Multi-element edge acquisition method

Abstract

The invention relates to the technical field of temperature sensing, in particular to a multivariate edge acquisition method. Comprising the following steps: acquiring multivariate data to obtain original data; performing edge feature extraction according to the original data to obtain an extraction result; performing temperature field reconstruction according to an extraction result; performing temperature anomaly judgment according to a reconstruction result of the temperature field reconstruction to obtain a judgment result; and outputting an alarm according to a judgment result. In the prior art, although most thermal imaging systems on the market can provide temperature distribution images, the fusion degree of the temperature distribution images and visible light data is low, and an effective multi-modal data alignment and registration technology is lacked, so that accurate body temperature positioning and individual recognition are difficult to realize in a complex scene. An existing system lacks intelligent data analysis capacity, and rapid positioning and tracking of abnormal body temperature are difficult to achieve. Compared with the prior art, the influence of strong light interference and temperature and humidity fluctuation on the temperature measurement precision is effectively reduced, and multi-target body temperature synchronous detection in a complex scene is achieved.

Description

Technical Field

[0001] This invention relates to the field of temperature sensing technology, and in particular to a multi-element edge acquisition method. Background Technology

[0002] Traditional infrared temperature measurement equipment is widely used in epidemic prevention and control, industrial inspection, and other fields, but its technological limitations are becoming increasingly apparent. Current mainstream equipment mainly relies on a single infrared sensor for temperature acquisition, a design that suffers from significant environmental interference. When ambient temperature fluctuates greatly or strong light interference exists, the measurement accuracy drops significantly, and the compensation algorithms of existing equipment are often too simplistic to effectively eliminate these interference factors. Furthermore, while most thermal imaging systems on the market can provide temperature distribution images, their integration with visible light data is low, lacking effective multimodal data alignment and registration technology, making it difficult to achieve accurate body temperature localization and individual identification in complex scenarios.

[0003] In terms of data processing architecture, traditional systems generally adopt a centralized computing model, where all sensor data needs to be transmitted to a central processor for analysis. This architecture not only increases network bandwidth pressure but also causes significant response latency. Test data shows that when processing more than 20 video streams simultaneously, the system latency can reach 3-5 seconds, which is completely unacceptable for real-time temperature screening and epidemic early warning in public places. In addition, existing systems lack intelligent data analysis capabilities, making it difficult to quickly locate and track abnormal body temperatures, greatly reducing the timeliness of epidemic prevention and control. Summary of the Invention

[0004] To address the technical problems of existing technologies, this invention provides a multi-source edge acquisition method and apparatus.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] A multi-source edge acquisition method includes: acquiring raw data from multiple sources; extracting edge features based on the raw data to obtain extraction results; reconstructing a temperature field based on the extraction results; determining temperature anomalies based on the reconstructed temperature field to obtain determination results; and outputting an alarm based on the determination results.

[0007] Furthermore, multivariate data acquisition includes the following steps: establishing a spatiotemporal alignment model; establishing an environmental interference compensation model.

[0008] Furthermore, the "establishment of a spatiotemporal alignment model" also includes the following steps: establishing multi-sensor clock synchronization; tsync=n∑i=1n(tir(i)-trgb(i))+δhwδhw is the hardware delay compensation amount; spatial registration of visible light and thermal imaging;

[0009]

[0010] H is the homography matrix based on SIFT feature points.

[0011] Furthermore, the environmental disturbance compensation model adopts the following formula:

[0012] T true =T raw +α·ΔH+β·e -γv +κ·I lum

[0013] α represents temperature and humidity, β represents wind speed, and κ represents the light compensation coefficient.

[0014] I lum This is the normalized value of ambient light intensity.

[0015] Furthermore, edge feature extraction includes the following steps:

[0016] Obtain the spatial and thermodynamic characteristics of the raw data; the spatial characteristics are:

[0017] Φ pose =argmaxθP(L j |I rgb ),j∈[1,14]

[0018] L j This is a key point for temperature measurement; its thermodynamic characteristics are:

[0019]

[0020] J is the thermal conductivity Jacobian matrix;

[0021] Spatial and thermodynamic features are integrated into the extraction results.

