A method and system for generating a camera image for automatic fire location

Abstract

This invention discloses a method and system for generating automatic fire location camera images, relating to the field of intelligent fire monitoring technology. The method includes: acquiring visible light and infrared radiation images of the scene under test; inverting the infrared radiation image into a temperature matrix based on the Stefan-Boltzmann law and registering it with the visible light image to obtain a multimodal fused image; extracting the temperature change rate, spatial temperature gradient, and boundary diffusion characteristics of the high-temperature region based on continuous time-series images to construct thermal diffusion dynamics characteristic parameters; identifying the actual fire source region based on the thermal diffusion dynamics characteristic parameters and performing function fitting on the fire source region in conjunction with the temperature gradient distribution to determine the sub-pixel-level center coordinates of the fire source; constructing an angular constraint relationship using the positional changes of the fire source in continuous time-series and the attitude angle changes of the acquisition device to calculate the fire source distance, and determining the spatial location of the fire source based on the sub-pixel-level center coordinates and the distance.

Description

Technical Field

[0001] This invention relates to the field of intelligent fire monitoring technology, and in particular to a method and system for generating automatic fire location camera images. Background Technology

[0002] With the increasing demand for intelligent public safety, fire detection technology has evolved from traditional point-based heat / smoke detection to image-based fire detection based on computer vision. Early image-based detection technologies relied primarily on the color characteristics, flicker frequency, and irregular edges of flames for identification. While achieving non-contact monitoring, these technologies had weak anti-interference capabilities and were easily affected by similar-colored light sources or complex background textures in the environment. In recent years, with the development of deep learning and multimodal perception technologies, fire recognition algorithms based on convolutional neural networks have improved detection accuracy. Some systems have achieved collaborative discrimination of flames and smoke by fusing visible light and infrared images.

[0003] Most methods directly use the grayscale values ​​of infrared images or pseudo-color images without physical calibration for analysis, ignoring the nonlinear mapping relationship between infrared radiation intensity and the real temperature field. This results in poor physical consistency of temperature characteristics, making it difficult to support accurate thermodynamic modeling. Existing algorithms mostly use static analysis of single-frame images, lacking time-series dynamic modeling of the fire heat diffusion process. They cannot effectively distinguish between the continuous heat accumulation and transient interference of the real fire source, and it is difficult to achieve sub-pixel-level high-precision center positioning. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an automatic fire location camera image generation method to solve the problem that most methods directly use the grayscale values ​​of infrared images or pseudo-color images without physical calibration for analysis, ignoring the nonlinear mapping relationship between infrared radiation intensity and the real temperature field, resulting in poor physical consistency of temperature characteristics and difficulty in supporting accurate thermodynamic modeling. Existing algorithms mostly use static analysis of single-frame images, lacking time-series dynamic modeling of the fire heat diffusion process, and cannot effectively distinguish between the continuous heat accumulation and transient interference of the real fire source, and it is difficult to achieve sub-pixel level high-precision center positioning.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for generating automatic fire location camera images, which includes acquiring a visible light image and an infrared radiation image of the scene to be tested, inverting the infrared radiation image into a temperature matrix based on the Stefan-Boltzmann law, and registering it with the visible light image to obtain a multimodal fusion image; Based on continuous time-series images, the temperature change rate, spatial temperature gradient and boundary diffusion characteristics of high-temperature regions are extracted to construct thermal diffusion dynamics characteristic parameters. The actual fire source region is determined based on the thermal diffusion dynamics characteristic parameters, and the fire source region is fitted with a function based on the temperature gradient distribution to determine the sub-pixel level center coordinates of the fire source. By utilizing the positional changes of the fire source in a continuous time sequence and the attitude angle changes of the acquisition device, an angular constraint relationship is constructed to calculate the distance to the fire source, and the spatial position of the fire source is determined based on the sub-pixel-level center coordinates and the distance.

[0007] As a preferred embodiment of the automatic fire location camera image generation method of the present invention, the construction of thermal diffusion dynamics characteristic parameters includes: The high-temperature regions in continuous time-series images are structurally divided, and the regions with relatively concentrated temperature distribution are identified as heat core regions, while the regions that expand outward from the heat core regions and whose temperature gradually decreases are identified as diffusion regions. The boundary position of the diffusion region is extracted frame by frame in consecutive time moments, and the displacement trajectory of the boundary between adjacent time moments is recorded. By statistically analyzing the directional distribution of the displacement trajectory at multiple time moments, the main offset direction of the diffusion region in the image plane is determined. The environmental wind field information corresponding to the scene to be tested is obtained, and the environmental wind field information is converted into a wind direction representation consistent with the image coordinate system, so that the wind direction information and the diffusion area are spatially correlated. The changes in the environmental wind field information over continuous time are analyzed. By comparing the magnitude of wind direction changes at multiple times and the degree of matching with the direction of diffusion area shift, the stability of the wind field information is determined, and a corresponding confidence weight is assigned to the wind field information based on the determination result.

