Low-power consumption wildfire monitoring pan-tilt method and system based on image difference pre-screening

Abstract

The application relates to a low-power consumption forest fire monitoring cloud platform method and system based on image difference pre-screening, and relates to the field of digital data processing.The method comprises the following steps: collecting a reference image frame and cloud platform metadata, collecting a to-be-detected image frame, performing metadata preliminary screening based on the cloud platform metadata, judging whether the reference image frame and the to-be-detected image frame satisfy preset comparison conditions, when the preset comparison conditions are satisfied, calculating the multi-dimensional difference value of the pre-processed reference image frame and the to-be-detected image frame, and obtaining the image difference value by weighting and fusing the multi-dimensional difference value; storing the image difference value in a historical difference value cache area, calculating a dynamic threshold value based on the statistical distribution of the historical difference value in the historical difference value cache area, and when it is detected that the image difference value exceeds the dynamic threshold value for a preset number of times, performing forest fire target identification to obtain the difference ROI region image data and transmitting the difference ROI region image data to a client. By implementing the method, the energy consumption of the forest fire monitoring cloud platform camera equipment can be reduced.

Description

Technical Field

[0001] This application relates to the field of electronic digital data processing, and in particular to a low-power wildfire monitoring pan-tilt-zoom (PTZ) method and system based on image difference pre-screening. Background Technology

[0002] With the increasing demand for forest resource protection and ecological environment monitoring, wildfire monitoring has become an important issue in ecological environment protection. In remote mountainous areas and unattended scenarios, timely detection and early warning of wildfires are of great significance in reducing forest fire losses.

[0003] Current wildfire monitoring primarily relies on pan-tilt-zoom (PTZ) cameras for real-time image acquisition and anomaly detection. These monitoring systems employ continuous high-resolution image acquisition and real-time transmission, using full-frame image analysis to detect potential ignition sources such as smoke or flames, thereby enabling early warning of wildfires.

[0004] However, this method of continuous full-frame image acquisition and real-time analysis has significant power consumption issues in practical applications. Especially in remote mountainous areas powered by batteries, continuous image processing and transmission drastically increase equipment power consumption, making it difficult for monitoring equipment to operate stably for extended periods. Summary of the Invention

[0005] This application provides a low-power pan-tilt-zoom (PTZ) method and system for wildfire monitoring based on image difference pre-screening, which can reduce the energy consumption of pan-tilt-zoom camera equipment for wildfire monitoring.

[0006] Firstly, this application provides a low-power wildfire monitoring pan-tilt-zoom (PTZ) method based on image difference pre-screening, applied to a wildfire monitoring PTG system. The method includes: acquiring reference image frames and PTG metadata, whereby the PTG metadata includes attitude angle, zoom parameters, and ambient temperature data; acquiring image frames to be detected; performing initial metadata screening based on the PTG metadata; determining whether the reference image frame and the image frame to be detected meet preset comparison conditions, wherein the preset comparison conditions include both attitude angle changes and zoom parameter changes being less than a first preset threshold; when the preset comparison conditions are met, performing image preprocessing on the reference image frame and the image frame to be detected, and calculating the preprocessed reference image frame. The multi-dimensional difference values ​​of the image frame to be detected are weighted and fused to obtain the image difference value. The image difference value is stored in the historical difference value cache. A dynamic threshold is calculated based on the statistical distribution of the historical difference values ​​in the historical difference value cache. The dynamic threshold is calculated by the mean, standard deviation and environmental adaptation coefficient of the historical difference values. The environmental adaptation coefficient is dynamically adjusted between a first preset coefficient and a second preset coefficient according to the rate of change of the environmental temperature data. When the image difference value is detected to exceed the dynamic threshold for a preset number of consecutive times, wildfire target identification is performed to obtain the image data of the difference ROI region and the difference ROI region image data is transmitted to the client.

[0007] In the above embodiments, initial metadata screening is performed based on gimbal metadata, and image difference analysis is only conducted when preset comparison conditions are met, reducing unnecessary image processing operations. Dynamically adjusting thresholds based on ambient temperature data makes difference detection more environmentally adaptable. Through weighted fusion of multi-dimensional difference values ​​and dynamic threshold judgment, accurate abnormal area identification is achieved, while reducing system power consumption, enabling the device to operate stably for extended periods in battery-powered scenarios.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, the step of acquiring image frames to be detected and performing initial screening of metadata based on gimbal metadata specifically includes: constructing a time-series sliding window based on the timestamps in the gimbal metadata; calculating the trend characteristics of attitude angle and zoom parameters within the time-series sliding window; performing steady-state determination on the gimbal metadata based on the trend characteristics; calculating the attitude angle deviation and zoom parameter deviation between adjacent image frames when the determination result is steady-state; calculating the environmental disturbance coefficient based on the ambient temperature data; adaptively adjusting a first preset threshold based on the environmental disturbance coefficient to obtain a compensation threshold; determining that a preset comparison condition is met when both the attitude angle deviation and zoom parameter deviation are less than the compensation threshold; and calculating a gimbal adjustment time window based on the trend characteristics when the determination does not meet the preset comparison condition, and re-acquiring a new reference image frame after the adjustment time window ends.

[0009] In the above embodiments, a time-series sliding window is constructed to calculate the changing trend characteristics, and the disturbance coefficient is calculated in combination with ambient temperature data to adaptively adjust the preset threshold. When the gimbal is unstable, the time window is automatically calculated and adjusted, and the reference frame is re-acquired, avoiding misjudgments caused by gimbal jitter and temperature changes. This metadata screening method based on a multi-judgment mechanism ensures the effectiveness of image comparison while reducing the computational load of the system.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, the steps of performing image preprocessing on the reference image frame and the image frame to be detected, calculating the multi-dimensional difference values ​​between the preprocessed reference image frame and the image frame to be detected, and weightedly fusing the multi-dimensional difference values ​​to obtain the image difference value specifically include: performing adaptive downsampling and grayscale processing on the reference image frame and the image frame to be detected based on the zoom parameters in the gimbal metadata; establishing an image spatial mapping relationship based on the attitude angle information in the gimbal metadata to determine the effective monitoring area; within the effective monitoring area, establishing a temperature compensation model based on the ambient temperature data to perform temperature compensation on the image grayscale value; dividing the effective monitoring area into multiple sub-regions, calculating the brightness difference value, texture difference value, and structural difference value in each sub-region to obtain the multi-dimensional difference value; calculating the temperature influence weight of each sub-region based on the changing trend characteristics of the ambient temperature data; and weightedly combining the multi-dimensional difference values ​​of each sub-region based on the temperature influence weight to obtain the image difference value.

[0011] In the above embodiments, the image undergoes adaptive downsampling and grayscale processing, and a temperature compensation model is established within the effective monitoring area. Multi-dimensional difference values ​​are calculated through sub-region division and weighted based on temperature influence. This multi-dimensional difference calculation method, which considers the influence of temperature, reduces computational complexity while improving the robustness and accuracy of difference detection.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, image difference values ​​are stored in a historical difference value cache. A dynamic threshold is calculated based on the statistical distribution of historical difference values ​​in the historical difference value cache. The dynamic threshold is calculated from the mean, standard deviation, and environmental adaptation coefficient of the historical comprehensive difference values. The environmental adaptation coefficient is dynamically adjusted between a first preset coefficient and a second preset coefficient according to the rate of change of the environmental temperature data. Specifically, this includes: storing image difference values ​​in the historical difference value cache in chronological order; acquiring multiple temperature sampling values ​​of the environmental temperature data within a preset time window, calculating the change in adjacent temperature sampling values ​​to obtain the rate of change of the environmental temperature data; setting the environmental adaptation coefficient to the first preset coefficient when the rate of change of the environmental temperature data is less than a preset rate of change threshold; setting the environmental adaptation coefficient to the second preset coefficient when the rate of change of the environmental temperature data is greater than or equal to the preset rate of change threshold; calculating the mean and standard deviation of the historical difference values ​​in the historical difference value cache, and calculating the dynamic threshold based on the mean, standard deviation, and environmental adaptation coefficient.

[0013] In the above embodiments, image difference values ​​are stored in a buffer in chronological order, and an environmental adaptation coefficient is dynamically adjusted based on the rate of change of ambient temperature. A dynamic threshold is calculated using the mean and standard deviation of historical difference values, combined with the environmental adaptation coefficient, establishing an adaptive judgment benchmark. This dynamic threshold calculation method, which combines historical data distribution and environmental changes, improves the adaptability of anomaly detection and effectively reduces false alarms caused by environmental disturbances.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, when the image difference value is detected to continuously exceed a dynamic threshold a preset number of times, the steps of identifying wildfire targets, obtaining image data of the difference ROI region, and transmitting the image data of the difference ROI region to the client specifically include: establishing a spatial coordinate mapping model based on PTZ metadata, calibrating the spatial positional relationship between the reference image frame and the image frame to be detected, and using the spatial coordinate mapping model to locate the difference region in the image frame that continuously exceeds the dynamic threshold; identifying target areas that conform to preset wildfire characteristics from the difference region based on the wildfire fire point and smoke feature model, and obtaining image data of the difference ROI region; extracting geographic coordinate information and monitoring parameter information from PTZ metadata, and sending the geographic coordinate information, monitoring parameter information, and image data of the difference ROI region to the client through a preset communication protocol.