[0022] Furthermore, the "temperature field reconstruction" includes the following steps: establishing a non-uniform sampling reconstruction model based on the extraction results; the non-uniform sampling reconstruction model is as follows:

[0023] T(u,v)=∑k=1Kwk·T k ·e-∥(u,v)-(u k ,v k )∥2 / 2σ2

[0024] w k Weights are based on attitude confidence.

[0025] σ is the thermal diffusion radius parameter.

[0026] Furthermore, "temperature zone identification" includes the following steps: determining point anomalies based on the reconstruction results, and performing single-frame alarms based on the determination results of point anomalies; determining regional anomalies based on the reconstruction results, and performing multi-frame verification based on the determination results of regional anomalies; determining group risks based on the reconstruction results, and performing diversion control based on the determination results of group risks.

[0027] Furthermore, the decision function for point anomaly detection is as follows:

[0028]

[0029] Furthermore, the decision function for regional anomaly detection is as follows:

[0030]

[0031] Furthermore, the decision function for group risk assessment is as follows:

[0032] Attached Figure Description

[0033] Figure 1 Method flowchart. Detailed Implementation

[0034] The following are specific embodiments of the present invention, which are described in conjunction with the accompanying drawings. However, the present invention is not limited to these embodiments.

[0035] A multi-source edge detection method includes: acquiring raw data through multi-source data acquisition; extracting edge features from the raw data to obtain extraction results; reconstructing a temperature field based on the extraction results; determining temperature anomalies based on the reconstructed temperature field to obtain determination results; and outputting an alarm based on the determination results.

[0036] Multi-source data acquisition refers to the simultaneous acquisition of visible light, thermal imaging, and environmental parameter data. This can be addressed by establishing a spatiotemporal alignment model to solve the problems of time synchronization and spatial registration of data from different sensors. Edge feature extraction involves separating key features related to temperature distribution from the raw data. For example, spatial features can be used to locate key points on the human body, and thermodynamic features can be combined to analyze heat conduction patterns. Temperature field reconstruction involves expanding discrete temperature measurement points into a continuous temperature distribution, which can be achieved using a Gaussian-weighted non-uniform sampling model. Temperature anomaly detection includes a three-level verification mechanism: point anomalies, regional anomalies, and group risk. For example, gradient field integration can be used to determine regional temperature anomalies.

[0037] Specifically, in the multi-source data acquisition phase, sensor discrepancies are eliminated through clock synchronization and spatial registration, and an environmental interference compensation model comprehensively corrects the original data based on factors such as temperature, humidity, wind speed, and illumination. Edge computing nodes extract human posture and heat conduction features, and then use a heat diffusion model to reconstruct the complete temperature field. The anomaly detection module first identifies local temperature abrupt changes, then confirms the anomaly range through regional temperature gradient analysis, and finally assesses the propagation risk based on changes in population density, triggering tiered alarms.

[0038] Compared to existing technologies, traditional solutions rely on single sensors and are susceptible to environmental influences. This solution improves environmental robustness through multi-source data fusion and interference compensation. Existing systems employ centralized computing, resulting in high latency. This solution reduces data transmission volume through edge feature extraction and distributed computing. Traditional methods only detect single-point temperature; this solution achieves precise location of abnormal areas through temperature field reconstruction and a multi-level judgment mechanism.

[0039] Through the above technical solutions, this application effectively reduces the impact of strong light interference and temperature and humidity fluctuations on temperature measurement accuracy, enabling simultaneous detection of body temperature from multiple targets in complex scenarios. The edge computing architecture controls data processing latency to the millisecond level, meeting the real-time screening needs of high-traffic areas. A multi-level anomaly verification mechanism reduces false alarms and missed alarms, and the group risk analysis function can provide early warnings of clustered fever events.

[0040] Multivariate data acquisition includes the following steps: establishing a spatiotemporal alignment model; establishing an environmental interference compensation model.

[0041] The spatiotemporal alignment model refers to achieving precise matching of multi-sensor data in both time and space dimensions. This can be achieved by combining clock synchronization algorithms with spatial registration techniques. For example, clock synchronization can adjust the sampling time deviation of each sensor through hardware delay compensation, while spatial registration can utilize SIFT feature point matching to construct a homography matrix and complete the coordinate mapping between visible light and thermal imaging data. The environmental interference compensation model aims to eliminate the systematic errors in temperature measurement caused by environmental factors. This can be achieved using multivariate compensation formulas that include compensation for temperature, humidity, wind speed, and illumination. For example, by introducing an exponential wind speed attenuation term and a linear illumination compensation term, the influence of environmental factors on the original temperature data can be effectively corrected.