[0008] A continuous time series analysis is performed on the angle change between the main offset direction and the wind direction. Offset components that maintain a consistent direction relationship at multiple times are selected, and boundary displacements that do not satisfy the consistent direction relationship are eliminated, thereby obtaining a stable diffusion direction driven by the wind field. The diffusion path is extended in the opposite direction of the stable diffusion direction, and the diffusion trajectories at multiple times are combined to form a convergence constraint, so that the reverse extension path gradually converges to the same region. The convergence region is determined as the fire source location inversion result, thereby forming the fire source location inversion characteristic parameters under wind field constraints.

[0009] In a preferred embodiment of the automatic fire location camera image generation method of the present invention, the calculation of the fire source distance includes: The subpixel-level center position of the flame region in the image is converted into the corresponding observation direction, and combined with the attitude change of the acquisition device at different acquisition times, multiple spatial observation directions pointing to the flame region are formed at different times. The spatial observation directions at each time point are combined and solved. By analyzing the spatial intersection relationship between different observation directions, the initial spatial position corresponding to the flame area is obtained. The changes in spatial position at consecutive time points are compared to identify abnormal results that deviate from the continuous change trend. By introducing the fire source location inversion feature parameters under the wind field constraint, the initial spatial location and the fire source location obtained by reverse extension are analyzed to decompose the spatial offset relationship between the flame location and the fire source location, so that the offset direction and the wind direction are established to correspond. Based on the credibility weight of the wind field information, the correction magnitude of the spatial observation direction is adaptively adjusted. When the credibility weight of the wind field information is high, the correction degree of the fire source location inversion result to the observation direction is increased. When the credibility weight of the wind field information is low, the correction degree is reduced and the influence of the original observation direction is retained.

[0010] Based on the offset relationship and wind direction constraints, the spatial observation direction is adjusted so that the adjusted observation direction changes from pointing to the flame area to being consistent with the fire source location inversion result, and the adjustment process satisfies the direction consistency between the diffusion direction and the wind direction. By reconstructing the spatial geometric relationship using the adjusted observation direction, the distance to the fire source is calculated. A directional consistency constraint is applied to the calculation results at consecutive time intervals to ensure that the spatial position of the fire source remains stable over time, thereby obtaining the final fire source location result.

[0011] In a preferred embodiment of the automatic fire location camera image generation method of the present invention, the step of identifying the actual fire source area includes: By jointly analyzing the temporal variation characteristics and spatial distribution characteristics of high-temperature areas, target areas that conform to the characteristics of fire source propagation are identified. The stability of the target area in continuous time series is verified to obtain the real fire source area.

[0012] As a preferred embodiment of the automatic fire location camera image generation method of the present invention, the determination of the sub-pixel-level center coordinates of the fire source includes: A continuous variation model is constructed based on the temperature distribution characteristics and boundary variation characteristics within the fire source region; The precise location of the fire source in the image is obtained by solving the extreme value of the continuously changing model.

[0013] In a preferred embodiment of the automatic fire location camera image generation method of the present invention, the calculation of the fire source distance includes: By establishing the relationship between the observed direction changes of the fire source in a continuous time series, spatial geometric constraints are formed between multiple time points; The spatial geometric constraints are then jointly solved to obtain the distance information of the fire source relative to the acquisition device.

[0014] As a preferred embodiment of the automatic fire location camera image generation method of the present invention, the step of determining the spatial location of the fire source includes: The image location information of the fire source is spatially mapped and transformed with the distance information to form the spatial location result of the fire source. And output the location information of the fire source based on the spatial location result.

[0015] Secondly, the present invention provides an automatic fire location camera image generation system, including a multimodal image fusion module, a heat diffusion feature construction module, a fire source precise location module, and a spatial location calculation module; The multimodal image fusion module is used to acquire visible light images and infrared radiation images of the scene under test, invert the infrared radiation image into a temperature matrix based on the Stefan-Boltzmann law, and register it with the visible light image to obtain a multimodal fused image; The thermal diffusion feature construction module extracts the temperature change rate, spatial temperature gradient and boundary diffusion features of the high-temperature region based on continuous time-series images, and constructs thermal diffusion dynamic feature parameters. The precise fire source positioning module is used to identify the actual fire source area based on the thermal diffusion dynamics characteristic parameters, and to perform function fitting on the fire source area in combination with the temperature gradient distribution to determine the sub-pixel-level center coordinates of the fire source. The spatial location calculation module is used to construct an angle constraint relationship to calculate the distance to the fire source by utilizing the position change of the fire source in a continuous time sequence and the attitude angle change of the acquisition device, and to determine the spatial location of the fire source based on the sub-pixel level center coordinates and the distance.

[0016] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the automatic fire location camera image generation method as described in the first aspect of the present invention.

[0017] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the automatic fire location camera image generation method as described in the first aspect of the present invention.

