Human body state monitoring method and system based on thermal imaging technology

Abstract

The invention relates to the field of human body state monitoring, in particular to a human body state monitoring method and system based on a thermal imaging technology, and the method comprises the steps: obtaining an infrared thermal image and a visible light image of a human body; registering the infrared thermal image to a coordinate system of the visible light image and extracting a temperature field; performing posture evaluation on the visible light image to extract skeleton key points, and dividing the human body into a plurality of eudipleural regions of interest based on the skeleton key points; calculating a corrected thermal symmetry difference value between the eudipleural regions of interest to obtain a thermal symmetry difference chart; carrying out region growth based on the point with the maximum pixel value in the thermal symmetry difference image to obtain an abnormal hot spot region; calculating the product of the inflammation activeness index and the boundary confidence of the abnormal hot spot region to obtain a state risk value; and generating an alarm signal in response to that the state risk value is greater than a set threshold. The problem of high false alarm rate of human body state monitoring in a dynamic complex environment is solved.

Description

Technical Field

[0001] This invention relates to the field of human condition monitoring, and in particular to a method and system for human condition monitoring based on thermal imaging technology. Background Technology

[0002] The spatial distribution characteristics of human body temperature are important physiological indicators reflecting the body's metabolic level, blood circulation status, and inflammatory response. In scenarios such as epidemiological screening and clinical auxiliary diagnosis, non-contact infrared thermal imaging technology is being applied to auxiliary diagnostic scenarios due to its rapid and non-invasive characteristics.

[0003] Existing infrared human body monitoring technologies mainly fall into two categories: one is screening methods based on absolute temperature thresholds, which trigger an alarm when the temperature of a certain part of the human body exceeds a preset value (e.g., 37.3℃); the other is region analysis methods based on image segmentation, which use region growing or edge detection to extract high-temperature areas. However, these technologies have limitations in real-world, complex dynamic monitoring environments. First, human surface temperature is greatly affected by ambient temperature, movement status, and individual basal metabolic rate, making a single absolute threshold highly susceptible to false negatives or false positives. Second, existing image segmentation algorithms often use globally fixed thresholds, failing to distinguish between human skin and its coverings, resulting in extracted abnormal areas often containing background noise or clothing wrinkles and textures.

[0004] Furthermore, existing quantitative assessment indicators are too simplistic, focusing only on extreme temperature values ​​and ignoring the geometry of heat sources and the distribution characteristics of temperature gradients. This makes it difficult to effectively distinguish between diffuse temperature rises caused by pathological inflammation and localized high temperatures caused by external heat sources, and it lacks the ability to identify artifacts. Therefore, how to accurately distinguish complex textured backgrounds from actual lesions while eliminating interference from environmental and individual differences, and how to effectively eliminate false thermal anomaly alarms caused by non-pathological factors from multi-dimensional features, are current technical problems that need to be solved. Summary of the Invention

[0005] To address the problem in existing technologies of accurately distinguishing complex textured backgrounds from actual lesions while eliminating interference from environmental and individual differences, this invention provides a human condition monitoring method and system based on thermal imaging technology.

[0006] In a first aspect, the present invention provides a method for monitoring human body status based on thermal imaging technology, which adopts the following technical solution: A human condition monitoring method based on thermal imaging technology includes the following steps: The process involves acquiring infrared and visible light images of the human body; registering the infrared thermal image to the coordinate system of the visible light image and extracting the temperature field; performing pose assessment on the visible light image to extract skeletal key points, and dividing the human body into multiple symmetrical regions of interest (ROIs) based on these key points; calculating the corrected thermal symmetry difference value between the symmetrical ROIs to obtain a thermal symmetry difference map; the corrected thermal symmetry difference value is positively correlated with the correction coefficient; the correction coefficient is positively correlated with the depth difference of the corresponding limb relative to the lens in the symmetrical ROIs, and negatively correlated with the cosine of the angle between the normal to the surface of the corresponding limb and the optical axis of the lens; performing region growing based on the point with the largest pixel value in the thermal symmetry difference map to obtain abnormal hot spot regions; the criteria for region growing include the growth resistance coefficient; the growth resistance coefficient is positively correlated with the biological thermal texture entropy; the biological thermal texture entropy is a normalized value calculated based on the contrast and energy of the gray-level co-occurrence matrix within the neighborhood window of the largest pixel, and the average modulus of the temperature gradient within the neighborhood window; calculating the product of the inflammation activity index and the boundary confidence of the abnormal hot spot region to obtain the state risk value; and generating an alarm signal in response to the state risk value exceeding a set threshold.

