Billiard hall smoke and fire detection early warning notification method and system

Abstract

The invention provides a billiard hall smoke and fire detection early warning notification method and a billiard hall smoke and fire detection early warning notification system, which are used for realizing high-precision and low-false-alarm regionalized early smoke and fire early warning of a billiard hall. The method comprises the steps of performing regional smoke concentration monitoring and screening on a billiard hall according to an environment data stream to obtain a smoke exceeding region, and performing temperature analysis and abnormal region identification on the smoke exceeding region to obtain a temperature abnormal region and a corresponding infrared region image, performing edge fitting of a heat source shape and rule evaluation on the infrared region image to obtain a potential smoke and fire list in each temperature abnormal region, and performing type identification on each potential smoke and fire in the potential smoke and fire list to obtain a smoke and fire type of each potential smoke and fire; and when the smoke and fire type is a preset fire alarm type set, smoke motion analysis and scene identification verification are performed on the potential smoke and fire to generate alarm decision data of each temperature abnormal area, and the temperature abnormal areas are alarmed according to the alarm decision data in a graded alarm mode.

Description

Technical Field

[0001] This application relates to the field of smoke and fire warning technology, and in particular to a method and system for detecting, warning and notifying smoke and fire in a billiard hall. Background Technology

[0002] Fire prevention and early warning systems in billiard halls need to focus on three aspects: First, early detection and accurate location of suspected smoke and fire are crucial for timely response and rapid evacuation, preventing small fires from escalating into large ones. Second, while ensuring safety, false alarm rates and disruption to the customer experience must be minimized, as billiard halls are open entertainment spaces and frequent false alarms can negatively impact business and image. Third, tiered response and traceability are required, meaning alarms should be pushed to on-site staff or fire departments according to risk level, and the process and evidence should be recorded for post-incident analysis and system optimization, while also ensuring privacy protection and regulatory compliance.

[0003] Currently, most billiard halls rely on single sensors or manual supervision for detection and notification: photoelectric or ionization smoke detectors, local temperature detectors, and linked audible and visual alarms are installed on the ceiling or in designated areas, supplemented by closed-circuit television for manual viewing by security guards or management. Some venues are equipped with centralized fire alarm controllers and automatically linked ventilation or fire extinguishing systems, as well as simple notification processes that push notifications to management personnel via SMS / APP. When staffing permits, regular patrols and inspection records are also relied upon to compensate for the inadequacy of equipment monitoring.

[0004] The existing methods mentioned above lack the ability to fuse multimodal and multi-evidence data in real time and make adaptive judgments. This leads to a large number of false alarms or missed alarms in complex indoor scenarios such as smoking, steam, lighting heat sources, or dense crowds. This not only affects the customer experience but also increases the burden of manual verification and delays the response to real fires. Summary of the Invention

[0005] This application provides a method and system for detecting and issuing early warnings of smoke and fire in billiard halls, which can achieve high-precision, low-false-alarm regional early warning of smoke and fire in billiard halls.

[0006] Firstly, this application provides a method for detecting and issuing early warnings of fireworks in a billiard hall, the method comprising: Based on the real-time acquired environmental data stream, the billiard hall is monitored for regional smoke concentration and screened to identify areas with excessive smoke. Temperature analysis and abnormal area identification are then performed on these areas to obtain abnormal temperature areas and corresponding infrared images. Edge fitting and rule evaluation of heat source shapes are performed on the infrared region image to obtain a list of potential fireworks in each temperature anomaly region. According to the preset fireworks feature classification rules, the type of each potential fireworks in the potential fireworks list is identified to obtain the fireworks type of each potential fireworks. When the type of smoke and fire is a preset set of fire alarm types, smoke movement analysis and scene recognition verification are performed on the potential smoke and fire to generate alarm decision data for each temperature anomaly area. The alarm is triggered in the temperature anomaly area based on the alarm decision data using a preset tiered alarm method.

[0007] In one possible implementation, the step of monitoring and filtering the area smoke concentration in the billiard hall based on real-time acquired environmental data streams to identify areas with excessive smoke includes: Based on the smoke concentration sequence, the short-term and long-term smoke concentration baselines of each monitoring area in the billiard hall are calculated using a preset sliding time window set. Based on environmental humidity data, personnel density data, the short-term smoke concentration baseline, and the long-term smoke concentration baseline, the smoke concentration threshold is adjusted to generate the smoke concentration threshold for each monitoring area. When the real-time smoke concentration data in the smoke concentration sequence is greater than the smoke concentration threshold, the corresponding monitoring area is determined as the area to be verified, and the area to be verified is continuously monitored for smoke concentration according to the preset monitoring period and the smoke concentration threshold. When the real-time smoke concentration data within the monitoring period is greater than the smoke concentration threshold, the corresponding area to be verified is determined as a smoke exceeding the standard area.

[0008] In one possible implementation, the step of performing temperature analysis and abnormal region identification on the smoke-exceeding area to obtain temperature abnormality areas and corresponding infrared region images includes: Infrared images of the area with excessive smoke are acquired, and median filtering is applied to the infrared images to obtain noise-reduced area images. By using a preset thermodynamic gradient enhancement algorithm, the temperature gradient of the noise reduction region image is optimized to obtain an enhanced temperature gradient field image. The enhanced temperature gradient field image is segmented according to a preset temperature gradient threshold to obtain a set of candidate thermal anomaly regions; Based on the candidate thermal anomaly region set, the temperature difference feature extraction process between adjacent regions of the enhanced temperature gradient field image is performed to obtain the temperature difference feature set of each candidate thermal anomaly region. By using a preset non-maximum suppression method based on the temperature difference feature set, the candidate thermal anomaly regions in the candidate thermal anomaly region set are fused to generate a temperature anomaly region. The enhanced temperature gradient field image corresponding to the temperature anomaly region is used as the infrared region image corresponding to the temperature anomaly region.

[0009] In one possible implementation, the step of performing edge fitting and rule evaluation on the infrared region image to obtain a list of potential fireworks within each temperature anomaly region includes: The infrared region image is subjected to heat source identification and heat source boundary optimization processing to obtain a regional heat source image; Heat source edge contours are extracted from the heat source image of the region to obtain the contour boundary point set of each heat source; The contour boundary point set is edge-fitted by a preset polygon approximation algorithm to obtain the contour feature point sequence of each heat source. Based on the contour feature point sequence, extract the geometric feature set of each heat source from the regional heat source image; Based on the geometric feature set, the geometric feature set is used to evaluate the shape regularity of each heat source in the temperature anomaly region and obtain the shape anomaly probability score of each heat source. Based on the preset fire probability threshold and the shape anomaly probability score, the heat sources are classified and screened to obtain a list of potential fires in each temperature anomaly area.

[0010] In one possible implementation, the step of identifying the type of each potential firework in the potential firework list according to a preset firework feature classification rule, and obtaining the firework type of each potential firework, includes: Acquire a visible light image of the temperature anomaly region, extract the color space feature set of each potential firework from the visible light image according to the potential firework list, and extract the thermodynamic feature set of each potential firework from the heat source image of the region according to the potential firework list; The color space feature set and the thermodynamic feature set are combined to form the photothermal feature set corresponding to potential fireworks; Calculate the similarity between the photothermal feature set and each type feature subset in the preset fireworks type feature set to obtain the type probability distribution of each potential fireworks; Based on the type probability distribution, the potential fireworks are identified using preset fireworks feature classification rules to determine the type of each potential fireworks.

