidity levels, which could represent outliers, were removed using the Interquartile Range (IQR) method [27]. The regions where the AWS did not measure humidity were also excluded. After these adjustments, the dataset was reduced to 24,412 preprocessed data points. Next, temperature was restricted to the range of −40 °C to +40 °C and binned into 5° intervals, while humidity was restricted to 20–90% and binned into 5% intervals. The number of fire occurrences within the categorized ranges was calculated, and the frequency of occurrences within each temperature and humidity range in the AWS data over a year was recorded. To correct for potential disparities between the fire occurrence data and AWS data, the number of fire occurrences was adjusted by multiplying the number of occurrences by 109, representing the total number of regions covered by the AWS data. Finally, the adjusted number of fire occurrences was divided by the frequency of the AWS data points within each temperature and humidity range to calculate the fire occurrence rate. The relationship between the fire occurrence rate and each temperature/humidity range is illustrated in Figure 4 and Figure 5.

It is not necessary for temperature and humidity to each have a linear relationship with the fire occurrence rate in multiple linear regression. Therefore, using these preprocessed data, a multiple linear regression model was constructed, as illustrated in Figure 6. The model produced the following equation, which captures the relationship between the fire occurrence frequency (y) and temperature and humidity variables, as represented in Equation (1):

y = 0.0715 × Temperature − 0.380 × Humidity + 0.416

(1)

Equation (1) shows the impact of specific temperatures and humidity levels on fire occurrence.

3.1.3. Statistical Evaluation

In this study, we employed various statistical methods to evaluate the accuracy and fit of the multiple linear regression model represented by Equation (1). Validation procedures were conducted to confirm the adherence of the model to its fundamental assumptions. To achieve this, the Ordinary Least Squares (OLS) method, which is a standard approach in multiple linear regression analysis, was employed [28].

In Figure 7, the R-squared value, which measures the model’s goodness of fit, is 0.863, and the adjusted R-squared value is also 0.863. This indicates that the model explains approximately 86.3% of the variance, reflecting a high level of explanatory power. However, assessing a model’s reliability based solely on this value is inadequate; therefore, it is necessary to examine other fundamental assumptions of linear regression [29].

First, the F-statistic is 4.910 × 10⁴, with a corresponding p-value (Prob(F-statistic)) of 0.00%. Given that the p-value is less than the significance level (0.05), the regression model is considered statistically significant. Furthermore, an examination of the t-statistic values and their associated p-values (p > |t|) for each variable confirmed that all variables were statistically significant.

Next, we assessed whether the basic assumptions of multiple linear regression were satisfied using a residual plot (Figure 8), Q-Q plot (Figure 9), Variance Inflation Factor (VIF), and the Durbin–Watson test, employing the bootstrap method. The VIF, which measures multicollinearity in multiple regression, was calculated based on the R-squared values obtained by regressing each independent variable against the others [30]. The VIF values were 1.00 for the constant term and 1.027525 for both ×1 and ×2, indicating that multicollinearity is not a concern because these values are close to 1. Furthermore, the Q–Q plot and residual plot were examined to verify the assumptions of homoscedasticity, normality, linearity, and independence. As shown in Figure 8 and Figure 9, the Q–Q plot shows points clustered along a 45° line, and the residual plot indicates a random distribution of residuals, confirming that the model adheres to these assumptions.

However, the Durbin–Watson statistic of 1.165, along with the elevated Omnibus and Jarque-Bera (JB) test values, indicates that the residuals may not fully satisfy the assumptions of normality and independence. Nevertheless, the residual plot indicates a very weak positive autocorrelation, which is acceptable given that the fire occurrence rate relative to temperature and humidity may exhibit some nonlinearity, especially at extreme values.

In conclusion, the proposed multiple linear regression model, as represented by Equation (1), is statistically significant and satisfies the fundamental assumptions. These results validate the reliability of the proposed model.

3.2. Multi Sensor Photoelectric Fire Detection System Design

3.2.1. Photoelectric Fire Detector

A photoelectric smoke detector is a type of fire alarm that uses light sensors to detect smoke. Figure 10 [31] illustrates the structure of a typical photoelectric fire detector. The fundamental operating principle involves light beams, either infrared or visible, scattered by smoke particles within a dark chamber. The sensors detect this light scattering, which plays a critical role in fire detection.