[0018] The beneficial effects of this invention are as follows: by inverting infrared radiation into a physical temperature matrix using the Stefan-Boltzmann law and registering it with a visible light image, the transformation from device-related grayscale signals to a unified physical dimension is realized, laying a physical consistency foundation for subsequent analysis. By constructing thermal diffusion dynamic characteristic parameters using continuous time-series images, static temperature identification is upgraded to dynamic modeling of the flame propagation and evolution process, effectively distinguishing between real fire sources and instantaneous interference. Combined with dynamic parameter discrimination and function fitting techniques, while eliminating false fire sources through time-series consistency, sub-pixel-level center positioning is achieved by breaking through hardware resolution limitations. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart for a method of generating automatic fire location camera images; Figure 2 A schematic diagram of an automatic fire location camera image generation system. Detailed Implementation

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0022] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0023] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0024] Reference Figures 1-2 As an embodiment of the present invention, this embodiment provides a method for generating automatic fire location camera images, including the following steps: S1. Acquire visible light and infrared radiation images of the scene to be tested. Based on the Stefan-Boltzmann law, invert the infrared radiation image into a temperature matrix and register it with the visible light image to obtain a multimodal fusion image.

[0025] Furthermore, after obtaining the multimodal fused image, a deep learning-based target detection step is also included, which uses a convolutional neural network to identify and locate specific targets in the fused image, and marks the abnormally high temperature areas of the targets based on temperature matrix data.

[0026] It should be noted that in existing fire detection technologies, target identification usually relies solely on visible light images or infrared images. Visible light images are easily affected by factors such as changes in lighting and smoke obstruction, while infrared images, although capable of reflecting heat distribution information, lack texture details and are difficult to accurately locate targets. Single-modal data is prone to false detection or missed detection in complex scenarios.

[0027] To ensure physical consistency of thermal information, infrared image grayscale values ​​were not directly used for analysis. Instead, temperature inversion was performed based on the thermal radiation mechanism. By converting equipment-related signals into a physical temperature field and eliminating the influence of different equipment and environments through physical modeling, subsequent diffusion analysis was established on a unified physical foundation. The expression is as follows: ; This transforms infrared radiation information into a temperature matrix with physical meaning, thereby providing a unified and stable input for subsequent thermal diffusion modeling.

[0028] To avoid the lack of a unified physical meaning of thermal information due to the direct use of infrared image grayscale values, infrared images were not directly used as ordinary image features for detection. Instead, temperature inversion was first performed on the infrared radiation information based on the thermal radiation mechanism.

[0029] By converting the radiation intensity output by the sensor into a physically meaningful temperature matrix, thermal features are transformed from device-related signals into comparable physical quantities. Spatial registration of the temperature matrix with a visible light image establishes a pixel-level correspondence between temperature and texture information, thereby forming a multimodal fused image. The expression is: ; in, Pixels in an infrared radiation image The corresponding radiation intensity For the target surface emissivity, It is the Stefan-Boltzmann constant. This represents the temperature value corresponding to each pixel.

[0030] This allows the radiation information in infrared images to be uniformly mapped into a temperature field distribution, thus providing a physically consistent input basis for subsequent high-temperature region extraction, thermal diffusion analysis, and abnormal target labeling.

[0031] After obtaining the multimodal fused image, a convolutional neural network is introduced to perform target detection on the fused image. This is to utilize the ability of deep learning models to express the appearance structure of targets in complex scenes, to perform preliminary screening of candidate targets such as flames and smoke, and to jointly judge the temperature distribution of the corresponding regions of candidate targets through a temperature matrix. This ensures that the target recognition results simultaneously meet image feature constraints and thermal feature constraints, thereby improving the accuracy of screening abnormally high temperature targets.

[0032] S2. Based on continuous time-series images, extract the temperature change rate, spatial temperature gradient and boundary diffusion characteristics of the high-temperature region, and construct thermal diffusion dynamic characteristic parameters.

[0033] Furthermore, the high-temperature regions in the continuous time-series images are structurally divided, with the regions where the temperature distribution is relatively concentrated identified as the heat core region, and the regions that expand outward from the heat core region and whose temperature gradually decreases identified as the diffusion region.

[0034] The boundary position of the diffusion region is extracted frame by frame in consecutive time steps, and the displacement trajectory of the boundary between adjacent time steps is recorded. By statistically analyzing the directional distribution of the displacement trajectory at multiple time steps, the main offset direction of the diffusion region in the image plane is determined.

[0035] The system acquires the environmental wind field information corresponding to the scene under test and converts the environmental wind field information into a wind direction representation consistent with the image coordinate system, so that the wind direction information and the diffusion area are spatially correlated.

[0036] The changes in environmental wind field information over continuous time are analyzed. By comparing the magnitude of wind direction changes at multiple times and the degree of matching with the direction of diffusion area shift, the stability of wind field information is determined, and corresponding confidence weights are assigned to the wind field information based on the determination results.

[0037] Continuous time-series analysis was performed on the angle change between the main offset direction and the wind direction. Offset components that maintained a consistent direction relationship at multiple time points were selected, and boundary displacements that did not satisfy the consistent direction relationship were eliminated, thereby obtaining the stable diffusion direction driven by the wind field.