[0015] In the above embodiments, a spatial coordinate mapping model is established to calibrate the positional relationships of images and extract multidimensional features of dissimilar regions. Regions exhibiting wildfire characteristics are identified based on a wildfire feature model, and data is transmitted in conjunction with geographic coordinate information and monitoring parameters. This target detection method based on spatial mapping and feature recognition ensures both detection accuracy and efficient data transmission, reducing system resource consumption.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, after the steps of identifying wildfire targets, obtaining image data of the differentiated Region of Interest (ROI), and transmitting the image data of the differentiated ROI to the client when the image difference value continuously exceeds a preset number of times, the method further includes: performing edge enhancement processing on the reference image frame, extracting contour features in the horizontal, vertical, and diagonal directions, establishing a tree contour feature template, and dynamically adjusting the extraction parameters of the contour features according to the ambient temperature data; performing contour feature extraction on continuous image frames of the differentiated ROI, calculating the degree of difference with the tree contour feature template, and obtaining contour change features, which include contour clarity index, contour continuity index, and contour stability index; identifying blurred contour regions based on the contour change features, extracting grayscale gradient features, texture evolution features, and temporal change features of the blurred contour regions, and calculating the diffusion speed and diffusion direction of the blurred contour regions; confirming a smoke target region when the grayscale gradient features, texture evolution features, and temporal change features conform to the smoke feature pattern, and the diffusion speed and diffusion direction meet preset diffusion conditions; calculating the spatial range, contour blur degree, and diffusion trend of the smoke target region, and generating early warning data containing geographic coordinate information.

[0017] In the above embodiments, edge enhancement is performed on the reference frame and a contour feature template is established to extract contour change features of the differential ROI regions. By combining grayscale gradient, texture evolution, and temporal variation features, the diffusion speed and direction are analyzed. Based on this comprehensive analysis of features, accurate identification of smoke targets is achieved, and early warning data including spatial positioning is generated, improving the monitoring reliability of the system.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, after the steps of identifying wildfire targets, obtaining image data of the differentiated Region of Interest (ROI), and transmitting the image data of the differentiated ROI to the client when the image difference value continuously exceeds a preset number of times, the method further includes: calculating image spatial distortion parameters based on ambient temperature data, performing temperature compensation correction on contour features to obtain corrected contour feature data; performing temporal tracking of the smoke target area based on the corrected contour feature data to obtain smoke target diffusion velocity parameters; calculating the smoke trajectory based on wind direction data and diffusion velocity parameters to obtain smoke spread trend data; calculating the smoke diffusion range based on smoke spread trend data and geographic information data to obtain affected area data; and generating graded early warning information based on affected area data and diffusion velocity parameters.

[0019] In the above embodiments, image spatial distortion parameters are calculated based on ambient temperature data, and temperature compensation correction is applied to contour features. Spread velocity parameters are obtained through temporal tracking of smoke targets, and the trajectory is calculated using wind direction data. The spread range is calculated based on smoke spread trends and geographic information, generating tiered early warning information. This establishes a comprehensive smoke situation assessment mechanism, enabling accurate prediction and tiered early warning of wildfire development trends.

[0020] Secondly, embodiments of this application provide a wildfire monitoring pan-tilt system, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, which includes computer instructions, and the one or more processors call the computer instructions to cause the wildfire monitoring pan-tilt system to perform the method described in the first aspect and any possible implementation thereof.

[0021] Thirdly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on a wildfire monitoring PTZ system, cause the wildfire monitoring PTZ system to execute the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a wildfire monitoring PTZ system, cause the wildfire monitoring PTZ system to perform the method described in the first aspect and any possible implementation thereof.

[0023] Understandably, the wildfire monitoring pan-tilt system provided in the second aspect, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the methods provided in the embodiments of this application. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0024] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0025] 1. This application uses gimbal metadata for initial screening, performing image difference analysis only when preset comparison conditions are met, thus reducing unnecessary image processing operations. Dynamically adjusting thresholds based on ambient temperature data makes difference detection more environmentally adaptable. Through weighted fusion of multi-dimensional difference values ​​and dynamic threshold judgment, accurate abnormal area identification is achieved, while reducing system power consumption, enabling the device to operate stably for extended periods in battery-powered scenarios.

[0026] 2. This application constructs a time-series sliding window to calculate the changing trend characteristics, combines this with ambient temperature data to calculate the disturbance coefficient, and adaptively adjusts the preset threshold. When the gimbal is unstable, it automatically calculates and adjusts the time window and re-acquires reference frames, avoiding misjudgments caused by gimbal jitter and temperature changes. This metadata screening method based on a multi-judgment mechanism ensures the effectiveness of image comparison while reducing the computational load of the system.

[0027] 3. This application establishes a temperature compensation model within the effective monitoring area by adaptively downsampling and grayscale processing the image. Multi-dimensional difference values ​​are calculated through sub-region division and weighted based on temperature influence. This multi-dimensional difference calculation method, which considers the influence of temperature, reduces computational complexity while improving the robustness and accuracy of difference detection. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating a low-power wildfire monitoring PTZ method based on image difference pre-screening in an embodiment of this application.

[0029] Figure 2 This is another flowchart illustrating the low-power wildfire monitoring PTZ method based on image difference pre-screening in this application embodiment;

[0030] Figure 3 This is a schematic diagram of the physical device structure of a wildfire monitoring pan-tilt system in this application embodiment. Detailed Implementation

[0031] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification of this application, the singular expressions “a,” “an,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0032] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0033] To facilitate understanding, the application scenarios of the embodiments of this application are described below.

[0034] In a remote mountainous nature reserve, the management department needs to deploy a wildfire monitoring system across 500 square kilometers of forest. Due to the complex terrain and limited power and communication resources, traditional monitoring equipment struggles to operate continuously and stably. Especially during the peak wildfire season in summer, daytime temperatures can reach 35-40℃, with significant diurnal temperature variations, leading to fluctuations in image quality. Existing pan-tilt-zoom (PTZ) equipment continuously acquires 4K high-definition images 24 / 7 and uploads them for analysis in real time, consuming approximately 80-100 watt-hours of power daily. Limited by the solar power system, the equipment frequently shuts down due to insufficient power. Furthermore, the limited bandwidth of the mountainous communication network often causes congestion and delays in image data transmission, affecting monitoring effectiveness. In addition, natural disturbances such as swaying trees and changing cloud shadows also lead to frequent false alarms, increasing management costs. Therefore, a low-power, low-bandwidth wildfire monitoring solution with strong environmental adaptability is urgently needed.

[0035] A traditional wildfire monitoring system used in a certain forest area employs a fixed-interval sampling method, acquiring a 1920×1080 resolution image frame every 5 seconds. The system detects smoke by setting fixed thresholds for brightness, color, and shape, triggering an alarm when an area matching the preset characteristics appears in the image. However, this approach has significant drawbacks: First, continuous high-resolution image acquisition results in an average power consumption of 120 watts, making it difficult to maintain 24-hour operation even with a 200W solar panel and a 100Ah battery, especially on cloudy or rainy days. Second, each monitoring point generates approximately 100GB of image data daily, which is difficult to transmit under the 4G network in mountainous areas, frequently causing data backlogs. Third, due to the use of fixed thresholds, the system cannot adapt to changes in environmental conditions such as light and temperature, resulting in a false alarm rate as high as 30%. Especially in light fog at dawn or dusk or in windy weather, the system misjudges swaying trees or natural fog as smoke, severely impacting its effectiveness.

[0036] After adopting the low-power wildfire monitoring scheme of this invention, the system first establishes a pre-screening mechanism based on gimbal metadata. The device acquires low-resolution (640×480) preview images at 1-second intervals, while simultaneously recording parameters such as attitude angle and zoom magnification. By comparing the gimbal parameters of adjacent frames, the system only performs image analysis when the device is stable (attitude angle change <0.5°), avoiding unnecessary calculations. In the image processing stage, the system divides the monitoring area into 48 sub-regions, independently calculates feature values ​​such as brightness and texture for each sub-region, and dynamically adjusts the judgment threshold according to changes in ambient temperature. For example, when the temperature rises rapidly (>5℃ / 30 minutes), the system automatically increases the detection sensitivity. When an anomaly is detected, the system only transmits local images of the different areas (typically <100KB in size), reducing bandwidth usage.

[0037] To facilitate understanding, the method provided in this implementation will be described in detail below, using the above scenario as an example. Please refer to [link / reference]. Figure 1This is a flowchart illustrating a low-power wildfire monitoring gimbal method based on image difference pre-screening in an embodiment of this application.

[0038] S101. Acquire reference image frames and gimbal metadata, which includes attitude angle, zoom parameters and ambient temperature data.

[0039] Among them, the reference image frame refers to the baseline image acquired by the gimbal device at a specific moment for subsequent difference comparison; gimbal metadata represents the device's working status parameters recorded synchronously with image acquisition; attitude angle refers to the spatial attitude parameters including the gimbal's pitch angle, yaw angle, and roll angle, used to represent the device's spatial orientation; zoom parameters represent optical parameters including zoom magnification and focal length, used to represent the degree of image magnification; ambient temperature data refers to the real-time temperature readings at the acquisition site, used for subsequent temperature compensation and threshold adjustment.