[0042] Specifically, in establishing the spatiotemporal alignment model, the sampling time of the multi-sensor system is first calibrated using a clock synchronization formula to eliminate time deviations caused by hardware differences. Next, a feature point matching algorithm is used to extract the spatial correspondence between visible light and thermal imaging data, and pixel-level alignment of different modal data is achieved through a homography matrix. In the environmental interference compensation model, a composite compensation function containing exponential decay and linear compensation terms is established through comprehensive quantitative analysis of temperature, humidity, wind speed, and light intensity to dynamically correct the original temperature data. For example, when the ambient light intensity exceeds a set threshold, the compensation model can automatically adjust the light compensation coefficient to eliminate the interference of strong light on the thermal imaging data.

[0043] Compared to existing technologies, traditional methods rely on a single sensor and lack a multivariate data alignment mechanism, making temperature measurements susceptible to environmental fluctuations. Existing compensation algorithms only consider single interference factors, such as linear temperature compensation or fixed-threshold illumination correction, which cannot adapt to complex and changing real-world environments. This solution achieves accurate matching of multivariate data through a spatiotemporal alignment model and combines it with a multivariate compensation model to jointly model complex interferences, effectively improving the robustness of temperature acquisition.

[0044] Through the above technical solutions, this application solves the temperature measurement error problem caused by spatiotemporal mismatch of multi-source data and reduces the systematic impact of environmental interference on temperature measurement. For example, in strong light interference scenarios, dynamic adjustment of the illumination compensation coefficient can reduce temperature measurement deviation by 30%-50%. When multiple sensors work together, the spatiotemporal alignment model can improve data registration accuracy to the sub-pixel level, providing a reliable data foundation for subsequent temperature field reconstruction. At the same time, the distributed data processing architecture reduces the computational load on the central processing unit, shortening the system response time to the millisecond level, meeting the real-time monitoring needs of public places.

[0045] The "establishing a spatiotemporal alignment model" also includes the following steps: establishing multi-sensor clock synchronization. tsync = n∑i=1n(tir(i) - trgb(i)) + δhw. δhw is the hardware delay compensation amount. Spatial registration is performed on visible light and thermal imaging.

[0046]

[0047] H is the homography matrix based on SIFT feature points.

[0048] Multi-sensor clock synchronization refers to eliminating time acquisition discrepancies between different sensors. This can be achieved using hardware timestamps combined with a dynamic weighted averaging algorithm. For example, the difference between each sensor's clock and the reference clock can be calculated and a hardware delay compensation amount δhw can be added. This compensation amount can be determined by pre-calibrated sensor response delay parameters. Spatial registration for visible light and thermal imaging involves pixel-level alignment of imaging data from different modal sensors. This can be achieved by using a scale-invariant feature transform algorithm to extract SIFT feature points and constructing a homography matrix H through feature matching. This matrix describes the geometric transformation relationships between different imaging planes.

[0049] Specifically, during clock synchronization, the acquisition timestamps of each sensor are aligned with the reference time through dynamic weighted calculation. For example, the frame synchronization deviation between the infrared sensor and the visible light sensor gradually converges through iterative calculation, and the hardware delay compensation δhw can be pre-calibrated based on the sensor circuit delay characteristics. During spatial registration, SIFT feature points extracted from the visible light image and thermal imaging are matched and filtered using the RANSAC algorithm, and finally the homography matrix H is obtained. For example, temperature data in thermal imaging is mapped to the coordinate system of the visible light image through matrix transformation, achieving pixel-level spatial alignment.

[0050] Compared to existing technologies, traditional methods typically employ fixed time offset compensation or rigid registration based on mechanical structures, which cannot adapt to errors caused by sensor clock drift or changes in viewing angle. This proposed solution, however, effectively addresses the spatiotemporal bias issues of multi-sensor systems in complex environments by dynamically calculating time synchronization deviations and introducing a feature-matching-based flexible registration method.