[0007] To address the issue of false alarms in existing infrared thermometry techniques due to changes in human posture and environmental background interference, this invention extracts key points of the skeleton and calculates limb depth differences and angles to perform physical geometric corrections for thermal symmetry differences, effectively eliminating non-pathological temperature measurement deviations caused by distance attenuation and viewing angle tilt. Simultaneously, it introduces bio-thermal texture entropy related to texture features as a criterion for region growth, enabling adaptive differentiation between smooth skin lesions and messy clothing or background textures, avoiding overgrowth or undersegmentation in image segmentation. Furthermore, by combining inflammation activity and boundary confidence to generate state risk values, it achieves accurate identification of real pathological temperature rises and external heat source artifacts, improving the robustness of human state monitoring in complex dynamic environments.

[0008] Preferably, the expression for the inflammation activity index is:

[0009] Where I represents the inflammation activity index; and These represent the average temperatures of the abnormal hotspot region and the opposite normal region, respectively. This is the normalization constant; This represents the total number of pixels contained in the abnormal hotspot region. This represents the total number of pixels in the image for the body part containing the anomalous hot spot. The mean thermal gradient modulus within the anomalous hotspot region. As the first weighting coefficient, This is the second weighting coefficient.

[0010] By constructing an inflammation activity index that includes relative temperature difference, area ratio, and thermal gradient modulus, the single temperature threshold judgment is upgraded to a multi-dimensional quantitative assessment. This index not only reflects the thermal intensity of abnormal areas, but also comprehensively considers the extent of lesions and the active trend of their edges, thus enabling a more objective and comprehensive characterization of the severity of inflammation and reducing misjudgments caused by individual differences in basal body temperature.

[0011] Preferably, the boundary confidence level has a positive Sigmoid function relationship with the average thermal gradient magnitude within the abnormal hotspot region. When the average thermal gradient magnitude is greater than a set gradient threshold, the boundary confidence level approaches 1, and when the average thermal gradient magnitude is less than the set gradient threshold, the boundary confidence level approaches 0.

[0012] By using the Sigmoid function to establish a mapping relationship between boundary confidence and thermal gradient modulus, the interference problem of thermal reflection artifacts is effectively solved. Since real lesions usually have significant temperature gradient transitions, while environmental reflection or illumination often forms a uniform and flat temperature distribution, this technical solution can automatically assign high confidence to high gradient regions and low confidence to low gradient regions, thereby effectively suppressing false hot spot alarms.

[0013] Preferably, the expression for the entropy of biological thermal texture is:

[0014] Wherein, ent represents the normalized entropy of biological thermal texture; Represents the contrast of the gray-level co-occurrence matrix; This represents the energy of the gray-level co-occurrence matrix; It is a very small positive number; This represents the average temperature gradient magnitude within the neighborhood window; , These are the first and second adjustment coefficients, respectively, and tanh is the hyperbolic tangent function.

[0015] By fusing the contrast, energy, and temperature gradient information of the gray-level co-occurrence matrix, a biological thermal texture entropy is constructed, enabling precise quantification of the local texture complexity of an image. This metric can keenly capture the difference between high-frequency interference features such as clothing wrinkles and hair edges and smooth skin surface features, providing reliable data support for subsequent dynamic adjustment of regional growth resistance and solving the problem of target extraction in complex backgrounds.

[0016] Preferably, the expression for correcting the thermal symmetry difference value is:

[0017] in, Represents pixels Corrected thermal symmetry difference value at the location; and These represent the temperature fields of the left and right regions of interest, respectively. This represents the attenuation coefficient of infrared radiation in the atmosphere; This indicates the depth difference between the left and right limbs relative to the camera lens; This indicates the angle between the surface normal of the left and right limbs and the optical axis; To prevent tiny constants with a denominator of zero.

[0018] By compensating for radiation attenuation caused by depth distance with an exponential term and for radiation loss caused by surface normal angle with a cosine term, it is ensured that the calculated temperature difference is caused only by the asymmetry of internal human metabolism, rather than external posture error, thus improving the effectiveness of bilateral limb temperature comparison in unrestrained state.

[0019] Preferably, the method for obtaining the depth difference is as follows: using a monocular vision depth recovery algorithm, the depth difference is calculated based on the difference in projection scale between the left and right symmetrical regions in the visible light image.