[0011] In one possible implementation, the step of performing smoke motion analysis and scene recognition verification on the potential fireworks to generate alarm decision data for each temperature anomaly area includes: Dense optical flow field analysis was performed on the visible light images corresponding to the potential fireworks to extract the particle flow feature set of each temperature anomaly region; Foreground detection processing is performed on the visible light image to extract background color and texture features of each temperature anomaly region; Based on the particle flow feature set and the background color texture features, the visible light image is analyzed for the smoke motion of each potential firework to obtain the smoke motion region of each potential firework and the corresponding smoke flow feature set. Based on the preset hazardous material types, the visible light image is used to perform semantic scene object recognition and smoke correlation calculation to obtain the spatial correlation index between each potential smoke and hazardous material. Using a pre-defined multi-evidence fusion algorithm, a fire threat confidence score is calculated based on the spatial correlation index, the smoke movement area, and the smoke flow feature set. Based on the fire threat confidence score and the preset confidence threshold level rules, the spatial correlation index, the smoke movement area, and the smoke flow feature set are encapsulated into alarm decision data for each temperature anomaly area.

[0012] Secondly, this application provides a billiard hall smoke detection and early warning notification device, the device comprising: The smoke screening module is used to monitor and screen the area smoke concentration in the billiard hall based on the real-time acquired environmental data stream, and to identify areas with excessive smoke. Anomaly detection module is used to perform temperature analysis and anomaly detection on the area where smoke exceeds the standard, and to obtain images of the temperature anomaly area and the corresponding infrared area. The smoke and fire detection module is used to perform edge fitting and rule evaluation of the heat source shape in the infrared region image to obtain a list of potential smoke and fire in each temperature anomaly region. The fireworks classification module is used to identify the type of each potential firework in the potential fireworks list according to the preset fireworks feature classification rules, and obtain the fireworks type of each potential firework. The fire alarm verification module is used to perform smoke movement analysis and scene recognition verification on the potential smoke when the smoke type is a preset set of fire alarm types, and generate alarm decision data for each temperature abnormality area. The graded alarm module is used to issue alarms to the temperature abnormality area based on the alarm decision data using a preset graded alarm method.

[0013] In summary, this application includes at least the following beneficial technical effects: 1. Through smoke threshold adaptation, thermal-visible multispectral fusion, and morphological and motion joint verification, multiple pieces of evidence can be used to corroborate each other, eliminating false alarms caused by steam, lighting, or human body heat sources.

[0014] 2. The system first triggers the smoke and then performs infrared temperature gradient and shape analysis, which can quickly locate abnormal areas and determine the type of smoke and fire, enabling graded (hazard / fire alarm) notifications to facilitate targeted response.

[0015] 3. Dynamically adjust the threshold based on the environmental context (humidity, personnel density) and record feedback samples for online learning, so that the system can gradually improve its robustness and detection accuracy as the scene changes. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a method for detecting and issuing early warnings of fireworks in a billiard hall, as provided in an embodiment of this application. Figure 2 This is a schematic diagram of the structure of a billiard hall fire detection and early warning notification device provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of the computing device provided in the embodiments of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. With the development of technology and the emergence of new scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0018] The terms "first," "second," etc., used in the specification, claims, and drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the description of embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not expressly listed or inherent to those processes, methods, products, or apparatuses.

[0019] like Figure 1 The diagram shown is a flowchart illustrating the billiard hall fire detection and early warning notification method provided in this embodiment of the application. The billiard hall fire detection and early warning notification method provided in this embodiment of the application includes the following steps.

[0020] Step S1: Monitor the regional smoke concentration of the billiard hall based on the real-time acquired environmental data stream and screen to obtain areas with excessive smoke. Then, perform temperature analysis and abnormal area identification on the areas with excessive smoke to obtain abnormal temperature areas and corresponding infrared area images.

[0021] It should be understood that the acquisition of environmental data streams is fundamental to this system, originating from a distributed sensor network deployed in the ceiling of the billiard hall. This network primarily includes three key data sources: a laser scattering smoke sensor array, a digital humidity sensor, and surveillance cameras with personnel counting capabilities. The smoke sensor continuously outputs a raw voltage signal at a sampling frequency of 10Hz, which is positively correlated with the concentration of particulate matter in the air. The infrared thermal imaging camera acquires 14-bit depth thermal radiation data at a rate of 25 frames per second. To ensure temporal and spatial consistency of this heterogeneous data, the system employs a GPS-disciplined clock module to provide a unified time reference for all sensing devices, achieving millisecond-level time synchronization. Subsequently, a sliding window buffer mechanism is used to align the data streams, and a mapping relationship between pixel coordinates and the actual physical space is established based on pre-calibrated camera parameters. The data set after time-space harmonic processing constitutes the aforementioned environmental data stream, which includes timestamp-aligned smoke concentration sequences, environmental humidity readings, and personnel density information calculated in real-time based on video analysis.

[0022] After obtaining a stable environmental data stream, to ensure a rapid response to potential fires and filter out recurring normal interference by incorporating historical patterns, a smoke concentration baseline for the corresponding monitoring area in the billiard hall is calculated based on the smoke concentration sequence. This application employs a sliding time window set composed of two different time scales: a short-term time window and a long-term time window, to calculate both the short-term and long-term smoke concentration baselines. The short-term smoke concentration baseline focuses on data changes within the last 30 seconds and is calculated using an exponentially weighted moving average algorithm. This algorithm assigns higher weight to recent data, enabling rapid response to sudden changes in smoke concentration. The long-term smoke concentration baseline focuses on historical data from the same time period over the past 24 hours, establishing a reference level by calculating its 75th percentile. This approach captures periodic concentration fluctuations in the billiard hall caused by daily operations (such as surges in customer traffic during specific periods). Furthermore, based on the aforementioned dual baselines, this application dynamically adjusts the smoke concentration threshold. The threshold adjustment in this application comprehensively considers the impact of real-time environmental humidity data and personnel density data. Since high humidity can cause water vapor condensation, interfering with the readings of the laser scattering sensor, the system appropriately raises the threshold to reduce false alarms when humidity exceeds 70%. The introduction of personnel density data addresses the potential dust disturbance and respiratory moisture effects caused by crowd gatherings. When the system detects a personnel density exceeding 0.5 people per square meter in a certain area via a camera, the threshold is also increased compensatorily. Specifically, the threshold generation operation in this application is performed by a linear weighted model. This model takes a short-term baseline, a long-term baseline, a humidity compensation term, and a personnel density compensation term as inputs, and its weight coefficients are determined through regression analysis of historical false alarm data. Through this adaptive threshold adjustment mechanism, the system can intelligently distinguish between normal smoking activities in a billiard hall and dangerous initial smoke from a fire. For example, it can effectively differentiate between low-concentration smoke from a few customers smoking and persistent high-concentration smoke from smoldering.

[0023] When the real-time smoke concentration data exceeds the smoke concentration threshold obtained from the above operations, the corresponding area is not immediately identified as a fire, but enters a pending verification state to prevent false alarms from the system. Areas marked as pending verification undergo more stringent continuous monitoring, typically with a preset monitoring period of 3 seconds. During this period, the system continuously collects smoke concentration data for the area at 100-millisecond intervals and compares it with the smoke concentration threshold. This short-term continuous monitoring strategy effectively filters out transient interference, such as a customer quickly lighting a match or briefly exhaling smoke rings, which usually only cause a momentary spike in smoke concentration rather than a sustained exceedance. Furthermore, the system makes a final determination based on the performance of the pending verification area throughout the complete monitoring period. Only when the real-time smoke concentration data for the area is consistently higher than the smoke concentration threshold at every sampling moment within the full 3-second monitoring period will the system ultimately identify the area as a smoke-exceeding area. This strict full-cycle exceedance judgment criterion can minimize false alarms caused by brief, intense human activity, transient sensor noise, or other random factors. For example, in a complex environment like a billiard hall, dust stirred up by customers waving their clothes or aerosol spray from opening carbonated drinks can trigger an instantaneous response from a smoke sensor, but rarely can it meet the stringent condition of continuous monitoring throughout the entire period and all sampling points exceeding the limit.