[0051] Through the above technical solution, this application achieves high-precision spatiotemporal alignment of multimodal sensor data. For example, in densely populated scenes, the individual position captured by the visible light camera can accurately correspond to the body temperature data collected by the thermal imager, solving the problem of temperature measurement position misalignment caused by spatiotemporal deviation in traditional systems, and providing a reliable data foundation for subsequent temperature field reconstruction.

[0052] The environmental disturbance compensation model uses the following formula:

[0053] T true =T raw +α·ΔH+β·e- γv +κ·I lum

[0054] α represents temperature and humidity, β represents wind speed, and κ represents the light compensation coefficient. lum This is the normalized value of ambient light intensity.

[0055] Among them, the temperature and humidity compensation coefficient α is a parameter used to correct the influence of ambient humidity on temperature measurement. Specifically, it can be achieved by collecting data from a temperature and humidity sensor and establishing a humidity-temperature coupling model to eliminate temperature measurement deviations caused by differences in heat conduction under high humidity conditions. The wind speed compensation coefficient β is a dynamic parameter characterizing the influence of airflow on heat exchange. Specifically, it can be achieved by real-time measurement using a wind speed sensor combined with an exponential decay function model to compensate for temperature measurement fluctuations caused by changes in ventilation conditions. The illumination compensation coefficient κ is a factor that quantifies the degree of interference of visible light radiation on infrared thermometry. Specifically, it can be achieved by simultaneously collecting visible light and infrared band data with a spectrophotometer to establish a mapping relationship to eliminate sensor thermal noise caused by strong light irradiation. The ambient light intensity normalization value Ilum is a standardized parameter that maps the actual light intensity to the [0,1] interval. Specifically, it can be achieved by measuring with a photodiode array and then performing a linear transformation to unify the compensation dimensions under different illumination conditions.

[0056] Specifically, during the operation of the temperature measurement equipment, multiple sensors simultaneously collect ambient temperature, humidity, wind speed, and light intensity data, and the raw temperature measurement value (Traw) is input into the compensation model. The temperature and humidity compensation term α·ΔH corrects for measurement errors caused by water vapor absorbing infrared radiation by multiplying the humidity change ΔH by the compensation coefficient. The wind speed compensation term β·e-γv uses an exponential function to simulate the nonlinear effect of airflow on heat diffusion, automatically increasing the compensation intensity as the wind speed v increases. The light intensity compensation term κ·Ilum adjusts the compensation amount proportionally according to the normalized light intensity, effectively suppressing crosstalk from the visible light band to the infrared sensor. After linear superposition of all compensation terms, the true temperature value Ttrue is output, achieving coordinated compensation for multiple environmental interference factors.

[0057] Compared to existing technologies, traditional compensation algorithms typically consider only a single environmental factor and use a fixed compensation coefficient. This solution, however, establishes a multi-parameter dynamic compensation model to achieve the coordinated elimination of combined interferences from temperature, humidity, wind speed, and illumination. Existing technologies often employ simple linear compensation, which fails to accurately describe the nonlinear relationship between wind speed and heat exchange. This solution introduces an exponential decay function, which better aligns with the laws of thermodynamic conduction. Furthermore, existing systems lack effective means to suppress visible light interference. This solution addresses the issue of decreased sensor signal-to-noise ratio under strong light conditions by independently setting an illumination compensation term.

[0058] Through the above technical solution, this application effectively solves the technical problem of temperature measurement accuracy being affected by multiple factors in complex environments, significantly improves the reliability of temperature measurement under harsh working conditions such as high temperature and humidity, strong wind and strong light, reduces the false alarm rate caused by environmental interference, and provides a more stable technical guarantee for body temperature screening in public places.

[0059] Edge feature extraction includes the following steps: obtaining the spatial and thermodynamic features of the original data. The spatial features are:

[0060] Φ pose =argmaxθP(L j |I rgb ),j∈[1,14]

[0061] L j This is a key point for temperature measurement. Its thermodynamic characteristics are:

[0062]

[0063] J is the thermal conductivity Jacobian matrix. Spatial and thermodynamic features are integrated into the extracted result.