[0040] This data acquisition step is performed when a new reference baseline needs to be established after the wildfire monitoring system has been activated or completed an anomaly detection. Specifically, the system first checks the operational status of the pan-tilt-zoom (PTZ) device to ensure it is in a stable working state. Then, it synchronously acquires one frame of compressed image data (such as a low-bitrate H.265 encoded image stream) and the corresponding PTZ metadata. The acquired PTZ metadata needs to include accurate timestamp information for subsequent time-series analysis. The system uses this data as the new reference baseline, stores it in the local cache, and updates the relevant status flags to prepare for subsequent difference detection.

[0041] In some embodiments, the acquisition process of reference frames and metadata can be implemented in several ways: Optionally, the system can adopt a timed triggering method, automatically acquiring a set of data at preset time intervals (e.g., 1 second), and selecting the optimal frame as the reference frame by judging quality indicators such as image clarity, exposure suitability, and device jitter; when a significant change in ambient light, temperature, or other conditions is detected, re-acquisition is automatically triggered; the optimal acquisition time is determined based on historical data analysis. Optionally, the system can also adopt an event-triggered method, determining the acquisition time by monitoring the stability of the gimbal attitude, triggering acquisition when the gimbal completes adjustment and maintains a stable state for more than a preset duration (e.g., 2 seconds); combining wind speed sensor data, acquisition is performed when the wind speed is low; the optimal acquisition time is selected by analyzing the image noise level. It is understood that other methods can also be used to implement the adaptive acquisition process of reference frames and metadata, which are not limited here.

[0042] In this step, the gimbal device collects core metadata in low-frequency mode, including attitude angles (such as pitch angle, yaw angle, and roll angle), zoom parameters (such as zoom magnification and focal length), shooting status (such as shutter speed and ISO value), and timestamps. At the same time, it collects compressed image frames (such as H.265 low bitrate streams) to minimize the power consumption and bandwidth requirements of the initial data acquisition. In the wildfire monitoring scenario, additional environmental auxiliary data is collected, such as wind speed and temperature sensor readings, to help determine natural disturbances (such as whether the swaying of leaves caused by wind can simulate the shape of smoke).

[0043] S102. Acquire the image frame to be detected, perform initial screening of metadata based on gimbal metadata, and determine whether the reference image frame and the image frame to be detected meet the preset comparison conditions. The preset comparison conditions include that the change in attitude angle and the change in zoom parameters are both less than the first preset threshold.

[0044] Among them, the image frame to be detected refers to the currently acquired image that needs to be analyzed for differences with the reference frame; metadata screening means that before image analysis, the comparability of two image frames is determined by comparing metadata parameters; preset comparison conditions refer to the parameter constraints that ensure image comparability; the first preset threshold represents the parameter limit for determining the stable working state of the device, which is used to filter out invalid comparisons caused by device adjustments; and reference frame update refers to the process of re-establishing the reference image when the detection conditions are not met.

[0045] After acquiring the reference frame, the system continuously acquires new images for difference detection. Specifically, the system acquires the image to be detected and its metadata at a preset sampling frequency (e.g., 1 second / frame). Then, by comparing the metadata of the frame to be detected with that of the reference frame, the system calculates the attitude angle change value (including the Euclidean distance between pitch, yaw, and roll angles) and the zoom parameter change rate. The system compares the calculated change values ​​with preset thresholds (e.g., attitude angle change not exceeding 0.5 degrees, zoom parameter change not exceeding 5%). When the judgment result does not meet the preset comparison conditions, it indicates that the gimbal may have been adjusted or interfered with. At this time, the system pauses the difference detection process and executes a reference frame update strategy: first, it analyzes the trend of gimbal parameter changes to predict the duration of the adjustment process; then, it continuously monitors the gimbal status until it detects that the device has stabilized again (e.g., attitude angle changes are less than the threshold for multiple consecutive frames); finally, it sets the latest stable frame as the new reference frame and restarts the difference detection process. Only when the judgment result meets the preset comparison conditions will subsequent image analysis steps continue.

[0046] In some embodiments, metadata screening and reference frame updates can be implemented in several ways: Optionally, the system can adopt a predictive update strategy, anticipating upcoming adjustments by analyzing gimbal control commands and pausing the use of the current reference frame in advance; while waiting for the gimbal adjustment to complete, image data is collected and cached; after the gimbal stabilizes, the optimal frame is selected from the cached image sequence as the new reference frame. Optionally, the system can also adopt a gradual update strategy, initiating the acquisition process of alternative reference frames in advance when slightly unmet conditions are detected (such as attitude angle changes approaching but not exceeding a threshold); while continuing to use the current reference frame for detection; when conditions are completely unmet, immediately switching to the prepared alternative reference frame. It is understood that other methods can also be used to implement the metadata screening and reference frame update process, which are not limited here.

[0047] In some embodiments, the difference score between the current frame and the reference frame can also be calculated using the collected metadata. For example, the comprehensive score of attitude angle change (single axis ≤ 0.5°), zoom change (≤ 5%) and time interval (≤ 1 second) can be calculated by weighted formula. If the score is ≤ 1.0, the gimbal is determined to be stable and enters the subsequent comparison. Otherwise, it is marked as actively adjusted and the reference frame is updated. In the wildfire monitoring extension, if the wind speed exceeds the threshold (e.g., 5 m / s), the initial attitude angle screening threshold is dynamically relaxed to 0.7° to adapt to the slight shaking caused by wind in the forest area and avoid frequent invalid screening.

[0048] In some embodiments, this step specifically includes the following steps:

[0049] A time-series sliding window is constructed based on the timestamps in the gimbal metadata, and the changing trend characteristics of attitude angle and zoom parameters are calculated within the time-series sliding window.

[0050] Among them, the time-series sliding window refers to a fixed-length data analysis interval that moves along the time axis; the timestamp indicates the precise moment of data acquisition; attitude angles include the gimbal's pitch angle, yaw angle, and roll angle; zoom parameters include focal length and zoom magnification; and the change trend characteristics describe the change pattern of parameters within the time window, including the rate of change, acceleration, and degree of fluctuation.

[0051] This step constructs a time-series sliding window of length T seconds (e.g., T=5), starting at the current time t and ending at tT. Within the window, the attitude angle sequence A(t) and zoom parameter sequence Z(t) are extracted according to timestamps. For each sequence, the following calculations are performed: first-order difference yields the rate of change v(t) = dA(t) / dt or dZ(t) / dt; second-order difference yields the acceleration a(t) = d²A(t) / dt² or d²Z(t) / dt²; the least squares method is used to fit the parameter change curve p(t) = kt + b, yielding the trend line slope k; the sequence mean μ, standard deviation σ, and extreme values ​​[min, max] are calculated. A feature vector F = [v, a, k, μ, σ, min, max] is generated for each parameter, and the feature vectors of all parameters are combined to form a complete set of trend features. The window slides forward one sampling period as new data arrives, and the above calculation process is repeated.

[0052] Based on the changing trend characteristics, the gimbal metadata is determined to be in a steady state. When the determination result is in a steady state, the attitude angle deviation and zoom parameter deviation between adjacent image frames are calculated.

[0053] Among them, steady state determination refers to the process of assessing whether the gimbal's working state is stable; attitude angle deviation indicates the degree of difference in attitude angle between two frames of images; zoom parameter deviation indicates the amount of change in zoom parameters between two frames of images; adjacent image frames refer to two frames of image data that are consecutive in time.

[0054] This step performs steady-state determination based on the trend feature vector: Calculate the rate of change of attitude angle |v_A| and the rate of change of zoom parameter |v_Z|, requiring |v_A| < θ_v1 (e.g., 0.1 degrees / second) and |v_Z| < θ_v2 (e.g., 0.05 / second); calculate the accelerations |a_A| and |a_Z|, requiring |a_A| < θ_a1 (e.g., 0.05 degrees / second²) and |a_Z| < θ_a2 (e.g., 0.02 / second²); calculate the slope of the trend line |k_A| and |k_Z|, requiring |k_A| < θ_k1 (e.g., 0.08) and |k_Z| < θ_k2 (e.g., 0.03); calculate the standard deviations σ_A and σ_Z, requiring σ_A < θ_σ1 (e.g., 0.2 degrees) and σ_Z < θ_σ2 (e.g., 0.1). When all conditions are met, the state is considered steady. The Euclidean distance between the attitude angle vectors of adjacent frames is calculated as D_A = √(Δφ² + Δθ² + Δψ²) and the relative rate of change of zoom parameters is calculated as D_Z = |Z2 - Z1| / Z1.

[0055] The environmental disturbance coefficient is calculated based on the ambient temperature data, and the compensation threshold is obtained by adaptively adjusting the first preset threshold based on the environmental disturbance coefficient.

[0056] Among them, ambient temperature data refers to the real-time temperature measurement value of the monitoring area; the environmental disturbance coefficient represents the degree of influence of temperature changes on system stability; the preset threshold refers to the benchmark value for judging whether the parameter change exceeds the limit; and the compensation threshold refers to the actual judgment threshold after adjustment for environmental factors.