[0024] Next, infrared images corresponding to each area with excessive smoke are acquired. These images are derived from 14-bit depth thermal radiation data collected by a radiation-type infrared thermal imaging camera. The infrared images are transmitted to the central processing system at a rate of 25 frames per second, with each pixel representing the temperature reading at that location. Due to the inherent thermal noise of the infrared sensor and environmental electromagnetic interference, the original infrared images contain randomly distributed bright and dark spots, which can severely affect the accuracy of subsequent analysis. Therefore, median filtering is required for the original infrared images. This application uses a 3×3 pixel sliding window, sorting the temperature values ​​of the nine pixels within the window by size and then replacing the center pixel value with the median value. This nonlinear filtering method effectively suppresses impulse noise while preserving the temperature boundary information in the image. The processed output is a noise-reduced area image. In the billiard hall scenario, the above noise reduction operation is particularly important because temperature fluctuations at the air conditioning vents and thermal radiation interference from lighting equipment can create noise points in the original infrared images. Without filtering, these noise points may be misidentified as potential heat sources in subsequent analysis.

[0025] After obtaining high-quality denoised region images, this application employs a thermodynamic gradient enhancement algorithm to optimize the temperature gradient field in the image. The algorithm first calculates the horizontal and vertical temperature gradient components using the Sobel operator, and then performs a nonlinear transformation on the gradient magnitude. Specifically, after calculating the gradient magnitude for each pixel, an exponential enhancement function is used to scale the magnitude, highlighting strong temperature boundaries while suppressing weaker gradient changes. This gradient enhancement process effectively highlights the thermodynamic boundary between potential smoldering and the surrounding environment, which is particularly important for identifying initial smoldering phenomena. This is because smoldering areas typically appear as hot spots with slightly higher temperatures than the background but blurred boundaries, making them difficult to detect effectively using conventional gradient calculations. In the actual environment of a billiard hall, the temperature gradients of the human body's heat dissipation area and the actual smoldering area often overlap. The gradient enhancement algorithm can strengthen the persistent thermal boundary while weakening short-lived temperature fluctuations, thereby improving the ability to identify actual smoldering.

[0026] Based on the enhanced temperature gradient field image obtained above, this application's system performs threshold segmentation to extract candidate thermal anomaly regions. This application employs an adaptive dual-threshold mechanism to extract candidate thermal anomaly regions, where the high threshold is set to the 92nd quantile of the gradient magnitude distribution, and the low threshold is set to the 78th quantile. Specifically, firstly, all pixels with gradient magnitudes exceeding the high threshold are marked as strong edge points. Then, a region growing algorithm connects adjacent pixels with gradient magnitudes exceeding the low threshold to form complete candidate regions. This dual-threshold strategy ensures the integrity of the real thermal boundary while avoiding region fragmentation or excessive expansion caused by improper setting of a single threshold. In the complex thermal environment of a billiard hall, this segmentation method can effectively distinguish between the continuous thermal boundary of real fireworks and the fragmented temperature changes caused by human heat dissipation. After obtaining the set of candidate thermal anomaly regions, this application's system needs to further extract the temperature difference feature set for each region, which includes two core indicators: absolute temperature difference and relative temperature difference. The absolute temperature difference is calculated as the difference between the highest temperature within the candidate area and the average temperature of the surrounding background area, reflecting the absolute intensity of the heat source. The relative temperature difference is calculated as the ratio of the absolute temperature difference to the standard deviation of the background temperature, characterizing the significance of the heat source relative to environmental fluctuations. This temperature difference feature extraction is used to distinguish between real fireworks and normal heat sources. For example, in a billiard hall, heat sources generated by lighting fixtures typically have a high absolute temperature difference, but their relative temperature difference is often low because the background temperature around the fixtures is also relatively high and stable. Real fireworks, on the other hand, exhibit both high absolute and relative temperature difference characteristics.

[0027] Subsequently, a non-maximum suppression algorithm is used to fuse the candidate thermal anomaly regions, generating the final temperature anomaly region. First, a comprehensive significance score is calculated based on the temperature difference feature set of each candidate region. Then, each region is processed in descending order of score. For candidate regions with overlapping spatial locations and similar thermal characteristics, the region with the highest comprehensive significance score is retained, while duplicate regions with lower scores are suppressed. This fusion mechanism solves the problem of the same heat source being repeatedly detected at different scale levels in multi-scale analysis, ensuring that each real heat source corresponds to only one temperature anomaly region in the output. In the practical application scenario of a billiard hall, non-maximum suppression can effectively handle the problem of multiple adjacent candidate regions caused by thermal radiation diffusion, avoiding repeated alarms for the same pyrotechnics.

[0028] Finally, the enhanced temperature gradient field image corresponding to the temperature anomaly region is output as the representative infrared image of that region. The enhanced gradient image is chosen over the original infrared image because it more clearly displays the structural features and boundary information of the temperature anomaly region, which is crucial for subsequent shape analysis. For example, when distinguishing the irregular shape of a real flame from the regular shape of common heat sources in a billiard hall (such as billiard lamps and electronic displays), the boundary information provided by the gradient image is more discriminative than that of the original temperature image. This output strategy lays a solid foundation for subsequent smoke and fire shape analysis, ensuring that the entire smoke and fire detection system can verify the authenticity of potential fires from multiple dimensions.

[0029] Step S2: Perform edge fitting and rule evaluation on the infrared region image to obtain a list of potential fireworks in each temperature anomaly region.

[0030] To achieve accurate identification of smoke and fire in a billiard hall environment, after obtaining the temperature anomaly area and its infrared image, it is necessary to further perform heat source shape analysis to distinguish between real smoke and fire and conventional heat sources.

[0031] First, the infrared region image is processed for heat source identification and boundary optimization to generate a regional heat source image. Based on the principle of temperature threshold segmentation, this application initially identifies regions with pixel temperatures exceeding the average ambient temperature by more than 5 degrees Celsius as heat sources. Due to inherent noise and temperature diffusion effects in infrared imaging, the boundaries of the initially identified heat sources often exhibit burrs and voids. Therefore, morphological closing operations are used for boundary optimization. This operation sequentially performs dilation and erosion operations, using a 3×3 circular structuring element to traverse the entire image. The dilation operation fills small voids within the heat source and connects adjacent fractured areas, while the subsequent erosion operation restores the approximate size of the heat source and smooths the edge contours. This boundary optimization is particularly necessary in a billiard hall scene because the edges of real flames are typically irregularly jagged, while the boundaries of electronic device heat sinks are relatively regular. The optimized boundary more accurately reflects the essential shape characteristics of the heat source. After obtaining the optimized regional heat source image, the system performs heat source edge contour extraction to obtain a precise set of contour boundary points. This application employs a gradient tracing-based boundary detection method. Starting from any edge point in each heat source region, the method progressively traces along the decreasing temperature gradient direction to ultimately form a closed contour boundary. During the tracing process, the eight-neighborhood connectivity criterion is used to ensure the continuity and integrity of the contour. Each boundary point records its two-dimensional coordinate position, and all boundary points are stored as a point set sequence ordered clockwise. In complex thermal environments, such as those with multiple adjacent heat sources or partial shading, this contour extraction method can effectively separate different heat source entities, providing fundamental data for subsequent shape analysis.

[0032] Next, this embodiment employs the Douglas-Peucker polygon approximation algorithm to perform edge fitting on the extracted contour boundary point set, generating a simplified contour feature point sequence. Specifically, firstly, a baseline is formed by connecting the start and end points of the contour, and then the vertical distance from all intermediate points to this baseline is calculated. The point with the largest distance is found. If this distance is greater than a preset tolerance threshold (usually set to 2 pixels), this point is retained as a key feature point, and the contour is divided into two segments using this point as the boundary. The same processing is recursively applied. The final retained feature point sequence can retain the shape features of the original contour to the maximum extent with the fewest points. The above polygon approximation processing operation significantly reduces the complexity of subsequent calculations while retaining the key geometric characteristics of the contour. In practical applications in billiard halls, this simplification can effectively capture the random fluctuation characteristics of real flame contours, while the contours of regular heat sources such as lamps or electronic devices will exhibit more regular polygonal features.