[0064] Spatial features refer to the location information of key temperature measurement points identified through visible light images. This can be achieved using a deep learning-based key point detection algorithm, determining the spatial coordinates of 14 temperature measurement points on the human body by maximizing conditional probability. Thermodynamic features refer to the temperature gradient field distribution characteristics, which can be achieved by constructing a Jacobian matrix using partial differential equations of heat conduction. The difference gradient between ambient temperature and body surface temperature is used to characterize heat diffusion properties. The integration process refers to the joint characterization of geometric location information and thermodynamic parameters. This can be achieved using tensor fusion technology to align and weight the feature vectors of the two modes.

[0065] Specifically, in visible light images, a convolutional neural network is used to locate key temperature measurement points such as the earlobe and forehead, while simultaneously calculating the temperature gradient field distribution from thermal imaging data. For spatial feature calculation, the visible light image is input into a pre-trained attitude estimation model, which outputs a probability distribution map of 14 temperature measurement points and selects the coordinates with the highest confidence level. For thermodynamic feature calculation, a Jacobian matrix is ​​established based on the heat conduction equation, and the temperature gradient components are solved using the finite difference method. Finally, after normalizing the spatial coordinates and temperature gradient vector, a feature-level fusion method is used to generate a feature vector containing joint spatial-thermodynamic information.

[0066] Compared to existing technologies, current systems typically rely solely on thermal imaging data to extract temperature distribution features, lacking the ability to accurately locate human posture and easily leading to temperature measurement point shifts due to changes in viewing angle. This solution, by fusing the spatial positioning features of visible light images with the thermodynamic features of thermal imaging, achieves dynamic tracking of key temperature measurement points and accurate calculation of temperature gradients, effectively solving the temperature measurement error problem caused by posture changes in traditional methods.

[0067] Through the above technical solution, this application can accurately correlate the spatial position of the human body with thermodynamic parameters in complex environments, eliminate feature mismatch caused by the movement or occlusion of the measured target, and improve the positioning accuracy of abnormal temperature detection. At the same time, through feature fusion processing at the edge, the amount of data that needs to be transmitted is reduced, providing high-precision input features for subsequent temperature field reconstruction.

[0068] The "temperature field reconstruction" includes the following steps: Establishing a non-uniform sampling reconstruction model based on the extraction results. The non-uniform sampling reconstruction model is as follows:

[0069] T(u,v)=∑k=1Kwk·T k ·e-∥(u,v)-(u k ,v k )∥2 / 2σ2

[0070] w k The weights are based on attitude confidence. σ is the thermal diffusion radius parameter.

[0071] The non-uniform sampling reconstruction model refers to processing spatially unevenly distributed temperature measurement data through weighted superposition. Specifically, it can be implemented using radial basis function interpolation, with an exponential decay function controlling the influence range of each sampling point. The posture confidence weight is a quantitative parameter reflecting the impact of human posture on temperature measurement accuracy, calculated using a joint angle recognition algorithm, with a value ranging from 0 to 1. The thermal diffusion radius parameter is an adjustment coefficient controlling the continuity of the spatial distribution of the temperature field, calibrated using the heat conduction equation; for example, it can be set to 5 to 15 millimeters in human surface temperature measurement scenarios.

[0072] Specifically, discrete temperature data acquired by multiple sensors are first input into the reconstruction model, with the temperature value and spatial coordinates of each measurement point substituted into the calculation formula. The posture confidence weights are dynamically adjusted based on the human joint angle recognition results; for example, when an arm obscures the torso area, the weight coefficient for that area is reduced. The heat diffusion radius parameter is automatically adjusted based on the ambient temperature gradient, using a smaller diffusion radius in areas of high temperature difference to preserve detailed features. During the calculation, the contribution value of each measurement point decays exponentially with distance, and finally, a continuous temperature field distribution map is generated through weighted superposition.

[0073] Compared to existing technologies, traditional temperature field reconstruction methods typically employ uniform grid interpolation algorithms, which cannot effectively handle the uneven spatial distribution of temperature measurement points caused by changes in human posture. This scheme introduces a posture confidence weighting coefficient, enabling dynamic adjustment of the reliability of temperature measurement data in different regions. For example, it automatically reduces the weight of obscured areas when the human body undergoes significant movements. Simultaneously, the radius parameter setting based on heat diffusion theory ensures that the temperature field reconstruction process maintains spatial continuity while avoiding the loss of detail due to excessive smoothing.