[0020] Preferably, the monitoring method further includes using the ratio of a preset baseline allowable temperature difference to the current growth resistance coefficient as a threshold for determining regional growth.

[0021] A dynamic region growth criterion controlled by the growth resistance coefficient was designed to achieve adaptive adjustment of the segmentation threshold. In non-skin regions with complex textures, the growth resistance increases, causing the threshold to tighten and growth to stop quickly to prevent the introduction of noise. In skin regions with smooth textures, the threshold is relaxed to ensure complete extraction of lesions, thereby minimizing interference from the environmental background while ensuring the integrity of the lesions.

[0022] Preferably, before calculating the corrected thermal symmetry difference value, the method further includes registering the left and right symmetrical regions of interest, including: horizontally flipping the left region of interest along the central axis to obtain a flipped region; defining limb axis vectors based on skeleton key points in the flipped region and the right region of interest respectively; calculating the affine transformation matrix that maps the limb axis vectors of the flipped region to the limb axis vectors of the right region of interest; and applying the affine transformation matrix to the flipped region to obtain the registered left temperature field.

[0023] Before calculating thermal differences, an affine transformation registration step based on skeleton key points is introduced to solve the spatial misalignment problem caused by inconsistent movement postures of the left and right limbs. Through translation, rotation and scaling operations, pixel-level precise alignment of the left and right regions of interest is achieved, providing an accurate spatial reference for subsequent point-to-point thermal symmetry analysis.

[0024] Preferably, the OpenPose algorithm is used to perform pose assessment on visible light images to extract skeleton key points.

[0025] By using the OpenPose algorithm to extract skeleton key points, human joints can be located quickly and accurately from a single visible light image. This method is highly robust to changes in human posture, ensuring the accuracy of region of interest segmentation and providing a stable anatomical localization basis for the entire multimodal analysis process.

[0026] Secondly, this invention provides a human body condition monitoring system based on thermal imaging technology, employing the following technical solution: A human condition monitoring system based on thermal imaging technology includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the human condition monitoring method based on thermal imaging technology as described above.

[0027] The aforementioned human condition monitoring method based on thermal imaging technology is generated into a computer program and stored in a memory so that it can be loaded and executed by a processor. Thus, a system can be created based on the memory and processor for convenient use.

[0028] The present invention has the following technical effects: By constructing a physical geometry correction model, infrared radiation errors are compensated using the relative depth of the limbs and the normal angle, thus eliminating human posture interference. Biothermal texture entropy is introduced to quantify the complexity of the body surface texture and adaptively adjust the regional growth parameters, effectively removing clothing and background noise. A multi-dimensional risk assessment system including inflammation activity and boundary confidence is established to accurately distinguish between pathological temperature rise and environmental artifacts, solving the problem of high false alarm rate in human status monitoring under dynamic and complex environments. Attached Figure Description

[0029] Figure 1 This is a flowchart of the human body condition monitoring method based on thermal imaging technology of the present invention.

[0030] Figure 2 This is a diagram showing the effect of the calculation of the thermal symmetry difference value correction of the present invention.

[0031] Figure 3 This is a diagram showing the effect of adaptive parameter adjustment based on texture entropy calculation in this invention.

[0032] Figure 4 This is a diagram illustrating the effect of the present invention based on state risk value detection. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] This invention discloses a human body condition monitoring method based on thermal imaging technology, referring to... Figure 1 This includes the following steps: S1: Obtain the initial image and perform preprocessing.

[0035] Infrared thermal images of the human body under test are acquired using a dual-spectrum industrial camera. and visible light images Calculate infrared thermal images The temperature field of each pixel in the infrared thermal image, i.e., the temperature value corresponding to each pixel, is used to register the infrared thermal image using a pre-calibrated registration matrix. Mapped to visible light image In a coordinate system, pixel-level alignment is achieved. A deep learning-based pose assessment algorithm, such as OpenPose, is executed on the visible light image to extract key skeletal nodes of the human body, including the shoulder, elbow, wrist, hip, knee, and ankle. Based on the skeletal connection vectors, a human anatomical topology is constructed, and the human body is divided into... For a region of interest (ROI) that is symmetrical about left and right, it is denoted as ,in, For the left region of the k-th pair of left-right symmetrical regions of interest, For the right region of the k-th pair of symmetrical regions of interest, an affine transformation based on skeleton keypoints is used to perform pixel-level alignment of the symmetrical regions of interest. The method is as follows: the left region of interest extracted from the infrared image is horizontally mirrored along the vertical central axis to obtain the flipped region. Using key points of the human skeleton, limb axis vectors are defined for the flipped areas. (From the flipped left elbow point to the left wrist point) and the right reference limb axis vector ,For example: The vector of the right reference limb axis points from the left elbow point to the left wrist point in the flipped region. To calculate the direction from the right elbow point to the right wrist point, Mapped to Required geometric transformation matrix The matrix contains translation, rotation, and scaling components; specifically, the translation component moves the centroid of the left limb to coincide with the corresponding point on the right, and the rotation component rotates the left image so that... direction and Parallel; the scaling component scales the left image along the axis and normal to make its length and width consistent with the right region of interest.