As shown in Figure 10, a standard photoelectric detector comprises a dark chamber containing a light-detecting sensor and an LED, along with a circuit connecting them. It is crucial for the chamber to allow smoke entry while simultaneously obstructing any external light sources, except for the light emitted by the LED. The circuit powers the LED to emit light at regular intervals, and the sensor measures the amount of light, enabling the detection of smoke and the assessment of potential fire.

3.2.2. Prototype Design and Implementation of the Proposed Multi-Sensor Photoelectric Fire Alarm

In this study, a prototype of a multi-sensor photoelectric fire alarm was developed using components commonly found in commercial fire alarms, as illustrated in Figure 11 and Figure 12. A key element of the design was to ensure that the chamber remained devoid of external light while allowing adequate smoke entry. To achieve this, the internal plastic components of the existing fire alarm models were designed with a curved structure, as shown in Figure 13, and were externally covered with a light-obstructing gray film to enhance the efficacy of the detection mechanism.

The primary components used in this study for fire detection included a light-emitting diode (LED) and a light sensor (specifically, an Arduino-light-photocell-Cds-light sensor (SZH-SSBH-011), SMG-A, Gwangju, South Korea), as shown in Figure 14. These components are essential for detecting changes in light transmission due to the presence of smoke, thereby indicating the potential occurrence of a fire. Furthermore, a combined temperature and humidity sensor (DHT11 [32]) was integrated into the system, as shown in Figure 15, and it was positioned at the rear of the fire alarm device to continuously monitor the temperature and humidity. An analog-to-digital converter (ADC) (MCP3008 [33]) was connected to the light sensor to enable the real-time collection of analog values. All components were integrated with a Raspberry PI 4 to form the core of the fire detection experimental setup, as outlined in the schematic in Figure 11. The experiment simulated dry summer conditions with an external temperature of approximately 27 °C and humidity of approximately 23%.

Moreover, based on the findings in Section 3.1.2, Equation (1) was employed to recalculate threshold values whenever fluctuations in temperature and humidity occurred, as illustrated by the code and schematic in Figure 16. In this framework, “current_temp” and “current_humidity” represent the present temperature and humidity measurements, and “previous_temp” and “previous_humidity” refer to the preceding recorded values. The variable “current_val” was calculated using the regression equation, Equation (1), and “bright” indicates the light intensity measured by the light sensor. This approach enabled the dynamic updating of threshold values “update_bright” for the fire alarm in real-time.

As highlighted in the Related Studies section, traditional photoelectric smoke detectors detect a fire when the light received by the photodetector falls below a certain threshold. In contrast, the proposed fire detector detects fires when the received light intensity drops below an adaptive threshold, referred to as update_bright. This threshold is not static; it adjusts dynamically based on variables that influence fire conditions, such as temperature and humidity. This adaptive approach allows the detector to respond more swiftly to potential fire incidents.

4. Experiments and Results

This section describes an experimental study that assesses the efficacy of the proposed multi-sensor photoelectric fire detector in a simulated camping environment, as shown in Figure 17. The experimental setup involved a shower booth tent as a stand-in for a typical camping site. In this controlled environment, a fire was initiated to test the detector response. Both a standard photoelectric fire detector [Stand-alone Photoelectric Type 2 Fire Detector by Hyundai Fire Protection Industry] and the proposed multi-sensor photoelectric fire detector were installed at the apex of the shower tent, adhering to common fire detector placement practices. Safety precautions were strictly followed, including the use of flame retardants, readily available fire extinguishers, and the presence of trained safety personnel throughout the experiment.

The experimental results, as illustrated in Figure 18, indicated a temperature increase from an initial temperature of 40 °C to a peak at 45 °C. Concurrently, the humidity level depicted in Figure 19 began at 40% and increased to 50% due to the evaporation at the ignition point shown in Figure 20 before gradually decreasing to 29%. The commercially available photoelectric fire detector took approximately 3 min and 28 s to detect the fire after it was ignited, and the average analog value of brightness recorded by the light sensor at that time was approximately 890.