[0057] This step calculates the temperature change rate R_T = dT / dt and the temperature gradient G_T by analyzing the temperature data T(t). The environmental disturbance coefficient is calculated using the formula: K = K0 × (1 + w1 × R'_T + w2 × G'_T), where K0 is the base coefficient (e.g., 1.0), w1 and w2 are weighting coefficients (e.g., w1 = 0.4, w2 = 0.3), and R'_T and G'_T are the normalized temperature change rate and gradient. When |R_T| > R_max (e.g., 1℃ / minute), w1 increases linearly; when |G_T| > G_max (e.g., 0.5℃ / meter), w2 increases linearly. The compensation threshold θ_c = K × θ0 is obtained by multiplying the preset threshold θ0 by the disturbance coefficient.

[0058] When both the attitude angle deviation and zoom parameter deviation are less than the compensation threshold, the preset comparison condition is determined to be met. When the preset comparison condition is not met, the gimbal adjustment time window is calculated based on the change trend characteristics, and a new reference image frame is re-acquired after the adjustment time window ends.

[0059] Among them, the preset comparison conditions refer to the criteria for judging whether an image meets the conditions for comparison analysis; the gimbal adjustment time window refers to the time interval expected to be required for the gimbal to complete the adjustment; and the reference image frame refers to the baseline image used for subsequent difference analysis.

[0060] This step compares the parameter deviation with the compensation threshold, requiring D_A < θ_cA and D_Z < θ_cZ. If the conditions are not met, the adjustment time is calculated based on the trend characteristics: T_adj = max(|Δθ / v_θ|, |ΔZ / v_Z|) × α, where Δθ and ΔZ are the parameter deviation values, v_θ and v_Z are the rates of change, and α is a safety factor (e.g., 1.2). The gimbal status is continuously monitored within the time window [t, t+T_adj] until the steady-state condition is met. The current image is acquired as a new reference frame, the reference frame timestamp and status marker are updated, relevant counters are reset, and a new detection cycle begins.

[0061] S103. When the preset comparison conditions are met, perform image preprocessing on the reference image frame and the image frame to be detected, calculate the multi-dimensional difference value between the preprocessed reference image frame and the image frame to be detected, and then weight and fuse the multi-dimensional difference value to obtain the image difference value.

[0062] Image preprocessing refers to basic processing operations such as downsampling and grayscale conversion of the original image; multi-dimensional difference value represents the image difference features calculated from multiple dimensions such as brightness, texture, and structure; weighted fusion refers to combining the difference values ​​of each dimension by assigning different weights according to their importance; and image difference value represents the final comprehensive difference evaluation index.

[0063] After confirming that the image frames meet the comparison criteria, the system needs to standardize the images and calculate the differences. Specifically, the images are first downsampled to a uniform resolution (e.g., 128×128 pixels) and converted to grayscale. Then, the monitoring area is divided into multiple sub-regions. For each sub-region, feature values ​​such as brightness histogram difference, Local Binary Pattern (LBP) texture difference, and Structural Similarity (SSIM) are calculated. Finally, based on the discriminative power of each feature in historical data, corresponding weight coefficients are determined, and the final image difference value is obtained through weighted summation.

[0064] In some embodiments, image difference calculation can be implemented in several ways: Optionally, the system can adopt a block-based processing approach, dividing the image into a fixed-size grid and calculating the difference features independently for each grid; using a Gaussian pyramid to extract multi-scale features; and combining gimbal parameters to perform spatial registration of the image before calculating the difference. Optionally, the system can also adopt a feature extraction approach, using HOG features to describe the local gradient distribution of the image; calculating the matching degree of SURF feature points; and extracting the color moment features of the image for comparison. It is understood that other methods can also be used to implement the image difference calculation process, which are not limited here.

[0065] In some embodiments, the image frames that pass the initial screening can also be preprocessed, including grayscale conversion, downsampling to 128×128 resolution and ROI cropping (focusing on the forest ground and canopy areas, excluding the sky). Then, the phase correlation method is used for image registration to correct minor offsets. Then, multi-dimensional difference values ​​are calculated, including histogram Bach distance, perceptual hash Hamming distance and simplified SSIM structural similarity, and weighted and fused into a comprehensive difference value D_total.

[0066] In the wildfire scene expansion, smoke-sensitive edge enhancement preprocessing (such as Sobel operator to extract tree outlines) is introduced to improve the detection sensitivity of blurred smoke changes. At the same time, the dHash algorithm is optimized to block hash (the image is divided into three parts, upper, middle and lower forest areas, for independent calculation) to further improve the accuracy of capturing differences of local fire sources (such as small flames on the ground).

[0067] In some embodiments, this step specifically includes the following steps:

[0068] Adaptive downsampling and grayscale processing are performed on the reference image frame and the image frame to be detected based on the zoom parameters in the gimbal metadata.

[0069] Among them, zoom parameters refer to the numerical values ​​describing the magnification of the image, including focal length and zoom magnification; adaptive downsampling refers to the process of dynamically adjusting the image resolution according to the zoom parameters; grayscale processing refers to the process of converting a color image into a grayscale image; reference image frame and image frame to be detected refer to the baseline image and the current image to be analyzed, respectively.

[0070] This step first calculates the downsampling ratio based on the zoom parameter Z: S = f(Z) = base_scale × (Z / Z_base), where base_scale is the baseline downsampling ratio (e.g., 0.5), and Z_base is the baseline zoom value. Downsampling is then performed on the original image I(w, h): the image resolution is adjusted to (w × S, h × S), and pixel resampling is performed using bilinear interpolation. Next, grayscale conversion is performed: a weighted average formula Gray = 0.299 × R + 0.587 × G + 0.114 × B is applied to the RGB three-channel image to generate a single-channel grayscale image. The same processing procedure is performed on the reference frame and the frame to be detected to ensure that the two frames have the same resolution and grayscale characteristics.

[0071] Based on the attitude angle information in the gimbal metadata, an image spatial mapping relationship is established to determine the effective monitoring area. Within the effective monitoring area, a temperature compensation model is established based on the ambient temperature data to perform temperature compensation on the image grayscale values.

[0072] The attitude angle information includes the gimbal's pitch angle, yaw angle, and roll angle; the image spatial mapping relationship describes the correspondence between pixels in different images; the effective monitoring area refers to the image area that needs to be analyzed; and the temperature compensation model is used to eliminate the influence of temperature changes on the image grayscale values.

[0073] This step establishes a spatial transformation matrix M = Rz(ψ) × Ry(θ) × Rx(φ) based on the attitude angles (φ, θ, ψ). The transformation matrix M is used to calculate the image coordinate mapping: for any point p(x, y) on the image, its projection point P in 3D space is calculated, and then its corresponding point p' in another frame is calculated. The effective monitoring area is determined: the overlapping area of ​​the two frames is calculated, and edge areas are removed to obtain the Region of Interest (ROI). A temperature compensation model is established: G' = G × (1 + k_t × ΔT), where G is the original grayscale value, ΔT is the temperature change value, and k_t is the temperature coefficient (e.g., 0.01 / ℃). Temperature compensation is performed on all pixels within the ROI.

[0074] The effective monitoring area is divided into multiple sub-regions, and the brightness difference value, texture difference value and structural difference value are calculated in each sub-region to obtain multi-dimensional difference values.

[0075] Among them, a sub-region refers to a small image region obtained by equally dividing the monitoring area; brightness difference value represents the degree of difference in grayscale distribution; texture difference value describes the change in local structural features; structural difference value reflects the change in the overall shape of the region; and multi-dimensional difference value is a collection of these feature differences.

[0076] This step uniformly divides the ROI into n×n sub-regions (e.g., 8×8). Within each sub-region, the following are calculated: brightness difference D_I = |μ1-μ2| / max(μ1, μ2), where μ is the average gray level of the region; texture difference D_T = ||LBP1-LBP2||2, using Local Binary Pattern (LBP) features; and structural difference D_S = 1-SSIM(I1, I2), using the Structural Similarity Index (SSIM). A difference feature vector D = [D_I, D_T, D_S] is generated for each sub-region.

[0077] Based on the changing trend characteristics of ambient temperature data, the temperature influence weight of each sub-region is calculated.

[0078] Among them, the temperature change trend characteristics describe the pattern of temperature change over time and space; the temperature influence weight represents the degree of influence of temperature changes on different regions.

[0079] This step analyzes the temperature data sequence T(t), calculating the local rate of change r_t = dT / dt and the spatial gradient g_t. For each sub-region (i, j), a weight is calculated: w(i, j) = 1 / (1 + α × |r_t(i, j)| + β × |g_t(i, j)|), where α and β are adjustment coefficients (e.g., α = 0.2, β = 0.3). The weight matrix is ​​then normalized: w'(i, j) = w(i, j) / Σw.

[0080] Image difference values ​​are obtained by weighting the multi-dimensional difference values ​​of each sub-region based on the influence of temperature.

[0081] Among them, weighted combination refers to the process of calculating the comprehensive difference value according to the weights; the image difference value represents the quantitative indicator of the overall image change.