[0033] Subsequently, based on the simplified contour feature point sequence obtained above, the system extracts a comprehensive geometric feature set from the regional heat source image. This geometric feature set includes four core geometric indices: rectangularity, calculated as the ratio of the heat source region's area to the area of ​​its smallest bounding rectangle, reflects the similarity between the heat source and a rectangle; roundness, calculated by multiplying the region's area by 4π and then dividing by the square of the perimeter, characterizes the heat source's approximation of a circle; elongation, defined as the ratio of the heat source region's major axis length to its minor axis length, describes the heat source's elongated characteristics; and edge irregularity, measured by the ratio of the actual contour perimeter to the equivalent circle's circumference, reflects the complexity of the boundary contour. These geometric features collectively constitute a digital description of the heat source's shape, effectively distinguishing different types of heat-generating bodies. For example, the radiators used in billiard halls typically exhibit high rectangularity (close to 1.0) and low irregularity (close to 1.0), while real flames often exhibit low rectangularity (less than 0.6) and high irregularity (greater than 1.3). Furthermore, this application employs a pre-trained gradient boosting decision tree as a classification model to analyze the geometric feature set and output a probability score for the shape anomaly of each heat source. It should be understood that the classification model used in this application is trained on a training set containing multiple known heat source types, with training samples covering common heat source categories such as real flames, electronic devices, and human body heat dissipation. The model receives a four-dimensional geometric feature vector as input, undergoes multiple rounds of decision tree ensemble calculations, and finally outputs a probability value between 0 and 1, representing the probability that the heat source belongs to an anomalous shape (i.e., real fireworks). The gradient boosting decision tree model used in this application's embodiments can automatically learn the nonlinear relationships between different geometric features and automatically weight the importance of features. In the complex thermal environment of a billiard hall, this machine learning method has stronger adaptability and discriminative ability than fixed threshold rules, and can identify heat sources that are not obvious in a single feature but exhibit anomalies in feature combinations.

[0034] Finally, based on the preset fire probability threshold and the obtained shape anomaly probability score, each heat source is classified and screened. In this embodiment, the fire probability threshold is a two-tiered boundary: a high threshold of 0.7 and a low threshold of 0.4. Heat sources with a shape anomaly probability score higher than the high threshold are directly marked as confirmed fire candidates, indicating that these heat sources have a high fire risk (i.e., these heat sources have significant abnormal shape characteristics) and need to immediately proceed to the subsequent verification process; heat sources with scores between the low and high thresholds are marked as observed fire candidates (i.e., these heat sources exhibit certain abnormal characteristics but are not obvious), requiring further verification; those with scores below the low threshold are excluded from the candidate list. This hierarchical screening mechanism ensures timely identification of high-risk heat sources and provides an opportunity for verification of boundary conditions. In the actual operation of billiard halls, this classification strategy can effectively distinguish between different risk levels, such as cigarette butts (medium probability) and real flames (high probability), achieving precise early warning management. Finally, the confirmed and observed fireworks candidates are combined as potential fireworks into a standardized data structure (i.e., a potential fireworks list). Each entry in the potential fireworks list contains a unique identifier for the heat source, spatial coordinates, a shape anomaly probability score, a classification category label, and a corresponding geometric feature vector. This structured data organization ensures that each heat source candidate in the list can be accurately identified and retrieved by subsequent processing modules. In the practical application scenario of a billiard hall, this hierarchical list achieves the goal of precise early warning: confirmed fireworks candidates can trigger a higher-level warning response, while observed fireworks candidates adopt a lower-priority monitoring strategy. For example, a confirmed fireworks candidate with a probability score of 0.85 will immediately initiate a rapid verification process, while an observed fireworks candidate with a probability score of 0.55 will enter the regular monitoring queue.

[0035] Step S3: According to the preset fireworks feature classification rules, identify the type of each potential fireworks in the potential fireworks list to obtain the fireworks type of each potential fireworks.

[0036] After establishing the potential fireworks list, further fireworks type identification is required to distinguish heat sources of different natures. This application achieves accurate classification by fusing visible light and infrared features, and its complete implementation process is shown below: It should be understood that the fireworks feature classification rule adopted in this application is a feature template library that has undergone rigorous training. This rule is constructed based on a large amount of labeled fireworks sample data and includes feature subsets corresponding to four main types of fireworks: open flame feature subset, smoldering feature subset, high-temperature object feature subset, and false alarm source feature subset. Each type feature subset consists of corresponding photothermal feature templates. These templates are obtained by collecting visible light and infrared data of various types of fireworks in a real billiard hall environment, and then performing feature extraction and cluster analysis. The role of the fireworks feature classification rule is to provide a comparison benchmark for subsequent type identification, enabling the system to calculate the similarity between the feature patterns of unknown heat sources detected in real time and known types, thereby determining its most likely category.

[0037] First, visible light images are acquired based on the identified temperature anomaly areas. These images are obtained from a high-definition color camera coaxially mounted with an infrared camera, capturing RGB images at a resolution of 1920×1080 at a rate of 25 frames per second. Due to the complex and variable lighting conditions in the billiard hall, the system first performs automatic white balance and exposure compensation processing on the visible light images to eliminate the influence of color cast and uneven brightness on color feature extraction. Based on the spatial coordinates recorded in the list of potential fireworks, image regions corresponding to each potential firework are extracted from the corrected visible light images. Color space feature sets are extracted from these regions. These feature sets include three dimensions of color information: the histogram statistical features of the hue channel reflect the overall color distribution pattern of the heat source; the mean and variance of the saturation channel characterize the purity and stability of the color; and the gradient features of the brightness channel describe the brightness variation pattern. In the billiard hall environment, this multi-dimensional color feature effectively distinguishes different types of heat sources; for example, candle flames exhibit a continuous warm color distribution, while electronic device indicator lights may display discrete single color blocks. Simultaneously, thermodynamic feature sets for each potential fire are extracted from the regional heat source image. These feature sets are constructed based on temperature change information within the regional heat source image and include both dynamic and static features. Dynamic features primarily describe the temperature change over time, including the temperature rise slope, temperature fluctuation frequency, and temperature change consistency within the last 3 seconds. Static features characterize the spatial distribution of temperature, including temperature uniformity within the heat source, edge temperature decay gradient, and spatial correlation of thermal radiation. The obtained thermodynamic features can capture the typical thermal behavior patterns of different potential fires; for example, open flames typically exhibit rapid and unstable temperature fluctuations, while electrical appliance heat dissipation shows a slow and stable temperature rise trend.

[0038] Furthermore, the obtained color space feature set and thermodynamic feature set are combined to form a photothermal feature set corresponding to each potential firework. This application employs a feature-level fusion strategy for the combination: first, the two types of features are normalized to eliminate dimensional differences, and then they are concatenated into a high-dimensional feature vector. The normalization process uses the z-score standardization method to convert the value of each feature into a standard distribution with a mean of 0 and a variance of 1. The fused photothermal feature set comprehensively reflects the optical appearance and thermodynamic behavior characteristics of the firework, forming a comprehensive description of the potential firework. In the complex environment of a billiard hall, this multimodal feature fusion can compensate for the deficiencies of single-modal information. For example, some high-temperature objects may have color characteristics similar to real flames, but their thermodynamic characteristics may differ significantly.

