Thermal imaging and visible light fused non-contact wild animal health monitoring method

Abstract

The invention belongs to the field of biomedical engineering, and particularly relates to a thermal imaging and visible light fused non-contact wild animal health monitoring method, which comprises the following steps of: processing a thermal imaging video frame by frame to identify a blood vessel dense region; mapping pixel coordinates of the blood vessel security level area to a corresponding position of the visible light image and taking the pixel coordinates as an ROI area; in the ROI area, calculating arithmetic mean values of R, G and B channels of all pixel points respectively, repeating the process at different time points t to obtain three light intensity curves R (t), G (t) and B (t) changing along with time, fusing the three light intensity curves to obtain a light intensity change curve I (t), performing heart rate calculation and blood pressure calculation according to the light intensity change curve I (t), and performing blood oxygen calculation by using R (t) and B (t); and monitoring the health state of the wild animal according to a calculation result. According to the invention, animal health non-damage monitoring is realized.

Description

Technical Field

[0001] The present invention belongs to the field of biomedical engineering, and specifically relates to a non-contact wildlife health monitoring method that integrates thermal imaging with visible light. Background Art

[0002] During a field visit to the Chengdu Giant Panda Breeding Base and research on pet health monitoring products, it was discovered that most existing animal health monitoring devices are similar to human medical devices, often requiring close contact with the animal during the measurement process. For example, veterinary blood pressure monitors are equipped with armbands of varying sizes to accommodate blood pressure measurements in animals of all sizes. For rare animals like giant pandas, the forelimbs are often used as the measurement site. Veterinary oximeters use clips to clamp onto the animal's ears, tongue tip, and other parts of the body to obtain blood oxygen concentration and heart rate data. However, these traditional measurement methods have significant drawbacks. Firstly, they require extremely high levels of animal cooperation. If the animal experiences a stress response, not only can the measurement data fluctuate and inaccurate, but it can even render the measurement impossible. Secondly, stressed animals can attack operators or damage the measuring instruments. Especially for highly aggressive wild animals, performing such contact measurements while they are awake is almost impossible. These limitations make traditional devices difficult to meet the health monitoring needs of wildlife rescue scenarios. Summary of the Invention

[0003] The purpose of the present invention is to provide a non-contact wildlife health monitoring method that integrates thermal imaging and visible light. This method does not require close contact with animals, which not only avoids the measurement obstacles and potential risks caused by animal stress reactions, but also provides safe and effective technical support for wildlife health detection and rescue work.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] A non-contact wildlife health monitoring method using thermal imaging and visible light fusion includes the following steps:

[0006] Step 1: Collect visible light video and thermal imaging video of the same area;

[0007] Step 2: Process the thermal imaging video frame by frame to identify areas with dense blood vessels. Simultaneously, perform noise reduction and illumination normalization on the visible light video frame by frame to eliminate noise and improve illumination consistency.

[0008] Step 3: Use the visible light and thermal imaging video matching algorithm to map the pixel coordinates of the vascular density area to the corresponding position in the visible light image and use it as the ROI area;

[0009] Step 4: Calculate the arithmetic mean of the R, G, and B channels of all pixels in the ROI area, and repeat this process at different time points t to obtain three time-varying light intensity curves R(t), G(t), and B(t), where R(t) represents the change of the average pixel value of the red channel over time, G(t) represents the change of the average pixel value of the green channel over time, and B(t) represents the change of the average pixel value of the blue channel over time. A weighted fusion method is used to assign different weights to R(t), G(t), and B(t) according to the degree of reflection of different physiological signals in each channel, and then perform weighted summation to obtain the fused light intensity change curve I(t).

[0010] Step 5: Calculate the heart rate and blood pressure based on the light intensity change curve I(t) obtained in step 4, and calculate the blood oxygen level using R(t) and B(t); monitor the health status of the wild animal based on the calculation results.

[0011] Furthermore, the specific process of step 2 of processing the thermal imaging video frame by frame to identify the blood vessel dense area and perform morphological operations includes:

[0012] Step 2.1: Extract facial foreground targets. Each frame of the thermal imaging video is grayscaled to generate a grayscale image. The grayscale image is then converted into a binary image using the Otsu binarization algorithm and morphologically manipulated to obtain the binary foreground target. The binary foreground target is then masked with the original thermal image to obtain the final facial foreground target of the wild animal.