[0082] This step performs a weighted average on the difference feature vector D and its corresponding weight w' for each sub-region: First, the intra-region weighted average D' = w_I × D_I + w_T × D_T + w_S × D_S is calculated, where w_I, w_T, and w_S are the feature weights (e.g., [0.3, 0.4, 0.3]); then, the inter-region weighted average is calculated: V = Σ(w'(i,j) × D'(i,j)). Finally, the normalized image difference value v = V / V_max is obtained, where V_max is the preset maximum difference value.

[0083] S104. Store the image difference values ​​in the historical difference value cache area. Calculate the dynamic threshold based on the statistical distribution of the historical difference values ​​in the historical difference value cache area. The dynamic threshold is calculated from the mean, standard deviation and environmental adaptation coefficient of the historical difference values. The environmental adaptation coefficient is dynamically adjusted between the first preset coefficient and the second preset coefficient according to the rate of change of the environmental temperature data.

[0084] Among them, the historical difference value cache refers to the data structure used to store the image difference values ​​calculated in the most recent period; the statistical distribution represents the numerical distribution characteristics of the historical difference values; the dynamic threshold refers to the judgment criteria that are adaptively adjusted according to historical data and environmental conditions; and the environmental adaptive coefficient represents the correction factor that is dynamically adjusted according to temperature changes.

[0085] After obtaining new image difference values, the system needs to update historical data and calculate a new judgment threshold. Specifically, the system adds the newly calculated difference values ​​to a fixed-length circular buffer while removing the oldest data to keep the buffer size unchanged. Then, it calculates the mean μ and standard deviation σ of the data in the buffer to obtain the recent temperature change rate ΔT. When ΔT is less than a preset change rate threshold, a smaller adaptive coefficient K1 (e.g., 2.0) is used; otherwise, a larger coefficient K2 (e.g., 3.0) is used. The final dynamic threshold calculation formula is: Threshold = μ + K × σ, where K is the currently selected adaptive coefficient.

[0086] In some embodiments, dynamic threshold calculation can be implemented in several ways: Optionally, the system can employ a time-weighted statistical method, assigning higher weights to recent differences in calculating the mean and standard deviation; adjusting the baseline threshold according to different time periods of the day; and dynamically optimizing the adaptive coefficient by combining historical alarm data. Optionally, the system can also employ a multi-parameter fusion method, simultaneously considering multiple environmental factors such as temperature change rate and light intensity change; establishing a mapping relationship between environmental parameters and the optimal threshold; and predicting the most suitable threshold using machine learning methods. It is understood that other methods can also be used to implement the dynamic threshold calculation process, which are not limited here.

[0087] In some embodiments, a historical difference buffer is maintained to calculate the mean μ and standard deviation σ. An adaptive threshold T = μ + K × σ (K is dynamically adjusted according to environmental fluctuations, such as K = 2.0 in stable forest areas and K = 3.0 in windy areas). If D_total > T and this condition is met for two consecutive frames, lightweight image recognition (such as the YOLOv8-nano model for smoke / flame categories) and minimized reporting (only the difference ROIJPEG, core metadata JSON, and recognition results are transmitted) are triggered. In the wildfire monitoring extension, a temperature linkage mechanism is added. If the temperature reading rises abnormally (>5°C / minute), the K value is reduced to improve threshold sensitivity, and anomalies are classified (such as only reporting coordinates for slight smoke and triggering an alarm for moderate flames), thereby enhancing the robustness of early fire source warnings.

[0088] In some embodiments, this step specifically includes the following steps:

[0089] Image difference values ​​are stored in the historical difference value cache in chronological order.

[0090] Among them, the image difference value refers to the quantified image change index; the historical difference value buffer is a fixed-length circular storage structure; and the time order indicates the order in which the data was collected.

[0091] This step uses a circular buffer of length N (e.g., N=100) to store the sequence of difference values. Each difference value record contains: the difference value v, the timestamp t, and the corresponding temperature value T. When a new difference value arrives, it is appended to the end of the buffer; when the buffer length exceeds N, the oldest record is removed. The buffer data is indexed by timestamp for quick access to data within a specific time range. Data statistics for the buffer are maintained, including the current data volume, the earliest record time, and the latest record time.

[0092] The system acquires multiple temperature samples within a preset time window and calculates the rate of change of the ambient temperature data by measuring the change between adjacent temperature samples.

[0093] Among them, the preset time window refers to a fixed data analysis period; the temperature sampling value is a discrete measurement of the ambient temperature; and the temperature change rate represents the amount of temperature change per unit time.

[0094] This step acquires the temperature sampling sequence {T1, T2, ..., T...} within a time window W (e.g., W = 300 seconds). n} and the corresponding timestamps {t1, t2, ..., t n}. Calculate the temperature change ΔTᵢ = Tᵢ₊1 - Tᵢ and the time interval Δtᵢ = tᵢ₊1 - tᵢ between adjacent sampling points. The temperature change rate is calculated using the formula: Rᵢ = ΔTᵢ / Δtᵢ. The average change rate within the calculated window is R = ΣRᵢ / n, which serves as the current indicator of the ambient temperature change rate.

[0095] When the change rate of the ambient temperature data is less than the preset change rate threshold, set the ambient adaptation coefficient to the first preset coefficient.

[0096] Among them, the preset change rate threshold refers to the standard value for judging the significance of temperature change; the ambient adaptation coefficient is used to adjust the response degree of the system to environmental changes; the first preset coefficient is applicable to the case of relatively stable temperature.

[0097] This step compares the calculated temperature change rate R with the preset threshold R_th (such as 1 °C / minute). When |R| < R_th, set the ambient adaptation coefficient K = K1, where K1 is the first preset coefficient (such as 2.0), indicating that a smaller adjustment amplitude is adopted when the temperature is stable.

[0098] When the change rate of the ambient temperature data is greater than or equal to the preset change rate threshold, set the ambient adaptation coefficient to the second preset coefficient.

[0099] Among them, the second preset coefficient is applicable to the case of significant temperature change, and its value is greater than the first preset coefficient.

[0100] This step sets the ambient adaptation coefficient K = K2 when |R| ≥ R_th, where K2 is the second preset coefficient (such as 3.0). Adopting a larger coefficient value enables the system to have a stronger adaptability to environmental changes. At the same time, record the coefficient switching event, including the switching moment and triggering conditions.

[0101] Calculate the mean and standard deviation of the historical difference values in the historical difference value buffer, and calculate the dynamic threshold according to the mean, standard deviation, and ambient adaptation coefficient.

[0102] Among them, the mean reflects the overall level of historical difference values; the standard deviation represents the degree of fluctuation of difference values; the dynamic threshold is an adaptive abnormal judgment standard.

[0103] This step reads the difference value sequence {v1, v2,..., v n} from the buffer. Calculate the mean μ = Σvᵢ / n and the standard deviation σ = √(Σ(vᵢ - μ)² / (n - 1)). The dynamic threshold calculation formula: θ = K × (μ + σ), where K is the current ambient adaptation coefficient. The threshold calculation takes into account the statistical characteristics of historical data and the influence of environmental changes, and realizes the adaptive adjustment of the judgment standard.

[0104] S105. When it is detected that the image difference values exceed the dynamic threshold continuously for the preset number of times, perform wildfire target recognition to obtain the image data of the differential ROI region and transmit the image data of the differential ROI region to the client.

[0105] Among them, the preset number of times refers to the number of anomaly detections that need to be continuously confirmed, in order to avoid occasional interference; continuously exceeding the dynamic threshold indicates that an abnormal state is continuously detected within multiple sampling periods; the difference ROI region refers to the image region where significant changes are detected; the difference ROI region image data represents only the local image data of the abnormal region and its location information.

[0106] After the system detects the first instance of exceeding the dynamic threshold, continuous verification is required to improve reliability. Specifically, the system continues to execute steps S101 to S104, namely, re-acquiring the frame to be detected and metadata, performing initial metadata screening, image preprocessing, and difference calculation. Each time a new image difference value is calculated, the dynamic threshold is recalculated based on the updated historical difference value cache, thus adapting to gradual environmental changes. The system records the number of consecutive instances exceeding the dynamic threshold; the target recognition process is only initiated when a preset number (e.g., 3 consecutive times) is reached. During the target recognition phase, the system locates areas with significant differences, extracts the image data of that area, and packages it along with geographic coordinate information before sending it to the client. If a detection during continuous verification does not exceed the threshold, the counter is reset and accumulation resumes.

[0107] In some embodiments, the continuous confirmation process can be implemented in several ways: Optionally, the system can adopt a progressive confirmation method, increasing the sampling frequency (e.g., from 1 second / frame to 2 frames / second) after the first threshold is exceeded; simultaneously reducing the threshold requirement for initial metadata screening to ensure continuity; and updating the position of the region of interest in real time during continuous confirmation. Optionally, the system can also adopt a multi-scale confirmation method, performing localized refined analysis on the suspected area after the first anomaly is detected; simultaneously performing contextual analysis over a larger area; and combining a collaborative confirmation mechanism with adjacent PTZs. It is understood that other methods can also be used to implement the continuous confirmation process for abnormal states, which are not limited here.

[0108] In some embodiments, this step specifically includes the following steps:

[0109] A spatial coordinate mapping model is established based on gimbal metadata to calibrate the spatial positional relationship between the reference image frame and the image frame to be detected. The spatial coordinate mapping model is used to locate the difference region in image frames that continuously exceed the dynamic threshold.