[0039] Subsequently, the similarity between the photothermal feature set and each feature subset of the fireworks type feature set is calculated. In this embodiment, the similarity calculation employs an improved cosine similarity metric, which considers not only the directional consistency of feature vectors but also the proportional relationship of vector magnitudes. For each potential fireworks feature set, the similarity score between its photothermal feature set and the center vector of the four fireworks type feature subsets is calculated. Since the spatial distribution density of different types of fireworks features varies, the system also performs a calibration process on the similarity scores based on category prior probabilities. These similarity scores are normalized using a softmax function and transformed into probability values ​​for each type, forming a complete type probability distribution. This similarity-based probability calculation accurately reflects the degree of matching between unknown heat sources and various standard fireworks, providing a quantitative basis for the final classification decision.

[0040] Finally, the obtained probability distribution of types is analyzed using preset firework feature classification rules to determine the firework type of each potential firework. This application employs a multi-level decision-making logic to analyze the type probability distribution: first, it checks whether the highest probability value exceeds a confidence threshold of 0.75; those that meet the condition are directly identified as the corresponding type. When the probability values ​​of multiple types are similar and none reach the confidence threshold, the system activates an auxiliary decision-making mechanism based on feature importance, prioritizing types that show significant discriminative features. The final output firework type includes a clear category label and classification confidence level, where the category label adopts a four-level classification system: open flame type represents a visible flame that is burning, smoldering type represents a slow burning phenomenon without an open flame, high-temperature object type refers to normal heating equipment, and false alarm type covers all non-firework heat sources. This refined type identification enables the system to adopt differentiated processing strategies for firework of different natures. For example, it immediately triggers an emergency alarm for open flame type, strengthens continuous monitoring for smoldering type, and records feature patterns and updates the whitelist for high-temperature object type.

[0041] Step S4: When the type of smoke and fire is a preset set of fire alarm types, perform smoke movement analysis and scene recognition verification on the potential smoke and fire, and generate alarm decision data for each temperature abnormality area.

[0042] After identifying the smoke type, if the identified smoke type belongs to a preset fire alarm type set, further smoke motion analysis and scene recognition verification are required to confirm the fire risk level. The preset fire alarm type set includes two fire modes requiring high attention: open flame and smoldering, while excluding high-temperature object types and false alarm types from the deep verification scope. This application generates reliable alarm decision data through multi-dimensional evidence fusion, and its complete implementation process is shown below: First, this application performs dense optical flow field analysis on the obtained visible light images. The application uses the Farneback algorithm to calculate the motion vector of each pixel between adjacent frames, establishing a complete two-dimensional flow field. Since the motion of smoke particles has specific hydrodynamic characteristics, the system extracts a particle flow feature set from the optical flow field. This feature set includes three core indicators: motion direction dispersion, velocity distribution entropy, and vortex intensity. Motion direction dispersion is characterized by calculating the variance of the motion direction within a local area; real smoke typically exhibits high directional dispersion. Velocity distribution entropy describes the randomness of the motion velocity; smoke areas often have high velocity entropy values. Vortex intensity quantifies the rotational motion characteristics by calculating the curl of the flow field, which is particularly important for identifying thermal convection phenomena caused by fires. In a billiard hall environment, this refined motion analysis can effectively distinguish the turbulent characteristics of real smoke from the regular motion patterns caused by people walking; for example, the optical flow field generated by customers walking typically exhibits low directional dispersion and velocity distribution entropy.

[0043] Simultaneously, foreground detection processing is performed on the visible light image to extract background color and texture features. This application employs an improved Gaussian mixture model, which separates the foreground moving region from the static background by establishing a color and texture distribution model for each pixel. The obtained background color and texture feature set includes three dimensions: color consistency, texture complexity, and edge preservation. Color consistency is measured by calculating the Bartlett distance between the foreground and background regions in the HSV color space; texture complexity is characterized by local binary mode variance; and edge preservation is evaluated by comparing the overlap ratio between foreground and background edges. The background color and texture features obtained above help the system distinguish the semi-transparent characteristics of real smoke from other opaque moving objects. For example, in a billiard hall, smoke areas typically exhibit low color consistency and high edge preservation, while human movement areas show the opposite feature pattern.

[0044] Subsequently, a comprehensive smoke motion analysis is performed based on the obtained particle flow feature set and background color and texture features to identify smoke movement regions. This application employs a random forest-based classifier to combine the particle flow feature set and background color and texture features into a 10-dimensional feature vector, outputting the probability that each movement region belongs to real smoke. For regions with a probability exceeding 0.65, the system further extracts its smoke flow feature set, which includes three key parameters: rising velocity component, diffusion rate, and stability index. The rising velocity component is calculated by analyzing the vertical velocity distribution of the optical flow field, reflecting the rising characteristics of the thermal plume; the diffusion rate is quantified by measuring the rate of change of the movement region's area; and the stability index characterizes the persistence of the movement pattern. This smoke flow feature set allows for the capture of unique smoke movement patterns. For example, candle smoke typically exhibits stable rising motion and slow diffusion characteristics, while smoke generated by electrical short circuits may exhibit rapid diffusion and unstable movement patterns.

[0045] Simultaneously, based on preset hazardous material types, the system performs semantic scene object recognition and smoke / fire correlation calculation on visible light images. Hazardous material types include flammable materials (such as paper and fabric), electrical equipment (such as chargers and sockets), and heat source containers (such as ashtrays and trash cans). This application's system uses a DeepLabV3+ semantic segmentation network trained on a dedicated dataset for billiard hall scenes to perform pixel-level semantic annotation on visible light images. Based on the segmentation results, the spatial correlation index between each potential smoke / fire and hazardous material is calculated. This index includes three components: distance correlation, directional correlation, and occlusion correlation. Distance correlation is obtained by calculating the nearest spatial distance between the smoke / fire and the hazardous material and normalizing it using an exponential function; directional correlation analyzes the orientation relationship of the smoke / fire relative to the hazardous material, considering the propagation direction characteristics of heat radiation; and occlusion correlation assesses the degree of visual occlusion between the two. Through the above spatial correlation analysis, high-risk scene configurations can be identified. For example, when open flame-type smoke and flammable paper-type materials are detected coexisting at close range, a high spatial correlation index will be generated.

[0046] Furthermore, the data obtained above is integrated through a multi-evidence fusion algorithm to calculate the corresponding fire threat confidence score. The multi-evidence fusion algorithm used in this application employs the DS evidence theory framework, treating the three evidence sources as independent trust functions. It should be understood that the spatial correlation index provides prior environmental risk evidence, the smoke movement area contributes direct evidence of motion characteristics, and the smoke flow feature set provides supplementary evidence of hydrodynamic properties. Each evidence source is assigned a different weight coefficient based on its reliability in historical data, and the joint trust score is calculated through the evidence combination rules in the framework. The final output fire threat confidence score is a continuous value between 0 and 1, with a higher value indicating a greater fire risk. This multi-evidence fusion mechanism can effectively handle uncertain situations; for example, when a smoke source exhibits weak motion characteristics but has a very high spatial correlation with hazardous materials, an appropriate threat assessment result can still be obtained.

[0047] Finally, the fire threat confidence score is analyzed using confidence threshold level rules to generate structured alarm decision data. The confidence threshold level rules in this application divide the confidence score into four levels: above 0.8 for emergency level, 0.6 to 0.8 for high-risk level, 0.4 to 0.6 for medium-risk level, and below 0.4 for low-risk level. Based on the score level, the system encapsulates spatial correlation indicators, smoke movement area boundary coordinates, and smoke flow feature sets into a standardized alarm decision data package. This data package is organized in JSON format and includes threat level labels, confidence scores, timestamps, spatial location information, and detailed parameters for each piece of evidence. In practical applications in billiard halls, this hierarchical decision-making mechanism enables precise emergency response. For example, it immediately activates audible and visual alarms and notifies the fire department for emergency-level smoke and fire, triggers on-site warnings and requires manual confirmation for high-risk-level smoke and fire, and logs and continuously monitors medium- and low-risk-level smoke and fire.

[0048] Step S5: Using a preset hierarchical alarm method, an alarm is triggered for the temperature abnormality area based on the alarm decision data.