[0013] Step 2.2: Calculate the temperature gradient. Treat each frame of the thermal imaging video as a two-dimensional matrix, where each matrix element corresponds to a pixel in the image, and its value represents the temperature detected at that pixel. Use the Sobel operator to calculate the horizontal and vertical temperature gradients of this matrix and synthesize them to obtain a gradient magnitude image.

[0014] In step 2.3, after normalizing the gradient amplitude image, the regions where the pixels are higher than the set threshold are extracted and morphological operations are performed on them, and finally the regions with dense blood vessel distribution are obtained.

[0015] Furthermore, the specific process of using visible light and thermal imaging video matching algorithm in step 3 to map the pixel coordinates of the blood vessel density area to the corresponding position of the visible light image and use it as the ROI area includes:

[0016] Step 3.1: Use a checkerboard or dot grid to calibrate the internal parameters of the visible light camera and thermal imaging camera, respectively. Then, perform spatial registration using a rigid spatial alignment algorithm to obtain the rotation matrix R and translation vector T between the two. The initial homography matrix H0 is constructed based on the rotation matrix R and translation vector T. The internal parameters include focal length, principal point position, and lens distortion internal parameters.

[0017] Step 3.2: Obtain the resolution of the thermal imaging camera and the visible light camera, extract the width and height parameters of the two images respectively, and then calculate the scaling coefficients Sx and Sy of the thermal imaging image relative to the visible light image in the horizontal and vertical coordinates;

[0018] Step 3.3: Normalize the pixels of the thermal image and the visible light image respectively. Based on the resolution ratios Sx and Sy calculated in step 3.2, map the normalized thermal image to the normalized visible light image.

[0019] Step 3.4, construct the objective function, and based on the objective function, use the Levenberg–Marquardt or ICP algorithm to iteratively update the homography matrix H0 and the scaling coefficient;

[0020] In step 3.5, the updated homography matrix H0 is used to extract the pixel points of the vascular region from the thermal image and project them into the visible light image using the coordinate transformation matrix. Then, based on the spatial distribution of the pixel points in the vascular region, the minimum enclosing rectangle is calculated to represent the geometric boundary of the region, thus completing the mapping from the thermal image to the visible light image.

[0021] Furthermore, in step 2, the illumination normalization method for the visible light image is histogram equalization or adaptive histogram equalization, and the noise reduction method for the visible light image is Gaussian filtering or median filtering.

[0022] Furthermore, in step 4, a low-pass filter or a band-pass filter is used to remove high-frequency noise during the extraction of the light intensity curves R(t), G(t), and B(t), and a chrominance-based method is used to suppress the influence of light intensity fluctuations.

[0023] Furthermore, the specific process of performing heart rate calculation, blood pressure calculation and blood sample calculation based on the obtained physiological signal data in step 5 includes:

[0024] Heart rate calculation: The light intensity variation curve I(t) is filtered using a Butterworth passband filter. Based on the periodic characteristics of the heart rate signal, the time domain signal is transformed into a frequency domain signal using a fast Fourier transform or a wavelet transform. The frequency value corresponding to the main frequency is determined by analyzing the spectral characteristics of the frequency domain signal, and the heart rate value is converted from this frequency value.

[0025] Blood pressure calculation: Two locations within the ROI are selected based on the blood flow path. By analyzing the time difference between the light intensity change curve I(t) reaching these two locations, the pulse wave propagation time PTT along the planned path is calculated. The systolic pressure value is estimated using the formula SBP to complete the blood pressure calculation; the formula SBP = a·PTT -b +c, where a, b, and c are empirical parameters obtained by experimental calibration;

[0026] Blood oxygen calculation: First, perform window processing on the light intensity signal curves R(t) and B(t), calculate their standard deviation AC and average value DC respectively; and substitute these into the blood oxygen saturation formula to obtain the blood oxygen value.