Figure 20 displays two graphs: red and blue. The red graph in Figure 20 illustrates the real-time adjustments of the fire alarm threshold based on the temperature and humidity data in Figure 18 and Figure 19, as computed using the algorithm shown in Figure 16. The blue graph in Figure 20 shows the actual light levels detected by the light sensor. A fire is identified when the threshold value depicted in the red graph drops below the light sensor’s reading (represented in the blue graph), and it is then registered as a new alarm.

The results revealed that conventional photoelectric fire detectors operate with a threshold of an analog value of approximately 890. However, in the proposed smoke detector, the threshold was adjusted to between 900 and 920 under normal, non-fire conditions. When fire conditions were detected, the threshold rapidly decreased to facilitate quicker fire detection. This indicates that maintaining a stable threshold value when the temperature and humidity remain relatively constant does not significantly increase the likelihood of false alarms. The detailed experimental results are provided in the table in Appendix A, which displays the temperature, humidity, brightness, and threshold values recorded by the fire alarm system between 17:55:40 and 17:56:28. The table also compares the fire detection time of our multi-sensor photoelectric fire detector with that of a commercially available model. Furthermore, the proposed fire alarm system, which considers humidity, was able to detect a fire approximately 40 s (20%) faster than a conventional alarm system, requiring approximately 2 min and 48 s from the onset of the fire. The reduced response time is a significant improvement, particularly in terms of enhancing evacuation times during a fire.

Table 4. Comparison with conventional fire alarms.

Detection Method	Ryu [15]	Saponara [16]	Hayashi [17]	Qiu [18]	Multi-Sensor Fire Alarm (Proposed)
Has the error rate decreased?	O	O	X	X	O
Has the accuracy increased?	O	O	O	O	O
Is no prior training required?	X	X	O	O	O
Is the construction cost affordable?	X	X	O	X	O

presents a comparative analysis between the proposed and existing detection systems based on various criteria, such as cost, error rate, accuracy, and the need for pre-learning for operation. The proposed multi-sensor fire alarm exhibited improvements in these comparison metrics compared with the other fire alarm types. For instance, a MOSFET-based fire alarm [18] exhibited a high error rate because of its sensitivity to temperature changes.

5. Conclusions

This study proposes a fire alarm system that incorporates both temperature and humidity sensors. The proposed alarm system has an enhanced capability to detect fires more rapidly than traditional alarm systems. This enhancement is achieved by dynamically adjusting the alarm threshold based on temperature and humidity changes, which is particularly advantageous in environments such as camping sites, where false alarms caused by smoke from sources such as mosquito coils and insecticides are common.

The experimental results revealed that the proposed fire alarm system detected fires approximately 40 s (20%) faster than conventional alarms. By incorporating temperature and humidity sensors, which are cost-effective components, the proposed approach provides a more affordable solution than more complex detection systems. This study introduces a novel concept for enhancing existing fire alarms with temperature and humidity sensing capabilities.

The study also incorporated an analysis of actual fire incident data from the National Fire Data System [13] and regional meteorological data from the Meteorological Administration’s disaster prevention system (AWS) [14]. This analysis culminated in the formulation of a multiple linear regression equation: y = 0.07145995281326634 × temperature − 0.37967797357088395 × humidity + 0.4160213392065553. Leveraging this equation, a fire alarm system was implemented using a Raspberry Pi, a DHT11 sensor, and a light sensor. This system uniquely addresses the environmental characteristics of camping sites where smoke is commonly present, allowing for threshold adjustment without increasing error rates, thereby facilitating faster fire detection. The experiments conducted in a shower tent demonstrated that the proposed fire alarm system detected fires significantly faster than traditional alarm systems.

Future research directions should focus on advancing this prototype into a fully operational fire alarm system, with extensive experiments in diverse environments to further validate the efficacy of integrating temperature and humidity sensors into fire detection systems.

In conclusion, this study posits that the findings and applications of this research are not confined to camping sites, but extend to any environment where temperature and humidity are critical factors in fire detection, such as small rooms or semi-basement apartments. This broadens the potential applicability of the proposed fire alarm system and makes it valuable in diverse settings where early fire detection is essential.