[0049] After generating alarm decision data, a tiered alarm mechanism needs to be implemented to ensure precise emergency response. This application ensures appropriate measures are taken for different types of fire risks through intelligent alarm strategy selection and multi-faceted notification execution. The complete implementation process is shown below: It should be understood that the core of the tiered alarm system adopted in this application lies in establishing an alarm strategy library. This library contains four levels of alarm plans, each corresponding to a threat level classification in the alarm decision data. The emergency-level alarm plan applies to fires with a confidence score of 0.8 or higher, initiating a full emergency response; the high-risk-level alarm plan addresses situations with a confidence score of 0.6 to 0.8, implementing localized early warning measures; the medium-risk-level alarm plan handles potential risks with a confidence score of 0.4 to 0.6, implementing preventative monitoring; and the low-risk-level alarm plan addresses events with a confidence score below 0.4, only requiring log recording. Each alarm plan specifies in detail the triggering mode of the audible and visual alarms, the content template of the notification message, the list of recipients, and the emergency equipment linkage scheme. In the actual operation of the billiard hall, this tiered plan ensures that the response measures are precisely matched to the risk level, avoiding unnecessary interference with normal operations.

[0050] First, based on the threat level labels in the alarm decision data, this application's system retrieves the corresponding alarm plans from the alarm strategy library and generates a specific alarm instruction set. This retrieval process employs a hash table-based fast lookup mechanism, completing a match within milliseconds using the threat level as the keyword. The generated alarm instruction set includes three categories: device control instructions, message push instructions, and system linkage instructions. Device control instructions specify the activation duration, flashing frequency, and tone mode of the audible and visual alarms; message push instructions define the priority of the notification content, the recipients, and the transmission channels; and system linkage instructions clearly define the auxiliary equipment to be activated and its operating parameters. This instruction-based alarm description makes the execution process highly operable. For example, for an emergency fire, the instruction set will explicitly require the activation of the entire hall's audible and visual alarms and simultaneous notification to the fire department, while for a medium-risk fire, only a visual prompt from the management end will be triggered.

[0051] Subsequently, the alarm instruction distribution module in the system of this application is responsible for transmitting the instruction set to each execution terminal. This module uses a message queue mechanism to ensure the reliable delivery of instructions. Instruction distribution follows the principle of proximity and load balancing strategy, and intelligently allocates tasks according to the current status of each execution terminal. For the audible and visual alarm devices, control instructions are directly sent through the Modbus-RTU protocol; for mobile terminals, structured messages are pushed using the MQTT protocol; for third-party systems, linkage requests are transmitted through the RESTful API interface. All instruction transmissions use TLS encryption to ensure security and are accompanied by digital signatures to prevent tampering. In the complex network environment of the billiards hall, this multi-protocol adaptation distribution mechanism can ensure that alarm instructions can be accurately delivered to the target devices under various conditions. The multi-channel alarm execution system synchronously drives various terminal devices to complete the alarm actions. The audible and visual alarm channel controls the alarms installed in each area of the billiards hall to generate specific audible and visual signals according to the instruction requirements. For the emergency-level alarm, a continuous sharp tone of 105 decibels is used in combination with a red strobe light; for the high-risk level, an intermittent tone of 95 decibels is used in combination with an orange warning; for the medium-risk level, a reminder tone of 75 decibels is used in combination with a yellow indicator light. The mobile notification channel pushes classified warning information to the smartphones of the management personnel. The emergency-level information includes vibration reminders and full-screen displays. The high-risk-level information is displayed at the top of the notification bar. The medium-risk-level information is included in the regular message list. The system linkage channel activates relevant emergency devices, including ventilation system control, access release, and emergency lighting activation, etc. This multi-channel collaborative alarm ensures that attention can be effectively drawn in any situation. For example, even in an area with relatively high background noise, the complementary use of audible and visual and mobile notifications can still ensure that the alarm information is promptly perceived.

[0052] The alarm feedback collection mechanism monitors the status of each execution terminal in real time to confirm the completion of the alarm actions. The system collects feedback data through three methods: device status feedback, message read confirmation, and manual response records. The audible and visual alarm devices return their working status once every second, including current values, volume, and light source intensity; the mobile notification system tracks the delivery status and reading time of messages; on-site personnel can send confirmation responses through the emergency button. All feedback data is accompanied by accurate timestamps to establish a corresponding relationship with the original alarm instructions. This closed-loop feedback mechanism can promptly detect execution anomalies. For example, when a certain audible and visual alarm fails to be activated as expected, the system will automatically attempt to use backup devices or upgrade the notification method.

[0053] Based on collected feedback data, this application's system performs dynamic alarm adjustments to optimize response performance. The adjustment strategies include alarm intensity adjustment, notification range expansion, and execution mode switching. When the system detects high population density in an area, it automatically increases the intensity of the audible and visual alarms; when the primary contact fails to respond promptly, notifications are sent sequentially to backup contacts; and when network communication is interrupted, alarm information is sent via SMS. All decisions during the adjustment process are recorded in the audit log, including the reason for the adjustment, the execution time, and the resulting effects. This adaptive adjustment mechanism ensures the alarm system maintains reliability under various abnormal conditions. For example, when an event is held in a billiard hall causing increased environmental noise, the system automatically increases the volume of the audible alarm to ensure penetration.

[0054] The alarm event archiving module in this application system records the complete alarm process to a distributed database, forming a traceable event archive. Each archive contains alarm decision data, the alarm plan used, the generated instruction set, the execution status of each terminal, feedback collection records, and dynamic adjustment logs. The archive data is stored in a time-series format, supporting multi-dimensional retrieval by time range, regional location, and event type. All sensitive data is encrypted, and access requires identity authentication and permission verification. These historical archives are not only used for post-event analysis but also provide data support for system optimization. For example, by analyzing the common characteristics of multiple false alarm events, the parameter settings of the detection algorithm can be improved.

[0055] The aforementioned tiered alarm process establishes a reliable emergency response system. This system fully leverages the multi-channel collaborative advantages of modern communication technologies and intelligent devices to ensure that fire warning information is accurately and promptly delivered to relevant personnel under any circumstances. This refined alarm management guarantees rapid response to actual fires while minimizing the impact on the billiard hall's normal operations, achieving an optimal balance between safety and practicality. By continuously collecting execution feedback and optimizing alarm strategies, the system can continuously improve itself, maintaining high reliability and adaptability during long-term operation.

[0056] Please see Figure 2 , Figure 2 This is a schematic diagram of a billiard hall smoke detection and early warning notification device provided in an embodiment of this application. Figure 2 As shown, the billiard hall smoke detection and early warning notification device 2 includes: a smoke screening module 21, an anomaly recognition module 22, a smoke detection module 23, a smoke classification module 24, a fire alarm verification module 25, and a graded alarm module 26.

[0057] The smoke screening module 21 is used to monitor and screen the area smoke concentration in the billiard hall based on the real-time acquired environmental data stream, and to identify areas where smoke exceeds the standard. Anomaly identification module 22 is used to perform temperature analysis and anomaly identification on the smoke-exceeding area, and obtain temperature anomaly area and corresponding infrared area image; The smoke and fire detection module 23 is used to perform edge fitting and rule evaluation of the heat source shape on the infrared region image to obtain a list of potential smoke and fire in each temperature anomaly region. The fireworks classification module 24 is used to identify the type of each potential firework in the potential fireworks list according to the preset fireworks feature classification rules, and obtain the fireworks type of each potential firework. The fire alarm verification module 25 is used to perform smoke movement analysis and scene recognition verification on the potential smoke when the smoke type is a preset fire alarm type set, and generate alarm decision data for each temperature abnormal area. The graded alarm module 26 is used to issue an alarm to the temperature abnormality area according to the alarm decision data through a preset graded alarm method.