[0038] The diffusion path is extended in the opposite direction of the stable diffusion direction, and the diffusion trajectories at multiple times are combined to form a convergence constraint, so that the reverse extension path gradually converges to the same region. The convergence region is determined as the fire source location inversion result, thus forming the fire source location inversion characteristic parameters under wind field constraints.

[0039] It should be noted that in existing fire detection and location technologies, the high-temperature area or the center of the detection frame is usually taken directly as the location of the fire source, without considering the propagation process of the flame over time.

[0040] However, in actual fire scenarios, the spatial distribution of flames is not static, but continuously evolves under the combined effects of wind field, pressure field and temperature gradient. The apparent high temperature area reflects the propagation result more than the actual ignition point.

[0041] Instead of directly locating the high-temperature region, the study modeled the flame propagation and evolution process based on continuous time-series data, deducing the "originating source" from the "propagation behavior." At the data level, flame propagation exhibits the following characteristics: the temperature field displays a clear gradient distribution, with the gradient direction reflecting the trend of heat diffusion. Wind fields are directional, causing deviations in the diffusion path, and changes in the pressure field affect the local diffusion rate and diffusion intensity.

[0042] However, if each physical field is used independently, it is difficult to depict the changes in the dominant relationship at different stages. By constructing a multi-physics coupled propagation driving model and introducing wind field, pressure gradient and temperature gradient, the propagation model has multi-source driving capability. By uniformly modeling each physical field, the propagation process can reflect the real fire dynamics mechanism, thereby transforming the description of flame spread from single-factor driving to multi-factor coupled driving.

[0043] The flame propagation driving field is defined as: ; This allows the wind field vector, pressure gradient vector, and temperature gradient vector to be spatially superimposed and fused using coefficients. , , The components of each physical field are quantitatively characterized, enabling the model to dynamically allocate the contribution of different physical fields according to environmental parameters. This allows the driving field $G_t(x,y)$ to accurately describe the direction and intensity of the resultant force of flame propagation under complex environments.

[0044] in, express Time and location The flame propagation driving field vector at that location, express The wind velocity vector at time t. express The pressure gradient vector at any given time reflects the thrust exerted on the flame by pressure changes. express The temperature gradient vector at any given time reflects the trend of thermal diffusion. , , These are the coupling weighting coefficients for the wind field, pressure field, and temperature field, respectively.

[0045] However, if fixed weights are used for modeling, it is impossible to adapt to the changes in propagation characteristics at different stages. By constructing an adaptive weight mechanism, the weights are dynamically adjusted according to the data by analyzing the changes of each physical field in the time dimension. By strengthening the weights of the dominant physical fields, the model can automatically adapt to the propagation state, thereby transforming the propagation model from being manually set to being data-driven.

[0046] The weight normalization constraint and the adaptive weight allocation function are expressed as follows: ; ; By setting a constraint that the sum of weights is 1, the proportion of each physical field in the coupled model is kept normalized, avoiding numerical divergence caused by differences in dimensions. This is achieved by constructing a model based on the changes in physical fields. Function mapping Function mapping The implementation relies on a graph neural network architecture, which abstracts different physical fields into nodes in a graph. By constructing a graph structure for these fields, the message passing mechanism of the graph neural network is used to analyze the dynamic influence between fields of different intensities. An attention mechanism is introduced, which enables the model to adaptively learn and allocate weights, thereby accurately capturing the key inter-field relationships that dominate the physical process. The weight coefficients can respond in real time to drastic fluctuations in environmental parameters, so that the model can automatically focus on the dominant factors at critical moments such as sudden changes in wind field or sudden temperature rise, thus improving the robustness of the model.

[0047] in, , , They represent The magnitude of changes in the wind field, pressure field, and temperature field at any given moment relative to the previous moment. This is the weight adaptive allocation function.

[0048] However, if statistics are based solely on the boundary displacement of the diffusion region, the diffusion direction only reflects changes in the observed data and is easily affected by noise. If only the propagation driving field is considered, the actual observed characteristics may be ignored. Instead of using a single judgment method, the diffusion direction is modeled as an optimization problem. By using an observation consistency term, the direction is made to fit the actual displacement trend. By using a physical driving consistency term, the direction is made to conform to the propagation mechanism. Thus, the diffusion direction is transformed from an empirical judgment into an optimization solution.

[0049] Expression for diffusion direction: ; By calculating the displacement vector difference between the centroid or boundary points of the flame region in consecutive frames. This allows the observation direction to capture the macroscopic movement trajectory of the flame over a short period of time by introducing a time window. Mean filtering is applied to smooth the calculation results from the jitter caused by instantaneous detection noise, thereby improving the observation of the diffusion direction. It can objectively reflect the geometric trend of the actual spread of flames.

[0050] in, This represents the average diffusion direction vector calculated based on observation data. express The coordinates of the feature points in the flame region at any given time. This represents the length of the time sliding window.

[0051] However, relying solely on the observation direction lacks physical constraints, while relying solely on the driving field may deviate from the actual observation. Modeling the diffusion direction as an optimization problem: through the observation consistency term, the direction is made to fit the actual data, and through the physical consistency term, the direction is made to conform to the propagation mechanism, thereby transforming the direction from an empirical judgment into an optimization solution.