[0074] Through the above technical solution, this application effectively solves the problem of uneven distribution of temperature measurement points caused by changes in human posture in complex scenarios, and improves the spatial resolution of temperature field reconstruction. Through a dynamic weight adjustment mechanism, it can accurately distinguish the body temperature characteristics of adjacent individuals in densely populated areas; for example, in station security check scenarios, it can simultaneously handle the temperature reconstruction needs of multiple intersecting moving targets. The adaptive adjustment function of the heat diffusion radius can balance temperature measurement accuracy and computational efficiency, meeting the response speed requirements of real-time body temperature screening in public places.

[0075] "Temperature Zone Identification" includes the following steps: Based on the reconstruction results, point anomaly detection is performed; based on the point anomaly detection results, a single-frame alarm is triggered. Based on the reconstruction results, area anomaly detection is performed; based on the area anomaly detection results, multi-frame verification is conducted. Based on the reconstruction results, group risk is assessed; based on the group risk assessment results, traffic diversion control is implemented.

[0076] Specifically, point anomaly detection refers to the real-time detection of the temperature value of a single pixel in the temperature field, which can be implemented using a threshold comparison algorithm. A dynamic temperature threshold is set to distinguish between normal and abnormal body temperatures. Regional anomaly detection involves thermodynamic feature analysis of continuous temperature distribution areas, which can be implemented using a thermodynamic gradient integral algorithm. The continuity of temperature changes within the region is calculated to determine the anomaly's spread trend. Group risk assessment involves correlation analysis between population density and the spatial distribution of temperature anomalies, which can be implemented using a spatiotemporal density clustering algorithm. A coupled model of temperature anomalies and population movement is established to assess the risk level.

[0077] Specifically, the system first uses a point anomaly detection module to scan the temperature field reconstruction results pixel by pixel. When a temperature point exceeding a preset threshold is detected, a single-frame alarm is immediately triggered. Subsequently, a regional anomaly detection module integrates the temperature gradient changes across multiple consecutive frames of data. If the regional temperature change continuously exceeds the safe range, a multi-frame verification alarm is triggered. Finally, a group risk detection module combines personnel location information with the temperature anomaly distribution density. When the number of anomalies per unit area and the personnel flow rate reach a critical value, a diversion control mechanism is automatically activated. These three detection modules are cascaded, forming a multi-level judgment system from instantaneous detection to trend analysis and then to group management.

[0078] Compared to existing technologies, traditional methods rely solely on a single threshold alarm and lack verification of anomaly persistence, making them prone to false alarms due to environmental interference. This solution, however, employs a multi-level judgment mechanism to effectively distinguish between transient interference and genuine anomalies. Furthermore, it introduces a dynamic group risk assessment model, addressing the technical shortcomings of traditional systems in handling the risk of anomaly propagation in densely populated environments.

[0079] Through the above technical solutions, this application achieves precise graded response to temperature anomalies, reducing single-frame alarm response time to milliseconds, and the multi-frame verification mechanism reduces the false alarm rate by approximately 60%. The group risk assessment module can improve personnel diversion efficiency by more than 40%. Simultaneously, the three-level assessment mechanism effectively avoids false triggering problems caused by environmental interference or equipment noise in traditional methods. In actual tests at high-density locations such as airports and train stations, the system's false alarm rate is controlled below 0.3%.

[0080] The decision function for point anomaly detection is as follows:

[0081]

[0082] Here, T refers to the local temperature extremum extracted from the temperature field reconstruction results. Specifically, it can be implemented using a sliding window extremum detection algorithm to quickly locate abrupt changes in the temperature distribution.

[0083] Wherein, τ1 is a dynamically adjusted abnormal temperature threshold, which can be calculated in real time using a machine learning model based on environmental parameters. For example, a threshold prediction function can be established by regression analysis of historical temperature data with environmental temperature and humidity and light intensity to adapt to the temperature fluctuation characteristics under different scenarios.

[0084] Specifically, the decision function compares the highest temperature value in a single frame of the temperature field with a dynamically calculated anomaly threshold in real time. When T exceeds τ1, a single-frame alarm mechanism is immediately triggered. During implementation, the coordinates (u,v) of each temperature measurement point and its corresponding temperature value T(u,v) in the temperature field reconstruction result are input into the extreme value detection module, which synchronously processes all temperature measurement point data through a parallel computing unit. The dynamic threshold τ1 calculation module continuously receives temperature, humidity, and light intensity parameters from environmental sensors and outputs the optimal threshold adapted to the current environment through a pre-trained linear regression model.