[0036] Applying geometric transformation matrix right Spatial resampling is performed, specifically using a bilinear interpolation algorithm, to obtain the temperature field of the finally registered left-side region of interest. At this point, the temperature field of the region of interest on the left side... Each pixel in In terms of anatomical structure, they are all related to the temperature field of the region of interest on the right. Correspondingly.

[0037] S2: Construct a thermally symmetric difference map.

[0038] In actual standing or walking, the human body is not an ideal two-dimensional plane. The distances of the left and right limbs relative to the camera, as well as the normal angles caused by slight body rotations, are often inconsistent. This geometric difference leads to temperature measurement deviations in infrared radiation intensity even under normal physiological conditions due to atmospheric attenuation or Lambert's cosine law. To eliminate this non-pathological geometric error and obtain the temperature difference between the left and right symmetrical regions of interest, a correction for thermal symmetry difference needs to be calculated, expressed as follows:

[0039] in, Represents pixels Corrected thermal symmetry difference value at the location; and These represent the temperature fields of the left and right regions of interest, respectively. This represents the attenuation coefficient of infrared radiation in the atmosphere, which is set to 0.01 in this embodiment. This indicates the depth difference between the left and right limbs relative to the camera lens; This indicates the angle between the surface normal of the left and right limbs and the optical axis; To prevent the use of tiny constants with a denominator of zero, a value of 0.001 is used. Anatomical feature points of the human body are extracted through the visible light channel, and combined with a monocular vision depth recovery algorithm, the relative depth difference is inferred by utilizing the difference in projection scale of symmetrical parts of the human body on the image plane. Meanwhile, a local coordinate system based on the skeleton topology is established to calculate the vector angle between the limb surface normal vector and the optical axis of the infrared detector.

[0040] when When the number increases, it indicates that one limb is farther from the other limb, and the index is higher. It will adaptively amplify the original temperature difference to compensate for radiation attenuation caused by distance; similarly, when the angle difference... As the value increases, the denominator decreases, and the correction coefficient increases, thus compensating for the loss of radiation caused by the tilt of the viewing angle. In summary, this formula, through physical geometric correction, ensures that the corrected thermal symmetry difference value is caused only by the asymmetry of internal thermal metabolism, eliminating the interference of external posture errors. Each pixel corresponds to a corrected thermal symmetry difference value, and the corrected thermal symmetry difference values ​​of all pixels constitute a thermal symmetry difference map.

[0041] S3: Calculate the entropy of biological thermal texture.

[0042] Anomaly regions need to be extracted from the thermally symmetric difference map using a region growing algorithm. However, the human body surface is often covered with interference such as clothing and hair, and the textures of these regions are complex. If a fixed parameter is used for growing, oversegmentation is very likely to occur. Therefore, this step introduces biological thermal texture entropy to quantify the complexity of local regions.

[0043] First, select the point with the largest pixel value in the thermal symmetry difference map as the initial seed point. .by Built around For example, using a neighborhood window where N is 5, calculate the gray-level co-occurrence matrix and extract the contrast ratio. and energy Subsequently, the entropy of the biological thermal texture is calculated, expressed as:

[0044] Where ent represents the normalized entropy of the biological thermal texture, and its value ranges from 1 to 1. ; Represents the contrast of the gray-level co-occurrence matrix; This represents the energy of the gray-level co-occurrence matrix; It is a very small positive number, to prevent the denominator from being 0, its value is 0.01; This represents the average temperature gradient magnitude within the neighborhood window; The first adjustment coefficient, This is the second adjustment coefficient, with units equal to the reciprocal of the average temperature gradient modulus, used to eliminate the influence of dimensions. In this embodiment... , The values ​​are set to 0.5 and 0.5 respectively, and tanh() is the hyperbolic tangent function.