[0052] The optimal expression for the diffusion direction is: ; By constructing a system that includes observation directions With physical driving field The cosine similarity objective function makes the final direction This can simultaneously maximize the consistency between the observed data and the physical mechanism by introducing a balance coefficient. This allows the model to find the optimal balance between data reliability and physical constraint strength, so that the solved diffusion direction not only conforms to visual observation facts, but also follows the laws of fluid mechanics and thermodynamics.

[0053] in, To optimize the final diffusion direction vector after the solution, This represents the calculation of cosine similarity between vectors. The adjustment weights for the physics-driven terms.

[0054] However, if diffusion analysis and fire source location are handled separately, the errors cannot be uniformly constrained. By constructing a unified inversion model: generating predicted diffusion results through a propagation model, constraining errors by comparing with actual observations, and reducing single-frame errors through time accumulation, fire source location is transformed from a local speculation problem into a global optimization problem.

[0055] In the process of fire source localization, if diffusion analysis and localization are processed separately, errors will gradually accumulate and cannot be uniformly constrained. A unified inversion model is constructed: predictive results are generated through a propagation model, constraints are formed by comparing with observation data, and errors are accumulated through the time dimension, thereby transforming localization from local speculation to global optimization.

[0056] The fire source inversion expression is: ; in, This represents the coordinates of the optimal initial position of the fire source obtained from the inversion solution. This indicates the location of candidate fire sources in the inversion search space. This represents the total number of frames in the time series. express Actual observation data at time, This represents the forward propagation evolution operator, used to determine the hypothetical fire source. and driving field Deduction The theoretical diffusion state at time t, express The multi-physics field coupling propagation driving field at any given time.

[0057] By constructing a global objective function based on the least squares method, the fire source localization problem is transformed into finding a source that satisfies the "actual observation data" "and theoretical evolution data" The optimization problem of minimizing the residuals between the two sides is solved by introducing a forward propagation evolution operator. This enables the model to be able to determine the assumed ignition point. and physical driving field Forward extrapolation of the spread pattern of flames over time, through time dimension 1 to... The accumulated error allows the inversion process to utilize the temporal information of the entire process rather than a single-frame snapshot to constrain the solution space, thereby enabling the solution to be more efficient. It becomes the globally optimal ignition point that best explains the observed phenomena in terms of physical mechanism, effectively eliminating the misleading effect of local noise and transient interference on the positioning results.

[0058] By constructing a multi-physics coupled propagation driving model and combining an adaptive weighting mechanism with a diffusion direction optimization algorithm, a precise dynamic model of the flame propagation process is achieved. This abandons the static assumption of relying solely on the center of the high-temperature region for positioning in traditional technologies, and instead deduces the "starting source" from the "propagation behavior". By introducing gradient information of wind field, pressure field and temperature field, the positioning drift problem caused by irregular flame shape and interference with the diffusion path is effectively solved in complex environments.

[0059] Meanwhile, the unified inversion model transforms fire source localization into a global optimization problem by constraining error accumulation in the time dimension, reducing the impact of single-frame detection noise on localization accuracy and significantly improving the accuracy and robustness of fire source localization.

[0060] S3. Based on the characteristic parameters of thermal diffusion dynamics, identify the actual fire source region, and combine the temperature gradient distribution to perform function fitting on the fire source region to determine the sub-pixel-level center coordinates of the fire source.

[0061] Furthermore, by jointly analyzing the temporal variation characteristics and spatial distribution characteristics of high-temperature areas, target areas that conform to the characteristics of fire source propagation can be identified.

[0062] The stability of the target area in continuous time series is verified to obtain the real fire source area.

[0063] A continuous variation model is constructed based on the temperature distribution characteristics and boundary variation characteristics within the fire source region.

[0064] The precise location of the fire source in the image is obtained by solving the extreme value of the continuously changing model.

[0065] It should be noted that existing fire source identification methods usually rely on single-frame temperature thresholds, which are easily affected by instantaneous interference. However, real fire sources have continuous and interpretable propagation and evolution characteristics.

[0066] Instead of using single-frame judgment, it is based on temporal consistency for discrimination: stability is judged by temporal continuity and rationality is judged by propagation model, thereby improving the accuracy of fire source identification.

[0067] However, if the pixel center is directly used as the fire source location, it is limited by resolution. By fitting a function to transform discrete pixels into a continuous space, and by solving for the extreme value, the accurate center can be obtained, thereby achieving sub-pixel level positioning.

[0068] ; By constructing a fitting function for the gray-level or temperature distribution in a continuous space. This transforms the discrete pixel matrix into a differentiable continuous surface, and the coordinates of the global extrema of the surface are then obtained by solving for the surface's coordinates. This enables fire source positioning accuracy to break through the limitations of the sensor's physical pixel grid, thereby achieving sub-pixel level precise positioning and eliminating quantization errors caused by insufficient image resolution.