[0110] Among them, the spatial coordinate mapping model describes the spatial correspondence of pixels between different images; the gimbal metadata includes spatial parameters such as attitude angle and installation position; the difference region refers to the region in the image that has undergone significant changes; and the spatial position relationship represents the geometric transformation relationship between two frames of images.

[0111] This step first establishes the camera coordinate system based on the gimbal mounting position (X0, Y0, Z0) and attitude angles (φ, θ, ψ). A spatial transformation matrix is ​​constructed: rotation matrix R = Rz(ψ) × Ry(θ) × Rx(φ) and translation vector T = (X0, Y0, Z0). Combined with the camera intrinsic parameter matrix K (containing focal length, principal point coordinates, etc.), a complete projection transformation model P = K[R|T] is established. For any point p1(u1, v1) on the image, its ray direction in the world coordinate system is calculated, and then its corresponding point p2(u2, v2) in another frame is calculated. In N consecutive frames (e.g., N=3) of images exceeding the dynamic threshold, this mapping model is used to align the images, and a region growing algorithm is used to locate the difference regions: using the point with the largest difference value as the seed point, the complete boundary of the difference region is gradually expanded.

[0112] Feature extraction is performed on the differential regions to obtain color features, shape features, edge contour features, and temporal change features. Based on the wildfire fire point and smoke feature model, regions with wildfire characteristics are identified from the differential regions to obtain differential ROI region image data.

[0113] Among them, color features describe the spectral properties of the region; shape features represent the geometric properties of the region; edge contour features describe the boundary properties of the region; temporal variation features represent the evolution of the region over time; and the wildfire feature model contains typical feature descriptions of fire points and smoke.

[0114] This step extracts multidimensional features for each dissimilar region: color features include statistical moments (mean, variance, skewness) in the RGB / HSV space, calculating the color histogram H(c) and color moment M(c); shape features include area S, perimeter L, roundness C=4πS / L², and rectangularity R; edge contour features are extracted using the Canny operator, calculating the edge intensity distribution E(x, y) and orientation histogram D(θ); temporal variation features are calculated by determining the time derivatives of the region's center position, area, and brightness. Feature matching process: The similarity score between the extracted feature vector F and the predefined fire point model F_f and smoke model F_s is calculated using a weighted Euclidean distance d=√(Σwᵢ(Fᵢ-F_mᵢ)²). When the similarity exceeds a threshold, the region is marked as a Region of Interest (ROI), and its location, size, and feature data are recorded.

[0115] Geographic coordinates and monitoring parameters are extracted from the PTZ metadata and sent to the client along with the image data of the region of difference (ROI) via a preset communication protocol.

[0116] The geographic coordinate information includes the latitude and longitude of the target area; the monitoring parameter information includes the working status of the PTZ and environmental parameters; and the preset communication protocol defines the format and method of data transmission.

[0117] This step first extracts information from the gimbal's metadata: based on the gimbal's position (X0, Y0, Z0) and the ROI's location in the image, the target's geographic coordinates (lon, lat) are calculated; monitoring parameters are extracted, including acquisition time t, gimbal attitude (φ, θ, ψ), zoom parameter Z, and ambient temperature T. A standard data packet is constructed: the header contains packet length, timestamp, and device ID; the data area contains ROI image data (using H.264 encoding), geographic coordinates, and monitoring parameters; the tail contains a checksum. A reliable connection is established using the TCP / IP protocol, data is organized according to a preset frame format, synchronization markers and error detection codes are added, and the data is sent to the client as a data stream. The connection status is monitored in real time during transmission, and a retransmission mechanism is initiated when an error occurs.

[0118] The following provides a more detailed description of the process of the method provided in this implementation. Please refer to [link / reference]. Figure 2 This is another flowchart illustrating the low-power wildfire monitoring gimbal method based on image difference pre-screening in this application embodiment.

[0119] S201. When the image difference value is detected to exceed the dynamic threshold for a preset number of consecutive times, wildfire target identification is performed to obtain the image data of the difference ROI region and the image data of the difference ROI region is transmitted to the client.

[0120] Understandably, this step is similar to step S105, and will not be described again here.

[0121] S202. Perform edge enhancement processing on the reference image frame, extract contour features in the horizontal, vertical and diagonal directions, establish tree contour feature templates, and dynamically adjust the extraction parameters of contour features according to the ambient temperature data.

[0122] Edge enhancement refers to the process of enhancing the boundaries of objects in an image through filtering and gradient calculation; contour features represent the shape, intensity, and orientation of object boundaries; contour feature templates are standard data structures used to describe the boundary features of normal trees; extraction parameters include key parameters that affect the feature extraction effect, such as gradient threshold and edge connectivity threshold.

[0123] This step first performs Gaussian filtering on the reference image frame for noise reduction, using a 3×3 or 5×5 filter kernel to eliminate image noise. Then, gradient maps are calculated in the horizontal (Sobel_x), vertical (Sobel_y), and diagonal directions (using Sobel operators at 45° and 135°). Non-maximum suppression is performed on the gradient map in each direction, preserving local maxima. A double-threshold method is used to binarize the gradient maps, with the lower threshold set to 0.4 times the higher threshold, and the threshold base value is linearly adjusted based on ambient temperature data. The processed edge map is then morphologically thinned to obtain a single-pixel width contour line. Key feature points are extracted from the contour line, and feature values ​​such as curvature, direction, and gradient intensity are calculated. These feature values ​​are organized into feature vectors to construct a tree contour feature template. When the ambient temperature increases, the gradient threshold is increased accordingly (e.g., the threshold increases by 2% for every 1°C increase in temperature) to compensate for the image quality degradation caused by thermal disturbances.

[0124] S203. Perform contour feature extraction on continuous image frames of the difference ROI region, calculate the degree of difference with the contour feature template, and obtain contour change features, which include contour sharpness index, contour continuity index and contour stability index.

[0125] Among them, the contour sharpness index represents the statistical characteristics of edge gradient intensity; the contour continuity index represents the integrity and smoothness of the contour line; the contour stability index represents the consistency of contour features between consecutive frames; and the degree of difference represents the quantized value of the deviation between the current contour feature and the template feature.

[0126] Perform the same edge enhancement and feature extraction process as the reference frame on each ROI image frame. Calculate the difference between the current frame's contour features and the template features: the sharpness index is obtained by calculating the gradient intensity ratio of corresponding contour points; the smaller the value, the more blurred the contour. The continuity index is obtained by statistically analyzing the number of breakpoints and curvature changes in the contour lines; the more breakpoints and the greater the curvature change, the worse the continuity. The stability index is obtained by calculating the standard deviation of the corresponding contour point positions over multiple consecutive frames (e.g., 5 frames); the larger the standard deviation, the more unstable the contour. Normalize the three indices to the [0, 1] interval, and obtain the comprehensive difference score through a weighted average. The weights can be set according to actual application requirements, such as [0.3, 0.3, 0.4].

[0127] S204. Identify blurred contour regions based on contour change characteristics, extract grayscale gradient features, texture evolution features, and temporal change features of blurred contour regions, and calculate the diffusion speed and diffusion direction of blurred contour regions.

[0128] Among them, the blurred contour region refers to the image region where the contour features are significantly degraded; the gray-level gradient feature describes the spatial variation trend of pixel gray values; the texture evolution feature represents the temporal variation features of local image structure; the temporal variation feature describes the variation law of feature parameters over time; and the diffusion speed and direction represent the motion features of the blurred region boundary.

[0129] First, regions with a contour sharpness index below a threshold within the ROI are identified as candidate blurred regions. Features are extracted from these candidate regions: radial grayscale profile curves are calculated to analyze the spatial distribution of grayscale values; texture features are extracted using Local Binary Pattern (LBP), and temporal differences of these texture features are calculated; a time series of feature parameters is established, and the trend is fitted using the least squares method. The diffusion velocity is obtained by calculating the displacement of the blurred region boundary points across consecutive frames; the direction of the velocity vector is the diffusion direction. Specifically, feature points on the blurred region boundary are selected, and their positional changes across consecutive frames are tracked. The displacement distance is divided by the time interval to obtain the velocity value, and the average of the velocity vectors of multiple feature points yields the overall diffusion feature of the region.

[0130] S205. When the grayscale gradient features, texture evolution features, and temporal change features match the smoke feature pattern, and the diffusion speed and diffusion direction meet the preset conditions, it is identified as a smoke target area.

[0131] Among them, the smoke feature pattern refers to the combination of feature parameters that describe typical smoke phenomena; the gray-scale gradient feature represents the gradual distribution of smoke regions from dense to light; the texture evolution feature describes the dynamic changes in smoke morphology; the temporal change feature represents the change law of feature parameters over time; and the preset conditions include speed threshold and direction constraint conditions.

[0132] This step confirms the smoke target through multi-dimensional feature matching. It analyzes grayscale gradation features: calculating the grayscale histogram of the target area to determine if it exhibits a unimodal distribution with a skewness within the range of [-0.5, 0.5]; calculating the grayscale spatial gradient to verify the existence of a gradation trend from the center outwards. It analyzes texture evolution features: calculating the temporal difference of the Local Binary Pattern (LBP) features to determine if the texture complexity increases over time; calculating the entropy change of the grayscale co-occurrence matrix to confirm whether it reflects the characteristics of smoke diffusion. It analyzes temporal variation features: calculating the autocorrelation coefficient of the feature parameters to determine if there are typical periodic changes. Simultaneously, it verifies diffusion characteristics: the diffusion velocity should be within the range of 0.5-5 m / s, and the direction should have an angle of less than 45 degrees with the current wind direction. When all feature judgment results are positive, the area is confirmed as a smoke target.