[0045] When the seed point is located in clothing folds or hair edges, the local texture becomes cluttered, resulting in poor contrast. Increase and energy The decrease, accompanied by a high gradient change, makes this... Increase, after After function mapping, The texture approaches 1; conversely, when the seed point is located on a smooth skin surface, the texture is uniform. Approaching 0.

[0046] In summary, The larger the value, the higher the probability that the area is a background disturbance. The seed point S area is more likely to be the edge of clothing or background, rather than a smooth skin lesion.

[0047] S4: Improve the region growth algorithm based on dynamic growth resistance coefficient.

[0048] In actual human monitoring scenarios, lesion areas typically exhibit smooth gradient changes, while background interference, such as clothing wrinkles and hair edges, exhibits high-frequency texture features. A single fixed threshold is insufficient to effectively suppress background leakage while ensuring the integrity of the lesion. Therefore, this step designs a dynamic growth resistance mechanism controlled by negative feedback of biothermal texture entropy. That is, by establishing a mapping relationship between local texture features and growth judgment threshold, the sensitivity of regional growth is adjusted in real time by calculating the dynamic growth resistance coefficient.

[0049] Calculate the dynamic growth resistance coefficient based on biothermal texture entropy. The expression is as follows:

[0050] in, This represents the current growth resistance coefficient; and These are the preset minimum and maximum resistance thresholds, which are set to 1.0 and 5.0 respectively in this embodiment. This represents the normalized entropy of biological thermal texture.

[0051] Then, region growing iterations are performed on the thermally symmetric difference map for the neighboring pixels to be examined. The criteria for classifying it into the abnormal region are as follows:

[0052] in, The temperature of the candidate pixel; This represents the average temperature of the currently grown area; The preset allowable temperature difference, for example, is 0.5℃. This represents the current growth resistance coefficient.

[0053] When the detection area has a complex texture, the biological thermal texture entropy ent approaches 1, and the growth resistance coefficient Z increases dramatically. This makes the threshold of the judgment criterion... The sensitivity of the region growing algorithm is significantly reduced. In areas with clothing or background, the algorithm becomes conservative and strict, allowing only minor temperature fluctuations, thus quickly stopping growth to prevent the inclusion of non-lesion areas. Conversely, in smooth skin areas, the growth resistance coefficient Z is smaller, and the threshold becomes larger, allowing the algorithm to completely extract the entire inflammatory plaque with a certain temperature gradient. In summary, this step achieves adaptive adjustment of the algorithm's sensitivity, ensuring lesion integrity while minimizing environmental noise interference.

[0054] S5: Calculate the inflammation activity index.

[0055] After region growth based on seed points and thresholds, abnormal hotspot regions are obtained. Then, it is necessary to further determine whether this region represents a genuine pathological reaction or a simple thermal reflection artifact. Here, a thermal reflection artifact refers to a false heat source that appears as an abnormally high temperature in an infrared thermogram but is actually formed by the reflection of environmental thermal radiation (such as light, sunlight, or the device's own radiation) from the surface of the object being measured. The method for distinguishing between genuine lesions and thermal reflection artifacts is to calculate the average thermal gradient modulus within the abnormal hotspot region. The boundary confidence score of the abnormal hotspot region is calculated using the following expression:

[0056] in, Indicates the confidence level of the boundary of the abnormal hot spot region; The slope parameter has a value range of [0.5, 5.0], and in this embodiment, the value is 2.0; This is the gradient threshold, for example, a value of 0.8. Due to the diffusion effect of subcutaneous heat, areas of pathological inflammation typically exhibit a significant gradient transition on thermograms. When the value of C approaches 1, it is considered a high confidence level; for spurious hotspots formed by environmental background or light reflection, their internal temperature distribution is often relatively uniform and flat. This leads to an increase in the exponential term, causing the value of C to decrease and approach 0, thereby effectively suppressing low-gradient artifacts.