[0069] in, Represents the coordinates of the fire source center with sub-pixel precision. This is a function for fitting grayscale or temperature distributions based on neighboring pixels.

[0070] By introducing temporal consistency discrimination and sub-pixel level localization algorithms, the problem that existing fire source discrimination methods rely on single-frame temperature thresholds and are easily affected by instantaneous interference is effectively overcome. The stability of the fire source is judged by temporal continuity, and the rationality of the evolution is verified by combining the propagation model, thereby reducing the false alarm rate.

[0071] Meanwhile, the traditional pixel center method was replaced by a function fitting and extreme value solving strategy, which broke the physical limitation of hardware resolution on positioning accuracy and realized the fine inversion of the fire source location, providing more accurate coordinate guidance for subsequent fire fighting and rescue operations.

[0072] S4. By utilizing the positional changes of the fire source in continuous time sequence and the attitude angle changes of the acquisition device, an angular constraint relationship is constructed to calculate the distance of the fire source, and the spatial position of the fire source is determined based on the sub-pixel level center coordinates and distance.

[0073] Furthermore, the sub-pixel-level center position of the flame region in the image is converted into the corresponding observation direction, and combined with the attitude changes of the acquisition device at different acquisition times, multiple spatial observation directions pointing to the flame region are formed at different times.

[0074] The spatial observation directions at each time point are combined to solve the problem. By analyzing the spatial intersection relationship between different observation directions, the initial spatial position corresponding to the flame area is obtained. The changes in spatial position at consecutive time points are compared to identify abnormal results that deviate from the continuous change trend.

[0075] By introducing characteristic parameters for fire source location inversion under wind field constraints, the initial spatial location and the fire source location obtained by reverse extension are analyzed to decompose the spatial offset relationship between the flame location and the fire source location, so that the offset direction and the wind direction are established to correspond.

[0076] Based on the credibility weight of wind field information, the correction magnitude of the spatial observation direction is adaptively adjusted. When the credibility weight of wind field information is high, the correction degree of the fire source location inversion result to the observation direction is increased. When the credibility weight of wind field information is low, the correction degree is reduced and the influence of the original observation direction is retained.

[0077] Based on the offset relationship and wind direction constraints, the spatial observation direction is adjusted so that the adjusted observation direction changes from pointing to the flame area to being consistent with the inversion result of the fire source location, and the adjustment process satisfies the directional consistency between the diffusion direction and the wind direction.

[0078] By reconstructing the spatial geometric relationship using the adjusted observation direction, the distance to the fire source is calculated. A directional consistency constraint is applied to the calculation results at consecutive time intervals to ensure that the spatial position of the fire source remains stable over time, thereby obtaining the final fire source location result.

[0079] By establishing the relationship between the observed direction changes of the fire source in a continuous time series, spatial geometric constraints are formed between multiple time points.

[0080] The spatial geometric constraints are solved jointly to obtain the distance information of the fire source relative to the acquisition device.

[0081] The image location information and distance information of the fire source are spatially mapped and transformed to form the spatial location result of the fire source.

[0082] It outputs the location information of the fire source based on the spatial location results.

[0083] It should be noted that the traditional method assumes that the observation direction always points to the fire source. However, due to the deviation of flame propagation, this assumption is not valid. The observation direction is corrected by introducing the inversion results.

[0084] However, if a fixed weight is used to fuse the observation direction and the inversion direction directly, the correction magnitude cannot be dynamically adjusted according to the propagation state. Instead of using simple linear weighting, an adaptive correction model is constructed: the correction intensity is controlled by the wind field confidence, the propagation rationality is judged by the direction consistency, and abrupt changes are avoided by the angle constraint, thereby achieving stable and reliable direction correction.

[0085] The direction smoothing interpolation equation, dynamic coupling correction coefficient, and angle constraint factor are respectively: ; ; ; By constructing a system based on correction coefficients The linear interpolation model makes the corrected observation direction Able to observe the original and inversion direction Smooth transition between them by introducing wind field credibility Consistency factor with direction This allows the correction intensity to dynamically decrease or increase based on the reliability of environmental data and the self-consistency of physical logic, through the angle constraint factor. By limiting the maximum correction angle, the direction correction process avoids drastic jumps caused by local anomalous data, thus achieving a robust correction effect that can both use the inversion results to correct biases and maintain the original characteristics of the observation data.

[0086] in, For the first Corrected direction vector for each observation point Original observation direction This is a theoretical direction based on inversion model calculations. The angle between the original observation direction and the theoretical direction. As a weight for the credibility of the wind field, This is the preset maximum allowable angle deviation.

[0087] However, multiple observation directions are usually not strictly intersecting. A joint optimization solution is adopted: the error is reduced by observing at multiple times and the reliability is improved by adjusting the weights, thereby stabilizing the positioning results.