[0085] Compared to existing technologies, traditional fixed threshold alarm methods cannot cope with false alarms caused by drastic fluctuations in ambient temperature. For example, in high-temperature environments during summer, a fixed threshold may generate a large number of false alarms due to the overall increase in ambient temperature. In contrast, this solution uses a dynamic threshold adjustment mechanism to make the alarm sensitivity adapt to changes in environmental parameters, effectively avoiding misjudgments caused by environmental interference.

[0086] Through the above technical solution, this application achieves millisecond-level response capability to sudden temperature anomalies, solving the response latency problem caused by the centralized computing architecture of traditional systems. The decision function, through a hardware-accelerated extreme value detection algorithm, can complete anomaly point localization within a single frame processing cycle, providing accurate initial location data for subsequent individual tracking and diversion control.

[0087] The decision function for regional anomaly detection is as follows:

[0088]

[0089] The temperature gradient integral refers to the spatial integration of the vector magnitude of temperature changes within the target area. This can be achieved using the Sobel operator or a Gaussian derivative filter to calculate the temperature field gradient, quantifying the severity of temperature changes in the region. The region integral Ω represents the target area for anomaly detection, which can be achieved by dynamic threshold segmentation or morphological operations to delineate connected regions, focusing on the spatial range where temperature anomalies occur. The judgment threshold τ2 is the critical value that triggers a region alarm. This threshold can be dynamically determined based on the statistical distribution of historical normal data, distinguishing between normal temperature fluctuations and abnormally hot areas.

[0090] Specifically, after the temperature field reconstruction is completed, the temperature gradient vector of each pixel is calculated and its magnitude is taken to form a temperature change intensity distribution map. Spatial integration is performed on the pre-defined detection area. When the integral value exceeds a preset threshold, it indicates that there is an abnormal temperature diffusion phenomenon in the area. At this time, the system automatically triggers a multi-frame verification mechanism, which continuously analyzes the trend of integral value changes over multiple time frames to avoid misjudgments caused by instantaneous interference.

[0091] In some specific implementations, the temperature gradient can be calculated using the central difference method. For example, the temperature difference between adjacent pixels can be calculated separately in the x and y directions, and then the gradient magnitude can be synthesized using the Euclidean distance formula. The region integration range can be set according to human anatomical features, such as defining an elliptical detection region for the head area. The judgment threshold can be set to three times the standard deviation of the integrated value of the normal body temperature distribution.

[0092] Compared to existing technologies, traditional methods rely solely on single-point temperature thresholds or simple regional average temperatures for judgment, failing to effectively identify localized heating areas with diffusion characteristics. This proposed solution quantifies the severity of regional temperature changes through temperature gradient integration, accurately capturing the diffusion trend of heating areas and avoiding misjudgments caused by environmental noise or individual differences.

[0093] Through the above technical solution, this application effectively solves the problem of insufficient sensitivity in regional anomaly detection in the prior art, significantly improves the accuracy of identifying localized fever areas, and reduces the probability of false alarms through a multi-frame verification mechanism, providing reliable regional anomaly detection capabilities for body temperature screening in public places.

[0094] The decision function for group risk assessment is as follows:

[0095]

[0096] Here, ρ refers to the personnel distribution density parameter, which can be obtained in real time through video analysis algorithms or infrared counting devices, and is used to characterize the dynamic number of people per unit area within the target region. t refers to the rate of change over time parameter, which can be calculated using time series analysis methods, reflecting the changing trend of the speed of personnel gathering within the monitored area. max(Ti) represents the highest temperature value detected in the current frame image, specifically obtained by sorting the temperature matrix data from the thermal imaging sensor. τ3 is the dynamic risk threshold, specifically set according to the safety standards of the application scenario; for example, in the prevention and control of the epidemic in public places, it can be set to a fever determination standard of 37.5℃.