[0057] The expression for the inflammation activity index is:

[0058] Among them, I is the inflammation activity index, which quantifies the severity of the lesions; and These represent the average temperatures of the abnormal hotspot region and the opposite normal region, respectively. This is a normalization constant used to eliminate the influence of dimensions; for example, its value is 1 degree Celsius. This represents the total number of pixels contained in the abnormal hotspot region. This refers to the total number of pixels in the image for the body part where the abnormal hot spot region is located, such as the entire forearm or the entire knee joint. The mean thermal gradient modulus within the anomalous hotspot region. This is the first weighting coefficient, which has no unit. This is the second weighting coefficient, its unit is the reciprocal of the average thermal gradient modulus, used to eliminate the influence of dimensions. and The sum is 1, for example. The value is 0.6. The value is 0.4.

[0059] In the formula, The abnormal heat intensity of the abnormal hot spot area was quantified. The larger the value, the greater the abnormal heat intensity of the abnormal hot spot area, and the greater the possibility that the abnormal hot spot area is a lesion area. Conversely, the smaller the value, the smaller the abnormal heat intensity of the abnormal hot spot area, and the less likely that the abnormal hot spot area is a lesion area, which may be a thermal reflection artifact. The value quantifies the breadth and activity trend of abnormal hot spot regions. The larger the value, the larger the area of ​​the abnormal hot spot region or the greater the temperature change, and the greater the possibility that it is a lesion region. Conversely, the smaller the value, the smaller the area of ​​the abnormal hot spot region or the smaller the temperature change, and the less likely that the abnormal hot spot region is a lesion region, that is, it may be a thermal reflection artifact.

[0060] S6: Calculate the state risk value to determine whether the corresponding area is abnormal.

[0061] A warning is issued in response to a status risk value exceeding a preset alarm threshold.

[0062] The expression for the state risk value is:

[0063] Where R represents the state risk value, The confidence level of the boundary of the abnormal hot spot region is represented by I, which is the inflammatory activity index.

[0064] R reflects the degree of risk of the abnormal hot spot area. The larger the value, the greater the degree of risk of the corresponding abnormal hot spot area; conversely, the smaller the value, the smaller the degree of risk of the corresponding abnormal hot spot area. If the value of R exceeds the preset alarm threshold, an audible and visual alarm signal is issued and the abnormal location is displayed. For example, the preset alarm threshold is 0.5.

[0065] For example: Suppose a person's left knee is found to have a high-temperature area.

[0066] Scenario A (Real Arthritis): Calculations show a large temperature difference and a significant edge gradient. ,lead to Meanwhile, due to the high temperature difference, the calculations show... .final The system correctly triggered the alarm.

[0067] Scenario B (Handheld hot water cup): Although the extreme temperature difference may cause I to reach as high as 1.2, the uniform temperature throughout the cup results in a low gradient. Model calculations show that its characteristics deviate from biological properties, leading to... .final It was determined to be an artifact, and no alarm was triggered.

[0068] This invention can also be illustrated with renderings, such as... Figure 2 As shown, the original observed temperature difference decreases exponentially with the increase of the depth difference between the left and right limbs, while the depth correction factor increases exponentially. The corrected thermal symmetry difference value after multiplying the two is represented by a stable solid line, indicating that step S2 successfully eliminated the temperature measurement deviation caused by incorrect standing position and restored the true physiological thermal difference.

[0069] like Figure 3 As shown, the growth resistance coefficient increases linearly with the increase of biological thermal texture entropy, while the dynamic growth threshold decreases with the increase of texture entropy. Smooth skin areas in the figure show low resistance, while complex texture areas show high resistance. This intuitively explains how step S4 automatically tightens the threshold in the clothing area to prevent false alarms.

[0070] like Figure 4 As shown, the horizontal axis represents the inflammation activity index, and the vertical axis represents the boundary confidence level. The actual lesions are concentrated in the upper right corner, located in the red area above the alarm threshold line, triggering the alarm.

[0071] Although the high-temperature artifact had a large horizontal axis value, its small vertical axis value fell below the threshold line, thus failing to trigger an alarm. This verified the effectiveness of the multiplication fusion formula in step S6, demonstrating that the method can effectively distinguish between high-temperature artifacts (such as hot water cups) and real inflammation.

[0072] This invention also discloses a human body condition monitoring system based on thermal imaging technology, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the human body condition monitoring method based on thermal imaging technology according to this invention.