[0088] The global optimal location inversion model and the adaptive composite weight coefficients are as follows: ; ; By constructing a least-squares optimization objective function, the final localization result is achieved. This minimizes the geometric distance error with all corrected rays by introducing a comprehensive weight. This enables the localization algorithm to prioritize high-confidence algorithms. and high quality The observation data, through coefficients , By adjusting the contribution ratio of different quality indicators, the optimization process can adaptively suppress the interference of low-quality observations, thereby making the fusion positioning results of multi-source observation data mathematically converge to the global optimal solution.

[0089] in, To obtain the optimal fire source location coordinates, For the first The coordinates of each observation point For ray parameters, For the first The overall weight of each observation point To score the quality of the observation data, , This is the weighting adjustment coefficient.

[0090] By constructing an adaptive direction correction model and a multi-observation direction joint optimization algorithm, the fundamental error caused by the assumption that the observation direction points directly to the fire source in traditional methods is resolved. By introducing inversion results to dynamically correct the observation direction and combining wind field reliability and angle constraints, the positioning interference caused by flame propagation offset is effectively eliminated. At the same time, a weighted joint optimization strategy is adopted to process multi-source observation data, which solves the geometric contradiction problem of non-intersecting lines of sight in multiple directions. By assigning higher weights to high-reliability data, the stability and convergence accuracy of the positioning system in complex interference environments are improved, ensuring the uniqueness and reliability of the fire source location calculation.

[0091] This embodiment also provides an automatic fire location camera image generation system, including: a multimodal image fusion module, a heat diffusion feature construction module, a fire source precise location module, and a spatial location calculation module.

[0092] The multimodal image fusion module is used to acquire visible light images and infrared radiation images of the scene under test. Based on the Stefan-Boltzmann law, the infrared radiation image is inverted into a temperature matrix and registered with the visible light image to obtain a multimodal fused image.

[0093] The thermal diffusion feature construction module extracts the temperature change rate, spatial temperature gradient, and boundary diffusion features of high-temperature regions based on continuous time-series images, and constructs thermal diffusion dynamic feature parameters.

[0094] The fire source precise positioning module is used to identify the real fire source area based on the thermal diffusion dynamics characteristic parameters, and to perform function fitting on the fire source area in combination with the temperature gradient distribution to determine the sub-pixel level center coordinates of the fire source.

[0095] The spatial location calculation module is used to construct angular constraint relationships by utilizing the positional changes of the fire source in continuous time and the attitude angle changes of the acquisition device to calculate the distance of the fire source, and determine the spatial location of the fire source based on the sub-pixel level center coordinates and distance.

[0096] This embodiment also provides a computer device applicable to the automatic fire location camera image generation method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the automatic fire location camera image generation method proposed in the above embodiment.

[0097] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0098] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the method for generating automatic fire location camera images as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0099] In summary, this invention achieves the transformation from device-related grayscale signals to unified physical dimensions by inverting infrared radiation into a physical temperature matrix using the Stefan-Boltzmann law and registering it with visible light images, thus laying a physical consistency foundation for subsequent analysis. It also utilizes continuous time-series images to construct thermal diffusion dynamics characteristic parameters, upgrading static temperature identification to dynamic modeling of the flame propagation and evolution process, effectively distinguishing between real fire sources and instantaneous interference. Furthermore, by combining dynamic parameter discrimination and function fitting techniques, it eliminates false fire sources through time-series consistency while overcoming hardware resolution limitations to achieve sub-pixel-level center positioning.

[0100] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for generating automatic fire location camera images, characterized in that: include: The visible light image and infrared radiation image of the scene to be tested are acquired. Based on the Stefan-Boltzmann law, the infrared radiation image is inverted into a temperature matrix and registered with the visible light image to obtain a multimodal fused image. Based on continuous time-series images, the temperature change rate, spatial temperature gradient and boundary diffusion characteristics of high-temperature regions are extracted to construct thermal diffusion dynamics characteristic parameters. The actual fire source region is determined based on the thermal diffusion dynamics characteristic parameters, and the fire source region is fitted with a function based on the temperature gradient distribution to determine the sub-pixel level center coordinates of the fire source. By utilizing the positional changes of the fire source in a continuous time sequence and the attitude angle changes of the acquisition device, an angular constraint relationship is constructed to calculate the distance to the fire source, and the spatial position of the fire source is determined based on the sub-pixel-level center coordinates and the distance.

2. The method for generating automatic fire location camera images as described in claim 1, characterized in that: The constructed thermal diffusion dynamics characteristic parameters include: The high-temperature regions in continuous time-series images are structurally divided, and the regions with relatively concentrated temperature distribution are identified as heat core regions, while the regions that expand outward from the heat core regions and whose temperature gradually decreases are identified as diffusion regions. The boundary position of the diffusion region is extracted frame by frame in consecutive time moments, and the displacement trajectory of the boundary between adjacent time moments is recorded. By statistically analyzing the directional distribution of the displacement trajectory at multiple time moments, the main offset direction of the diffusion region in the image plane is determined. The environmental wind field information corresponding to the scene to be tested is obtained, and the environmental wind field information is converted into a wind direction representation consistent with the image coordinate system, so that the wind direction information and the diffusion area are spatially correlated. The changes in the environmental wind field information over continuous time are analyzed. By comparing the magnitude of wind direction changes at multiple times and the degree of matching with the direction of diffusion area shift, the stability of the wind field information is determined, and a corresponding confidence weight is assigned to the wind field information based on the determination result. A continuous time series analysis is performed on the angle change between the main offset direction and the wind direction. Offset components that maintain a consistent direction relationship at multiple times are selected, and boundary displacements that do not satisfy the consistent direction relationship are eliminated, thereby obtaining a stable diffusion direction driven by the wind field. The diffusion path is extended in the opposite direction of the stable diffusion direction, and the diffusion trajectories at multiple times are combined to form a convergence constraint, so that the reverse extension path gradually converges to the same region. The convergence region is determined as the fire source location inversion result, thereby forming the fire source location inversion characteristic parameters under wind field constraints.