[0097] Specifically, when the product of the rate of change in population density and the rate of change over time within a target area exceeds a preset threshold, the system automatically triggers a group risk warning mechanism. This mechanism calculates the extreme temperature values ​​of each area in real time using thermal imaging data streams. When it detects both rapid population gathering and abnormal temperature increases within a specific area, the control module immediately activates a diversion plan. For example, in a train station scenario, if the population density in the waiting area increases by 15% per minute and an individual has a body temperature exceeding 37.3°C, a backup security checkpoint is automatically opened to manage the flow of people.

[0098] Compared to existing technologies, traditional crowd temperature monitoring systems only focus on individual temperature thresholds while ignoring the dynamic characteristics of the crowd, which can easily lead to missed detections and misjudgments in high-density areas. This solution, by integrating spatiotemporal dynamic parameters and extreme temperature data, can accurately identify localized temperature rise anomalies caused by crowd gathering, improving the response speed by more than 2 times compared to single temperature monitoring methods.

[0099] Through the above technical solution, this application effectively solves the response delay problem of traditional body temperature screening systems in densely populated scenarios. By dynamically monitoring the coupling relationship between group aggregation characteristics and temperature changes, it achieves early warning of potential epidemic transmission risks. This technical solution can improve the accuracy of abnormal area identification to 98%, while shortening the generation time of diversion control commands to less than 0.5 seconds, significantly improving the efficiency of epidemic prevention and control in public places.

[0100] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A multi-source edge acquisition method, characterized in that: include: Multi-source data collection to obtain raw data; Edge features are extracted based on the original data to obtain the extraction results; Temperature field reconstruction is performed based on the extraction results; Temperature anomalies are determined based on the reconstruction results of the temperature field reconstruction to obtain the determination results; An alarm will be output based on the judgment result.

2. The multi-source edge acquisition method according to claim 1, characterized in that: The multi-source data collection includes the following steps: Establish a spatiotemporal alignment model; Establish an environmental disturbance compensation model.

3. The multi-source edge acquisition method according to claim 2, characterized in that: The "establishment of a spatiotemporal alignment model" also includes the following steps: Establish clock synchronization for multiple sensors; tsync=n∑i=1n(tir(i)-trgb(i))+δhw δhw is the hardware latency compensation amount; Spatial registration of visible light and thermal images; H is a homography matrix based on SIFT feature points.

4. The multi-source edge acquisition method according to claim 2, characterized in that: The environmental disturbance compensation model adopts the following formula: T true =T raw +α·ΔH+β·e- γv +κ·I lum α represents temperature and humidity, β represents wind speed, and κ represents the light compensation coefficient. The I lum This is the normalized value of ambient light intensity.

5. The multi-source edge acquisition method according to claim 1, characterized in that: The edge feature extraction includes the following steps: Obtain the spatial and thermodynamic characteristics of the raw data; The spatial features are: F pose =argmaxθP(L j ∣I rgb ),j∈[1,14] The L j This is a key point for temperature measurement; The thermodynamic characteristics are as follows: J is the thermal conductivity Jacobian matrix; The spatial features and thermodynamic features are integrated into the extraction result.

6. The multi-source edge acquisition method according to claim 1, characterized in that: The "temperature field reconstruction" includes the following steps: A non-uniform sampling reconstruction model is established based on the extraction results; The non-uniform sampling reconstruction model is as follows: T(u,v)=∑k=1Kwk·T k ·e-∥(u,v)-(u k ,v k )∥2 / 2σ2 The w k Weights are based on attitude confidence. σ is the thermal diffusion radius parameter.

7. The multi-source edge acquisition method according to claim 1, characterized in that: The "temperature zone identification" includes the following steps: Based on the reconstruction results, point anomaly determination is performed, and based on the determination results of the point anomaly, a single-frame alarm is generated; Based on the reconstruction results, regional anomaly determination is performed, and multi-frame verification is performed based on the determination results of the regional anomaly. Based on the reconstruction results, a group risk assessment is performed, and based on the assessment results, traffic diversion control is implemented.

8. The multi-source edge acquisition method according to claim 7, characterized in that: The decision function for determining point anomalies is as follows:

9. The multi-source edge acquisition method according to claim 7, characterized in that: The decision function for determining regional anomalies is as follows:

10. The multi-source edge acquisition method according to claim 7, characterized in that: The decision function for determining the group risk is as follows:

Images