[0073] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0074] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for monitoring human condition based on thermal imaging technology, characterized in that, Including the following steps: The process involves acquiring infrared and visible light images of the human body; registering the infrared thermal image to the coordinate system of the visible light image and extracting the temperature field; performing pose assessment on the visible light image to extract skeletal key points, and dividing the human body into multiple symmetrical regions of interest (ROIs) based on these key points; calculating the corrected thermal symmetry difference value between the symmetrical ROIs to obtain a thermal symmetry difference map; the corrected thermal symmetry difference value is positively correlated with the correction coefficient, which is positively correlated with the depth difference between the corresponding limb and the lens in the symmetrical ROIs, and negatively correlated with the cosine of the angle between the surface normal of the corresponding limb and the optical axis of the lens; performing region growing based on the point with the largest pixel value in the thermal symmetry difference map to obtain abnormal hot spot regions; the criteria for region growing include the growth resistance coefficient; the growth resistance coefficient is positively correlated with the biological thermal texture entropy; the biological thermal texture entropy is a normalized value calculated based on the contrast and energy of the gray-level co-occurrence matrix within the neighborhood window of the largest pixel, and the average modulus of the temperature gradient within the neighborhood window. The state risk value is obtained by multiplying the inflammatory activity index of the abnormal hot spot region and the boundary confidence score. An alarm signal is generated in response to a state risk value exceeding a set threshold.

2. The human body condition monitoring method based on thermal imaging technology according to claim 1, characterized in that, The expression for the inflammation activity index is: Where I represents the inflammation activity index; and These represent the average temperatures of the abnormal hotspot region and the opposite normal region, respectively. This is a normalization constant; This represents the total number of pixels contained in the abnormal hotspot region. This represents the total number of pixels in the image for the body part containing the anomalous hot spot. The mean thermal gradient modulus within the anomalous hotspot region. As the first weighting coefficient, This is the second weighting coefficient.

3. The human body condition monitoring method based on thermal imaging technology according to claim 1, characterized in that, The boundary confidence level has a positive Sigmoid function relationship with the average thermal gradient magnitude within the abnormal hot spot region. When the average thermal gradient magnitude is greater than the set gradient threshold, the boundary confidence level approaches 1, and when the average thermal gradient magnitude is less than the set gradient threshold, the boundary confidence level approaches 0.

4. The human body condition monitoring method based on thermal imaging technology according to claim 1, characterized in that, The expression for the entropy of biological thermal texture is: Wherein, ent represents the normalized entropy of biological thermal texture; Represents the contrast of the gray-level co-occurrence matrix; This represents the energy of the gray-level co-occurrence matrix; It is a very small positive number; This represents the average temperature gradient magnitude within the neighborhood window; , These are the first and second adjustment coefficients, respectively, and tanh is the hyperbolic tangent function.

5. The human body condition monitoring method based on thermal imaging technology according to claim 1, characterized in that, The expression for correcting thermal symmetry differences is: in, Represents pixels Corrected thermal symmetry difference value at the location; and These represent the temperature fields of the left and right regions of interest, respectively. This represents the attenuation coefficient of infrared radiation in the atmosphere; This indicates the depth difference between the left and right limbs relative to the camera lens; This indicates the angle between the surface normal of the left and right limbs and the optical axis; To prevent tiny constants with a denominator of zero.

6. The human body condition monitoring method based on thermal imaging technology according to claim 1, characterized in that, The method for obtaining the depth difference is as follows: using a monocular vision depth recovery algorithm, the depth difference is calculated based on the difference in projection scale between the left and right symmetrical regions in the visible light image.

7. The human body condition monitoring method based on thermal imaging technology according to claim 1, characterized in that, The monitoring method also includes using the ratio of the preset basic allowable temperature difference to the current growth resistance coefficient as the threshold for regional growth determination criteria.

8. The human body condition monitoring method based on thermal imaging technology according to claim 1, characterized in that, Before calculating the corrected thermal symmetry difference value, the process also includes registering the left and right symmetrical regions of interest, including: horizontally flipping the left region of interest along the central axis to obtain the flipped region; defining limb axis vectors based on skeleton key points in the flipped region and the right region of interest respectively; calculating the affine transformation matrix that maps the limb axis vectors of the flipped region to the limb axis vectors of the right region of interest; and applying the affine transformation matrix to the flipped region to obtain the registered left temperature field.

9. The human body condition monitoring method based on thermal imaging technology according to claim 1, characterized in that, The OpenPose algorithm is used to evaluate the pose of visible light images in order to extract skeleton key points.

10. A human condition monitoring system based on thermal imaging technology, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement the human body condition monitoring method based on thermal imaging technology according to any one of claims 1-9.

Images