[0027] After adopting the above technical solution, the present invention has the following advantages:

[0028] 1. This invention captures wildlife images solely through thermal imaging cameras and optical cameras, and achieves a non-contact monitoring method by fusing images from other videos. No close contact with the animals is required during the entire monitoring period, making wildlife health monitoring safer and more efficient, effectively avoiding measurement obstacles and potential risks caused by stress reactions in animals, and addressing the limitations of traditional equipment in wildlife health monitoring scenarios.

[0029] 2. The present invention optimizes the processing flow of thermal imaging videos and visible light videos. It uses thermal imaging videos to identify areas with dense blood vessels, and then uses a matching algorithm for thermal imaging videos and visible light videos to map the dense blood vessel areas into visible light videos to extract the ROI area. Since the improved processing flow can accurately locate the target area and improve image quality, the subsequent physiological signal extraction is more accurate and reliable, thereby improving the accuracy of heart rate, blood pressure and blood oxygen calculations, providing more scientific data support for wildlife health status monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is a flow chart of a non-contact wildlife health monitoring method according to an embodiment;

[0031] Figure 2 This is the algorithm flow for matching visible light and thermal imaging videos. DETAILED DESCRIPTION

[0032] The technical solution of the present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0033] In the field of physiological signal monitoring, visible light cameras and thermal imaging cameras are two common non-contact heart rate detection devices, each with its own advantages and disadvantages.

[0034] Visible light cameras, with their widespread availability and low cost, are widely used in numerous fields. They can capture RGB images, providing essential data for subsequent physiological signal extraction. Complex algorithms can be used to extract IPPG signals from RGB images. Because hemoglobin in blood strongly absorbs green light, the green light channel is often the primary focus of analysis. As the heart contracts and relaxes, blood volume undergoes periodic fluctuations, causing fluctuations in reflected light intensity. These fluctuations contain physiological information such as heart rate. However, visible light cameras have significant limitations. First, they are extremely sensitive to lighting conditions. In strong light, images can easily become overexposed, leading to pixel saturation and loss of information about blood volume changes. In low light, the image signal-to-noise ratio decreases, compromising signal extraction accuracy. Second, while high resolution is an inherent characteristic, it can be a disadvantage in physiological signal monitoring. The rich image details can cause subtle movements such as breathing and blinking to generate motion artifacts, which can interfere with heart rate signals and complicate signal separation. Furthermore, the large amount of image data significantly increases the computational complexity of subsequent processing, placing extremely high demands on computing resources and processing speed, severely limiting their real-time monitoring performance.

[0035] Thermal imaging cameras extract heart rate signals by capturing skin temperature distribution. Facial temperature in animals fluctuates periodically due to the propagation of arterial waves, which correlates with heartbeats. Thermal imaging cameras can accurately capture these changes. The device's significant advantage lies in its unaffected lighting conditions, operating stably in both low-light and dark environments. This significantly expands the application scenarios for physiological signal monitoring, such as nighttime medical monitoring and motion monitoring in dark environments. However, thermal imaging cameras also have drawbacks. For one thing, their resolution is low, and a single pixel covers a large area of ​​skin, which can easily miss subtle information about local temperature changes. Furthermore, the devices are expensive. This is primarily due to the high production cost of the core thermal detector component—which requires specialized materials and precision manufacturing processes to achieve high sensitivity to even tiny temperature differences. This results in a high price tag, limiting widespread adoption.

[0036] This embodiment aims to address the advantages and disadvantages of thermal imaging cameras and visible light cameras in use, fuses the images of the two cameras, and proposes a non-contact wildlife health monitoring method that fuses thermal imaging and visible light. Figure 1 As shown, the following steps are included:

[0037] Step 1: Collect visible light video and thermal imaging video of the same area.

[0038] In step 2, the thermal imaging video is processed frame by frame to identify areas with dense blood vessels. At the same time, the visible light video is subjected to noise reduction and illumination normalization frame by frame to eliminate noise and improve illumination consistency.