3. The method for generating automatic fire location camera images as described in claim 2, characterized in that: The calculation of the fire source distance includes: The subpixel-level center position of the flame region in the image is converted into the corresponding observation direction, and combined with the attitude change of the acquisition device at different acquisition times, multiple spatial observation directions pointing to the flame region are formed at different times. The spatial observation directions at each time point are combined and solved. By analyzing the spatial intersection relationship between different observation directions, the initial spatial position corresponding to the flame area is obtained. The changes in spatial position at consecutive time points are compared to identify abnormal results that deviate from the continuous change trend. By introducing the fire source location inversion feature parameters under the wind field constraint, the initial spatial location and the fire source location obtained by reverse extension are analyzed to decompose the spatial offset relationship between the flame location and the fire source location, so that the offset direction and the wind direction are established to correspond. Based on the credibility weight of the wind field information, the correction magnitude of the spatial observation direction is adaptively adjusted. When the credibility weight of the wind field information is high, the correction degree of the fire source location inversion result to the observation direction is increased. When the credibility weight of the wind field information is low, the correction degree is reduced and the influence of the original observation direction is retained. Based on the offset relationship and wind direction constraints, the spatial observation direction is adjusted so that the adjusted observation direction changes from pointing to the flame area to being consistent with the fire source location inversion result, and the adjustment process satisfies the direction consistency between the diffusion direction and the wind direction. By reconstructing the spatial geometric relationship using the adjusted observation direction, the distance to the fire source is calculated. A directional consistency constraint is applied to the calculation results at consecutive time intervals to ensure that the spatial position of the fire source remains stable over time, thereby obtaining the final fire source location result.

4. The method for generating automatic fire location camera images as described in claim 3, characterized in that: The determination of the actual fire source area includes: By jointly analyzing the temporal variation characteristics and spatial distribution characteristics of high-temperature areas, target areas that conform to the characteristics of fire source propagation are identified. The stability of the target area in continuous time series is verified to obtain the real fire source area.

5. The method for generating automatic fire location camera images as described in claim 4, characterized in that: The determination of the sub-pixel-level center coordinates of the fire source includes: A continuous variation model is constructed based on the temperature distribution characteristics and boundary variation characteristics within the fire source region; The precise location of the fire source in the image is obtained by solving the extreme value of the continuously changing model.

6. The method for generating automatic fire location camera images as described in claim 5, characterized in that: The calculation of the fire source distance includes: By establishing the relationship between the observed direction changes of the fire source in a continuous time series, spatial geometric constraints are formed between multiple time points; The spatial geometric constraints are then jointly solved to obtain the distance information of the fire source relative to the acquisition device.

7. The method for generating automatic fire location camera images as described in claim 6, characterized in that: Determining the spatial location of the fire source includes: The image location information of the fire source is spatially mapped and transformed with the distance information to form the spatial location result of the fire source. And output the location information of the fire source based on the spatial location result.

8. A fire automatic positioning camera image generation system, based on the fire automatic positioning camera image generation method according to any one of claims 1 to 7, characterized in that: It includes a multimodal image fusion module, a heat diffusion feature construction module, a fire source precise location module, and a spatial location calculation module; The multimodal image fusion module is used to acquire visible light images and infrared radiation images of the scene under test, invert the infrared radiation image into a temperature matrix based on the Stefan-Boltzmann law, and register it with the visible light image to obtain a multimodal fused image; The thermal diffusion feature construction module extracts the temperature change rate, spatial temperature gradient and boundary diffusion features of the high-temperature region based on continuous time-series images, and constructs thermal diffusion dynamic feature parameters. The precise fire source positioning module is used to identify the actual fire source area based on the thermal diffusion dynamics characteristic parameters, and to perform function fitting on the fire source area in combination with the temperature gradient distribution to determine the sub-pixel-level center coordinates of the fire source. The spatial location calculation module is used to construct an angle constraint relationship to calculate the distance to the fire source by utilizing the position change of the fire source in a continuous time sequence and the attitude angle change of the acquisition device, and to determine the spatial location of the fire source based on the sub-pixel level center coordinates and the distance.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the automatic fire location camera image generation method according to any one of claims 1 to 7.

10. 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 automatic fire location camera image generation method according to any one of claims 1 to 7.

Images