[0058] In one possible implementation, the smoke filtering module 21 is specifically used for: Based on the smoke concentration sequence, the short-term and long-term smoke concentration baselines of each monitoring area in the billiard hall are calculated using a preset sliding time window set. Based on environmental humidity data, personnel density data, the short-term smoke concentration baseline, and the long-term smoke concentration baseline, the smoke concentration threshold is adjusted to generate the smoke concentration threshold for each monitoring area. When the real-time smoke concentration data in the smoke concentration sequence is greater than the smoke concentration threshold, the corresponding monitoring area is determined as the area to be verified, and the area to be verified is continuously monitored for smoke concentration according to the preset monitoring period and the smoke concentration threshold. When the real-time smoke concentration data within the monitoring period is greater than the smoke concentration threshold, the corresponding area to be verified is determined as a smoke exceeding the standard area.

[0059] In one possible implementation, the anomaly detection module 22 is specifically used for: Infrared images of the area with excessive smoke are acquired, and median filtering is applied to the infrared images to obtain noise-reduced area images. By using a preset thermodynamic gradient enhancement algorithm, the temperature gradient of the noise reduction region image is optimized to obtain an enhanced temperature gradient field image. The enhanced temperature gradient field image is segmented according to a preset temperature gradient threshold to obtain a set of candidate thermal anomaly regions; Based on the candidate thermal anomaly region set, the temperature difference feature extraction process between adjacent regions of the enhanced temperature gradient field image is performed to obtain the temperature difference feature set of each candidate thermal anomaly region. By using a preset non-maximum suppression method based on the temperature difference feature set, the candidate thermal anomaly regions in the candidate thermal anomaly region set are fused to generate a temperature anomaly region. The enhanced temperature gradient field image corresponding to the temperature anomaly region is used as the infrared region image corresponding to the temperature anomaly region.

[0060] In one possible implementation, the smoke detection module 23 is specifically used for: The infrared region image is subjected to heat source identification and heat source boundary optimization processing to obtain a regional heat source image; Heat source edge contours are extracted from the heat source image of the region to obtain the contour boundary point set of each heat source; The contour boundary point set is edge-fitted by a preset polygon approximation algorithm to obtain the contour feature point sequence of each heat source. Based on the contour feature point sequence, extract the geometric feature set of each heat source from the regional heat source image; Based on the geometric feature set, the geometric feature set is used to evaluate the shape regularity of each heat source in the temperature anomaly region and obtain the shape anomaly probability score of each heat source. Based on the preset fire probability threshold and the shape anomaly probability score, the heat sources are classified and screened to obtain a list of potential fires in each temperature anomaly area.

[0061] In one possible implementation, the fireworks classification module 24 is specifically used for: Acquire a visible light image of the temperature anomaly region, extract the color space feature set of each potential firework from the visible light image according to the potential firework list, and extract the thermodynamic feature set of each potential firework from the heat source image of the region according to the potential firework list; The color space feature set and the thermodynamic feature set are combined to form the photothermal feature set corresponding to potential fireworks; Calculate the similarity between the photothermal feature set and each type feature subset in the preset fireworks type feature set to obtain the type probability distribution of each potential fireworks; Based on the type probability distribution, the potential fireworks are identified using preset fireworks feature classification rules to determine the type of each potential fireworks.

[0062] In one possible implementation, the fire alarm verification module 25 is specifically used for: Dense optical flow field analysis was performed on the visible light images corresponding to the potential fireworks to extract the particle flow feature set of each temperature anomaly region; Foreground detection processing is performed on the visible light image to extract background color and texture features of each temperature anomaly region; Based on the particle flow feature set and the background color texture features, the visible light image is analyzed for the smoke motion of each potential firework to obtain the smoke motion region of each potential firework and the corresponding smoke flow feature set. Based on the preset hazardous material types, the visible light image is used to perform semantic scene object recognition and smoke correlation calculation to obtain the spatial correlation index between each potential smoke and hazardous material. Using a pre-defined multi-evidence fusion algorithm, a fire threat confidence score is calculated based on the spatial correlation index, the smoke movement area, and the smoke flow feature set. Based on the fire threat confidence score and the preset confidence threshold level rules, the spatial correlation index, the smoke movement area, and the smoke flow feature set are encapsulated into alarm decision data for each temperature anomaly area.

[0063] The smoke screening module 21, anomaly detection module 22, smoke and fire detection module 23, smoke and fire classification module 24, fire alarm verification module 25, and graded alarm module 26 can all be implemented in software or hardware. For example, the implementation of the smoke screening module 21 will be described below. Similarly, the implementation methods of the anomaly detection module 22, smoke and fire detection module 23, smoke and fire classification module 24, fire alarm verification module 25, and graded alarm module 26 can refer to the implementation method of the smoke screening module 21.

[0064] As an example of a software functional unit, the smoke filtering module 21 may include code running on a computing instance. The computing instance may include at least one of a physical host (computing device), a virtual machine, or a container. Further, the aforementioned computing instance may be one or more. For example, the smoke filtering module 21 may include code running on multiple hosts / virtual machines / containers. It should be noted that the multiple hosts / virtual machines / containers used to run the code may be distributed in the same region or in different regions. Further, the multiple hosts / virtual machines / containers used to run the code may be distributed in the same availability zone (AZ) or in different AZs, each AZ including one or more geographically proximate data centers. Typically, a region may include multiple AZs.

[0065] Similarly, multiple hosts / virtual machines / containers used to run this code can be distributed within the same Virtual Private Cloud (VPC) or across multiple VPCs. Typically, a VPC is set up within a region. Communication between two VPCs within the same region, as well as between VPCs in different regions, requires a communication gateway to be set up within each VPC to enable interconnection between VPCs.

[0066] As an example of a hardware functional unit, the smoke filtering module 21 may include at least one computing device, such as a server. Alternatively, the smoke filtering module 21 may also be a device implemented using an application-specific integrated circuit (ASIC) or a programmable logic device (PLD). The PLD may be implemented using a complex programmable logical device (CPLD), a field-programmable gate array (FPGA), generic array logic (GAL), or any combination thereof.

[0067] The smoke filtering module 21 includes multiple computing devices that can be distributed in the same region or in different regions. Similarly, the smoke filtering module 21 can be distributed within the same Availability Zone (AZ) or in different AZs. Likewise, the smoke filtering module 21 can be distributed within the same Virtual Private Cloud (VPC) or in multiple VPCs. These multiple computing devices can be any combination of computing devices such as servers, ASICs, PLDs, CPLDs, FPGAs, and GALs.

[0068] See Figure 3 As shown, Figure 3 This is a schematic diagram of a computing device provided in this application. The computing device 100 includes: a processor 104, a communication interface 108, a bus 102, and a memory 106. The processor 104, the communication interface 108, and the memory 106 communicate via the bus 102. In practical applications, communication can also be achieved through other means such as wireless transmission; however, this is not limited here.

[0069] The computing device 100 may be a server or a terminal device. It should be understood that this application does not limit the number of processors and memory in the computing device 100.

[0070] The processor 104 may include any one or more processors such as a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0071] The communication interface 108 uses transceiver modules such as, but not limited to, network interface cards and transceivers to enable communication between the computing device 100 and other devices or communication networks.

[0072] Bus 102 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 3 The bus 102 may be represented by a single line, but this does not mean that there is only one bus or one type of bus. The bus 102 may include a path for transmitting information between various components of the computing device 100 (e.g., memory 106, processor 104, communication interface 108).

[0073] Memory 106 may include volatile memory, such as random access memory (RAM). Memory 106 may also include non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (SSD).

[0074] The memory 106 stores executable program code, and the processor 104 executes the executable program code to implement the functions of the aforementioned smoke screening module 21, anomaly recognition module 22, smoke and fire detection module 23, smoke and fire classification module 24, fire alarm verification module 25, and graded alarm module 26, thereby realizing the billiard hall smoke and fire detection early warning notification method. That is, the memory 106 stores instructions for executing the billiard hall smoke and fire detection early warning notification method.