[0133] S206. Calculate the spatial extent, outline blurring degree, and diffusion trend of the smoke target area, and generate early warning data containing geographic coordinate information.

[0134] Among them, spatial range indicates the location and size of the smoke area in the image; contour blur describes the clarity of the smoke boundary; diffusion trend indicates the predicted direction and speed of smoke development; and early warning data refers to structured information containing multiple monitoring indicators.

[0135] The first step is to calculate the spatial extent: using a connected component labeling algorithm, the pixel set of the smoke region is obtained, and its minimum bounding rectangle and pixel area are calculated; combining gimbal parameters and camera calibration parameters, the pixel coordinates are converted into actual geographic coordinates and physical dimensions. Next, the degree of contour blurring is calculated: the gradient values ​​of the smoke boundary are extracted, and the mean and variance of the gradient intensity at boundary points are statistically analyzed; the fractal dimension of the boundary curve is calculated to quantify the complexity of the boundary. Then, the diffusion trend is calculated: a motion model is built based on the contour features of multiple consecutive frames to predict the location and size of the smoke region in the future. Finally, the calculation results are organized into a warning data package: containing information such as the latitude and longitude of the target location, coverage area, development direction, diffusion speed, and confidence level, along with current environmental parameters and image feature data.

[0136] S207. Calculate the image spatial distortion parameters based on the ambient temperature data, perform temperature compensation correction on the contour features, and obtain the corrected contour feature data.

[0137] Among them, the spatial distortion parameter represents the degree of image distortion caused by temperature changes; temperature compensation correction refers to the image correction process that eliminates the influence of temperature; and the corrected contour feature data refers to the standardized feature data after temperature compensation.

[0138] This step first establishes a model of the relationship between temperature and image distortion: analyzing the difference between the current temperature and the reference temperature, and calculating the change in atmospheric refractive index; estimating the light deflection angle based on the optical path length and temperature gradient. Then, image spatial distortion parameters are calculated: a grid transformation matrix is ​​established to describe the degree of distortion in each region of the image; for distant target areas, the influence of atmospheric turbulence is considered, and corresponding compensation coefficients are added. Finally, feature correction is performed: the coordinates of contour feature points are mapped to the correction space through an inverse transformation; contour curvature and orientation are recalculated; and temperature attenuation compensation is applied to the feature intensity, with the compensation coefficient proportional to the temperature difference and the target distance. This correction ensures the comparability of feature data under different temperature conditions.

[0139] S208. Based on the corrected contour feature data, perform time-series tracking of the smoke target area to obtain the diffusion rate parameters of the smoke target.

[0140] Among them, temporal tracking refers to tracking the change process of the target area in a continuous image sequence; diffusion velocity parameters include radial diffusion velocity and tangential motion velocity; corrected contour feature data refers to temperature-compensated standardized feature data; target area represents the confirmed smoke coverage range.

[0141] This step achieves target tracking by analyzing the evolution of the smoke region in consecutive image frames. First, the centroid position and boundary contour points of the smoke region are extracted in each frame, establishing a correspondence between feature points in consecutive frames. The inter-frame displacement of the centroid position is calculated to obtain the overall motion velocity; the centroid displacement distance divided by the time interval yields the velocity value. The diffusion velocity is calculated through the radial distance change of the boundary contour points: uniformly distributed contour points are selected, and the rate of change of their distance to the centroid over time is calculated. Simultaneously, the rotational and shearing velocities of the smoke region are calculated through the tangential displacement of the contour points. Statistical analysis is performed on all velocity components; after removing outliers, the average velocity and acceleration are calculated to obtain a velocity parameter vector describing the smoke's motion characteristics.

[0142] S209. Calculate the smoke trajectory based on wind direction data and diffusion speed parameters to obtain smoke spread trend data.

[0143] Among them, wind direction data represents the wind direction and speed information of the monitoring area; motion trajectory describes the expected motion path of the smoke centroid; spread trend data includes predictions of the direction, speed and range of smoke spread; and spread velocity parameter refers to the velocity vector characterizing the smoke spread characteristics.

[0144] This step establishes a smoke movement prediction model. First, wind direction data is decomposed into horizontal and vertical components, and a motion state vector is established by combining this with diffusion velocity parameters. A Kalman filter is used to track the position of the smoke centroid and predict its future coordinates. The influence of terrain on smoke movement is considered: an upward velocity component is added to hillsides, and a convergence effect is added to valleys. A diffusion range prediction model is established: the diffusion width perpendicular to the direction of movement is calculated using the centroid trajectory as the center line; the influence of atmospheric stability on diffusion is considered, and the diffusion coefficient is increased under unstable conditions. Numerical integration yields smoke distribution predictions for multiple future time points, forming complete spread trend data.

[0145] S210. Calculate the smoke diffusion range based on smoke spread trend data and geographic information data to obtain data on the affected area.

[0146] The geographic information data includes spatial information such as topographic elevation and vegetation distribution; the smoke diffusion range indicates the predicted smoke-covered area; the affected area data describes the specific geographic areas affected by the smoke and the degree of impact; and the spread trend data includes the predicted results of smoke movement and diffusion.

[0147] This step combines smoke dispersion prediction with geographic information for impact assessment. First, a computational grid is established in the geographic information system (GIS), with the grid cell size adaptively adjusted based on terrain complexity. The probability of being affected by smoke is calculated for each grid cell: a multi-layered weighted assessment model is established, incorporating predicted smoke concentration, terrain height, vegetation density, and other factors. The blocking and guiding effects of terrain on smoke dispersion are considered: the dispersion coefficient is increased on windward slopes and decreased on leeward slopes. The degree of impact for each grid cell is calculated: predicted smoke concentration is mapped to impact levels, and a spatiotemporal impact matrix is ​​generated considering exposure time. Finally, the calculation results are converted into polygonal data in the geographic coordinate system, including the boundary coordinates and classification information of the affected area.

[0148] S211. Generate graded early warning information based on the data of the affected area and the diffusion rate parameters.

[0149] Among them, the graded early warning information refers to different levels of alarm data divided according to the severity of the fire; the affected area data indicates the geographical range and degree of smoke impact; the diffusion rate parameter describes the dynamic characteristics of smoke diffusion; and the early warning level includes corresponding response measures and handling suggestions.

[0150] This step involves multi-dimensional assessment of early warning levels and generation of early warning information. First, a basic early warning level is set based on the affected area and spread rate: a Level 1 warning is issued when the affected area is less than 1000 square meters and the spread rate is less than 1 meter per second; a Level 2 warning is issued when the affected area is between 1000 and 5000 square meters or the spread rate is between 1 and 3 meters per second; and a Level 3 warning is issued when the affected area is greater than 5000 square meters or the spread rate is greater than 3 meters per second. Then, the early warning level is adjusted based on geographic location information: if the affected area includes residential areas, the warning level is raised by one level; if the affected area includes important facilities, it is directly raised to the highest warning level. A standardized early warning information package is generated for each warning level, including the latitude and longitude coordinates of the fire's location, geographic boundary data of the affected area, numerical descriptions of the spread direction and speed, estimated spread time and range, location information of surrounding important targets, suggested evacuation range and routes, and emergency response measures. Simultaneously, early warning information formats tailored to different users are generated: structured data suitable for command systems, concise instructions suitable for on-site personnel, and standardized messages suitable for early warning dissemination platforms. The warning information includes key identifying information such as timestamp, warning number, and issuing authority to ensure the traceability and authority of the warning information.

[0151] In some embodiments, the generation of tiered early warnings can be achieved in multiple ways: Optionally, the system can adopt a comprehensive scoring method, quantifying factors such as affected area, spread rate, and surrounding environment into scoring indicators; setting weight coefficients for different indicators; obtaining a comprehensive score through weighted calculation; and determining the early warning level based on the score range. The system also establishes a dynamic update mechanism: reassessing the fire situation at preset time intervals; adjusting the early warning level promptly when the assessment results show a significant change in the fire severity; and generating an early warning level change notification. Optionally, the system can also adopt a rule-based assessment method, establishing a multi-level judgment rule tree; matching various monitoring indicators with rule conditions; obtaining the early warning level through rule reasoning; and recording the rule triggering process as the basis for early warning. The system configures corresponding response plans for each early warning level: specifying the scope and method of information dissemination; defining the response department and responsibilities; and clarifying the handling process and time limits. It is understood that other methods can also be used to achieve the generation process of tiered early warning information, which are not limited here.

[0152] The wildfire monitoring pan-tilt system in the embodiments of this invention is described below from the perspective of hardware processing. Please refer to [link / reference needed]. Figure 3 This is a schematic diagram of the physical device structure of a wildfire monitoring pan-tilt system in this application embodiment.

[0153] It should be noted that, Figure 3 The structure of the wildfire monitoring pan-tilt system shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0154] like Figure 3 As shown, the wildfire monitoring PTZ system includes a Central Processing Unit (CPU) 301, which can perform various appropriate actions and processes based on programs stored in Read-Only Memory (ROM) 302 or programs loaded from storage section 308 into Random Access Memory (RAM) 303, such as executing the methods described in the above embodiments. The RAM 303 also stores various programs and data required for system operation. The CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An Input / Output (I / O) interface 305 is also connected to the bus 304.

[0155] The following components are connected to I / O interface 305: input section 306 including audio input devices, push-button switches, etc.; output section 307 including liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 308 including hard disks, etc.; and communication section 309 including network interface cards such as LAN (Local Area Network) cards, modems, etc. Communication section 309 performs communication processing via a network such as the Internet. Drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as disks, optical disks, magneto-optical disks, semiconductor memories, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.

[0156] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by central processing unit (CPU) 301, it performs the various functions defined in the present invention.

[0157] It should be noted that specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0158] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0159] Specifically, the wildfire monitoring PTZ system of this embodiment includes a processor and a memory. The memory stores a computer program. When the computer program is executed by the processor, it implements the low-power wildfire monitoring PTZ method based on image difference pre-screening provided in the above embodiment.

[0160] In another aspect, the present invention also provides a computer-readable storage medium, which may be included in the wildfire monitoring pan-tilt system described in the above embodiments; or it may exist independently and not assembled into the wildfire monitoring pan-tilt system. The storage medium carries one or more computer programs that, when executed by a processor of the wildfire monitoring pan-tilt system, cause the wildfire monitoring pan-tilt system to implement the low-power wildfire monitoring pan-tilt method based on image difference pre-screening provided in the above embodiments.

[0161] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application 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. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0162] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0163] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A low-power wildfire monitoring pan-tilt-zoom (PTZ) method based on image difference pre-screening, characterized in that, The method, applied to a pan-tilt-zoom (PTZ) system for wildfire monitoring, includes: Collect reference image frames and gimbal metadata, including attitude angle, zoom parameters and ambient temperature data; Acquire image frames to be detected, perform initial screening of metadata based on the gimbal metadata, and determine whether the reference image frame and the image frame to be detected meet preset comparison conditions, wherein the preset comparison conditions include that the attitude angle change and zoom parameter change are both less than a first preset threshold. When the preset comparison conditions are met, image preprocessing is performed on the reference image frame and the image frame to be detected, and the multi-dimensional difference value between the preprocessed reference image frame and the image frame to be detected is calculated. The multi-dimensional difference value is then weighted and fused to obtain the image difference value. The image difference values ​​are stored in a historical difference value cache. A dynamic threshold is calculated based on the statistical distribution of the historical difference values ​​in the historical difference value cache. The dynamic threshold is calculated from the mean, standard deviation and environmental adaptation coefficient of the historical difference values. The environmental adaptation coefficient is dynamically adjusted between a first preset coefficient and a second preset coefficient according to the rate of change of the environmental temperature data. When the image difference value is detected to exceed the dynamic threshold for a preset number of consecutive times, wildfire target identification is performed to obtain the difference ROI region image data and the difference ROI region image data is transmitted to the client.

2. The method according to claim 1, characterized in that, The step of acquiring the image frame to be detected and performing initial metadata screening based on the gimbal metadata specifically includes: A time-series sliding window is constructed based on the timestamp in the gimbal metadata, and the changing trend characteristics of the attitude angle and zoom parameters are calculated within the time-series sliding window. Based on the changing trend characteristics, the gimbal metadata is determined to be in a steady state. When the determination result is in a steady state, the attitude angle deviation and zoom parameter deviation between adjacent image frames are calculated. The environmental disturbance coefficient is calculated based on the ambient temperature data, and the first preset threshold is adaptively adjusted based on the environmental disturbance coefficient to obtain a compensation threshold. When both the attitude angle deviation and the zoom parameter deviation are less than the compensation threshold, it is determined that the preset comparison condition is met. When it is determined that the preset comparison condition is not met, the gimbal adjustment time window is calculated based on the change trend characteristics, and a new reference image frame is re-acquired after the adjustment time window ends.

3. The method according to claim 1, characterized in that, The steps of performing image preprocessing on the reference image frame and the image frame to be detected, calculating the multi-dimensional difference values ​​between the preprocessed reference image frame and the image frame to be detected, and weighted fusing the multi-dimensional difference values ​​to obtain the image difference value specifically include: Adaptive downsampling and grayscale processing are performed on the reference image frame and the image frame to be detected based on the zoom parameters in the gimbal metadata. Based on the attitude angle information in the gimbal metadata, an image spatial mapping relationship is established to determine the effective monitoring area. Within the effective monitoring area, a temperature compensation model is established based on the ambient temperature data to perform temperature compensation on the image grayscale value. The effective monitoring area is divided into multiple sub-regions, and the brightness difference value, texture difference value and structural difference value are calculated in each sub-region to obtain the multi-dimensional difference value. Based on the changing trend characteristics of the ambient temperature data, the temperature influence weight of each sub-region is calculated; The image difference value is obtained by weighting and combining the multi-dimensional difference values ​​of each sub-region based on the temperature influence weight.

4. The method according to claim 1, characterized in that, The step of storing the image difference values ​​in a historical difference value cache and calculating a dynamic threshold based on the statistical distribution of the historical difference values ​​in the historical difference value cache specifically includes: The image difference values ​​are stored in the historical difference value cache in chronological order. The ambient temperature data is obtained by acquiring multiple temperature sampling values ​​within a preset time window, and the rate of change of the ambient temperature data is obtained by calculating the change in adjacent temperature sampling values. When the rate of change of the ambient temperature data is less than a preset rate of change threshold, the environmental adaptive coefficient is set to the first preset coefficient. When the rate of change of the ambient temperature data is greater than or equal to the preset rate of change threshold, the environmental adaptive coefficient is set to the second preset coefficient. Calculate the mean and standard deviation of historical difference values ​​in the historical difference value buffer, and calculate the dynamic threshold based on the mean, the standard deviation and the environmental adaptation coefficient.

5. The method according to claim 1, characterized in that, The step of performing wildfire target identification, obtaining the difference ROI region image data, and transmitting the difference ROI region image data to the client when the image difference value continuously exceeds the dynamic threshold a preset number of times specifically includes: A spatial coordinate mapping model is established based on the gimbal metadata to determine the spatial positional relationship between the reference image frame and the image frame to be detected. The spatial coordinate mapping model is used to locate the difference region in image frames that continuously exceed the dynamic threshold. Based on the wildfire ignition point and smoke feature model, target areas that conform to preset wildfire characteristics are identified from the difference areas to obtain the image data of the difference ROI area; Geographic coordinate information and monitoring parameter information are extracted from the PTZ metadata, and the geographic coordinate information, the monitoring parameter information, and the differential ROI area image data are sent to the client through a preset communication protocol.

6. The method according to any one of claims 1-5, characterized in that, After the steps of identifying wildfire targets, obtaining image data of the region of interest (ROI) with differences, and transmitting the image data of the ROI to the client when the image difference value continuously exceeds the dynamic threshold a preset number of times, the method further includes: Edge enhancement processing is performed on the reference image frame to extract contour features in the horizontal, vertical and diagonal directions, establish a tree contour feature template, and dynamically adjust the extraction parameters of the contour features according to the ambient temperature data. Contour feature extraction is performed on consecutive image frames of the difference ROI region, and the degree of difference with the tree contour feature template is calculated to obtain contour change features, which include contour sharpness index, contour continuity index and contour stability index. Based on the contour change features, identify the blurred contour regions and extract the grayscale gradient features of the blurred contour regions. Texture evolution characteristics and temporal change characteristics are analyzed, and the diffusion speed and diffusion direction of the blurred contour region are calculated. When the grayscale gradient feature, the texture evolution feature, and the temporal change feature conform to the smoke feature pattern, and the diffusion speed and the diffusion direction meet the preset diffusion conditions, it is identified as a smoke target area; The spatial extent, degree of outline blurring, and diffusion trend of the smoke target area are calculated to generate early warning data containing geographic coordinate information.

7. The method according to claim 6, characterized in that, After the steps of identifying wildfire targets, obtaining image data of the region of interest (ROI) with differences, and transmitting the image data of the ROI to the client when the image difference value continuously exceeds the dynamic threshold a preset number of times, the method further includes: Based on the ambient temperature data, the image spatial distortion parameters are calculated, and the contour features are temperature-compensated and corrected to obtain the corrected contour feature data. The smoke target region is time-series tracked based on the corrected contour feature data to obtain the diffusion rate parameter of the smoke target; Calculate the smoke spread trend data based on wind direction data and the aforementioned diffusion speed parameters; The smoke diffusion range is calculated based on the smoke spread trend data and geographic information data to obtain the affected area data; A graded early warning information is generated based on the data of the affected area and the diffusion rate parameter.

8. A wildfire monitoring pan-tilt system, characterized in that, The wildfire monitoring PTZ system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to cause the wildfire monitoring PTZ system to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on the wildfire monitoring PTZ system, the wildfire monitoring PTZ system performs the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on the wildfire monitoring PTZ system, the wildfire monitoring PTZ system performs the method as described in any one of claims 1-7.

Images