[0075] This application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that a computing device can store, or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive). The computer-readable storage medium includes instructions that instruct the computing device to execute a billiard hall fireworks detection and early warning notification method, or instruct the computing device to execute a billiard hall fireworks detection and early warning notification method.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting and issuing early warnings of fireworks in a billiard hall, characterized in that, The method includes: Based on the real-time acquired environmental data stream, the billiard hall is monitored for regional smoke concentration and screened to identify areas with excessive smoke. Temperature analysis and abnormal area identification are then performed on these areas to obtain abnormal temperature areas and corresponding infrared images. Edge fitting and rule evaluation of heat source shapes are performed on the infrared region image to obtain a list of potential fireworks in each temperature anomaly region. According to the preset fireworks feature classification rules, the type of each potential fireworks in the potential fireworks list is identified to obtain the fireworks type of each potential fireworks. When the type of smoke and fire is a preset set of fire alarm types, smoke movement analysis and scene recognition verification are performed on the potential smoke and fire to generate alarm decision data for each temperature anomaly area. The alarm is triggered in the temperature anomaly area based on the alarm decision data using a preset tiered alarm method.

2. The billiard hall smoke detection and early warning notification method according to claim 1, wherein the environmental data stream includes smoke concentration sequence, environmental humidity data, and personnel density data, characterized in that, The method of monitoring and screening the regional smoke concentration in the billiard hall based on real-time acquired environmental data streams to identify areas with excessive smoke includes: Based on the smoke concentration sequence, the short-term and long-term smoke concentration baselines of each monitoring area in the billiard hall are calculated using a preset sliding time window set. Based on environmental humidity data, personnel density data, the short-term smoke concentration baseline, and the long-term smoke concentration baseline, the smoke concentration threshold is adjusted to generate the smoke concentration threshold for each monitoring area. When the real-time smoke concentration data in the smoke concentration sequence is greater than the smoke concentration threshold, the corresponding monitoring area is determined as the area to be verified, and the area to be verified is continuously monitored for smoke concentration according to the preset monitoring period and the smoke concentration threshold. When the real-time smoke concentration data within the monitoring period is greater than the smoke concentration threshold, the corresponding area to be verified is determined as a smoke exceeding the standard area.

3. The method for detecting and issuing early warnings of fireworks in a billiard hall according to claim 1, characterized in that, The step of performing temperature analysis and abnormal region identification on the smoke-exceeding area to obtain temperature abnormality areas and corresponding infrared region images includes: Infrared images of the area with excessive smoke are acquired, and median filtering is applied to the infrared images to obtain noise-reduced area images. By using a preset thermodynamic gradient enhancement algorithm, the temperature gradient of the noise reduction region image is optimized to obtain an enhanced temperature gradient field image. The enhanced temperature gradient field image is segmented according to a preset temperature gradient threshold to obtain a set of candidate thermal anomaly regions; Based on the candidate thermal anomaly region set, the temperature difference feature extraction process between adjacent regions of the enhanced temperature gradient field image is performed to obtain the temperature difference feature set of each candidate thermal anomaly region. By using a preset non-maximum suppression method based on the temperature difference feature set, the candidate thermal anomaly regions in the candidate thermal anomaly region set are fused to generate a temperature anomaly region. The enhanced temperature gradient field image corresponding to the temperature anomaly region is used as the infrared region image corresponding to the temperature anomaly region.

4. The method for detecting and issuing early warnings of fireworks in a billiard hall according to claim 1, characterized in that, The step of performing edge fitting and rule evaluation on the infrared region image to obtain a list of potential fireworks in each temperature anomaly region includes: The infrared region image is subjected to heat source identification and heat source boundary optimization processing to obtain a regional heat source image; Heat source edge contours are extracted from the heat source image of the region to obtain the contour boundary point set of each heat source; The contour boundary point set is edge-fitted by a preset polygon approximation algorithm to obtain the contour feature point sequence of each heat source. Based on the contour feature point sequence, extract the geometric feature set of each heat source from the regional heat source image; Based on the geometric feature set, the geometric feature set is used to evaluate the shape regularity of each heat source in the temperature anomaly region and obtain the shape anomaly probability score of each heat source. Based on the preset fire probability threshold and the shape anomaly probability score, the heat sources are classified and screened to obtain a list of potential fires in each temperature anomaly area.

5. The method for detecting and issuing early warnings of fireworks in a billiard hall according to claim 4, wherein the fireworks feature classification rules include a fireworks type feature set, characterized in that, The step of identifying the type of each potential firework in the potential firework list according to the preset firework feature classification rules, and obtaining the firework type of each potential firework, includes: Acquire a visible light image of the temperature anomaly region, extract the color space feature set of each potential firework from the visible light image according to the potential firework list, and extract the thermodynamic feature set of each potential firework from the heat source image of the region according to the potential firework list; The color space feature set and the thermodynamic feature set are combined to form the photothermal feature set corresponding to potential fireworks; Calculate the similarity between the photothermal feature set and each type feature subset in the preset fireworks type feature set to obtain the type probability distribution of each potential fireworks; Based on the type probability distribution, the potential fireworks are identified using preset fireworks feature classification rules to determine the type of each potential fireworks.

6. The method for detecting and issuing early warnings of fireworks in a billiard hall according to claim 5, characterized in that, The process of analyzing smoke movement and verifying scene recognition of the potential fireworks to generate alarm decision data for each temperature anomaly area includes: Dense optical flow field analysis was performed on the visible light images corresponding to the potential fireworks to extract the particle flow feature set of each temperature anomaly region; Foreground detection processing is performed on the visible light image to extract background color and texture features of each temperature anomaly region; Based on the particle flow feature set and the background color texture features, the smoke motion of each potential firework is analyzed in the visible light image to obtain the smoke motion region of each potential firework and the corresponding smoke flow feature set. Based on the preset hazardous material types, the visible light image is used to perform semantic scene object recognition and smoke correlation calculation to obtain the spatial correlation index between each potential smoke and hazardous material. Using a pre-defined multi-evidence fusion algorithm, a fire threat confidence score is calculated based on the spatial correlation index, the smoke movement area, and the smoke flow feature set. Based on the fire threat confidence score, the spatial correlation index, the smoke movement area, and the smoke flow feature set are encapsulated into alarm decision data for each temperature anomaly area according to the preset confidence threshold level rules.

7. A billiard hall fire detection and early warning notification device, applied to the billiard hall fire detection and early warning notification method according to claim 1, characterized in that, The device includes: The smoke screening module is used to monitor and screen the area smoke concentration in the billiard hall based on the real-time acquired environmental data stream, and to identify areas with excessive smoke. Anomaly detection module is used to perform temperature analysis and anomaly detection on the area where smoke exceeds the standard, and to obtain images of the temperature anomaly area and the corresponding infrared area. The smoke and fire detection module is used to perform edge fitting and rule evaluation of the heat source shape in the infrared region image to obtain a list of potential smoke and fire in each temperature anomaly region. The fireworks classification module is used to identify the type of each potential firework in the potential fireworks list according to the preset fireworks feature classification rules, and obtain the fireworks type of each potential firework. The fire alarm verification module is used to perform smoke movement analysis and scene recognition verification on the potential smoke when the smoke type is a preset set of fire alarm types, and generate alarm decision data for each temperature abnormality area. The graded alarm module is used to issue alarms to the temperature abnormality area based on the alarm decision data using a preset graded alarm method.

8. A computing device, characterized in that, The computing device includes: At least one processor; and, A memory and a communication interface that are communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the at least one processor implements the billiard hall fire detection and early warning notification method according to any one of claims 1 to 6 by executing the instructions stored in the memory.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the billiard hall fireworks detection and early warning notification method according to any one of claims 1 to 6.

Images