[0039] The specific process of processing thermal imaging videos frame by frame to identify areas with dense blood vessels includes:

[0040] Step 2.1, extracting facial foreground targets: This embodiment extracts facial foreground targets for two purposes: first, to meet real-time requirements; second, to effectively deal with the situation where the wild animal is slightly shaking, providing an accurate target area for subsequent heart rate measurement. The specific operation is as follows:

[0041] In step 2.1.1, grayscale processing is performed on each frame of the thermal imaging video in turn to generate a grayscale image to reduce the amount of data and reduce the computational complexity.

[0042] Step 2.1.2: Use the Otsu binarization algorithm to convert the grayscale image into a binary image. The specific conversion process is as follows:

[0043] Assume that the gray level of the image is L and the inter-class variance between the foreground and background is σ 2 ; Traverse all thresholds and find the class variance σ 2 The maximum threshold is calculated as follows:

[0044] σ 2 =ω0ω1(μ0-μ1) 2 ;

[0045] Among them, ω0 is the proportion of foreground pixels, ω1 is the proportion of background pixels, μ0 is the grayscale mean of foreground pixels, and μ1 is the grayscale mean of background pixels.

[0046] In step 2.1.3, morphological operations are performed on the binary image to remove noise and small holes and enhance the connectivity of the target area. The morphological operations in this embodiment include erosion and dilation operations. After the erosion and dilation operations, a contour extraction method is applied to the image to obtain the target boundary information and obtain the foreground object in the binary image.

[0047] In step 2.1.4, the foreground target of the binary image is subjected to a mask bit operation with the original image. That is, each pixel in the original image is extracted according to the set rules. If the corresponding position of the pixel in the binary image is a foreground pixel, the original color information of the pixel is retained; otherwise, it is set to the background color, thereby finally obtaining the facial foreground target of the wild animal.

[0048] Step 2.2: Calculate the temperature gradient and obtain the area with dense blood vessel distribution based on the temperature gradient. The implementation process is as follows:

[0049] Step 2.2.1: Treat each frame of the thermal imaging video as a two-dimensional matrix, where each matrix element corresponds to a pixel in the image, and its value represents the temperature value detected at that pixel;

[0050] In step 2.2.2, the Sobel operator is used to calculate the horizontal and vertical temperature gradients of the matrix and synthesize them to obtain a gradient magnitude image. The Sobel operator, a classic edge detection operator, calculates the gradient approximation of the image grayscale value through convolution. In this embodiment, this operator is used to calculate the temperature gradient.

[0051] The calculation formula for the transverse temperature gradient is:

[0052] G x (i,j)=T(i,j+1)-T(i,j-1);

[0053] The calculation formula for the longitudinal temperature gradient is:

[0054] G y (i,j)=T(i+1,j)-T(i-1,j);

[0055] Where T(i,j) represents the temperature value of the pixel in the i-th row and j-th column in the thermal imaging image.

[0056] The calculated transverse temperature gradient and longitudinal temperature gradient are synthesized to obtain the gradient amplitude image G mag , the specific operation formula is:

[0057]

[0058] Through the gradient magnitude image G mag The temperature change rate of each pixel in the thermal imaging image in the horizontal and vertical directions can be obtained.

[0059] Step 2.2.3: Normalize the gradient magnitude image so that its pixel values ​​are in the range of [0,1]. The normalization formula is:

[0060]

[0061] Among them, G max is the minimum value in the gradient magnitude image, G min is the maximum value in the gradient magnitude image.

[0062] Because areas with dense blood vessels in normalized thermal images often exhibit a cluster of high-gradient-amplitude pixels, this characteristic can be used to extract regions with pixels above a set threshold from the normalized gradient-amplitude image as initial regions of dense blood vessel distribution. The threshold is set based on the actual thermal image characteristics and vascular distribution; in this example, the threshold is set to 0.8.

[0063] In step 2.2.4, morphological processing is performed on the initial densely vascularized region to remove noise points or small non-vascular areas, smooth region boundaries, and connect adjacent vascular regions to obtain the final densely vascularized region. This example first performs a dilation operation using a 3×3 square structuring element. After multiple dilations expand the region, an erosion operation is then performed to reduce the region and remove small branches and noise.

[0064] Normalizing visible light images can be achieved using histogram equalization or adaptive histogram equalization. This embodiment uses histogram equalization to normalize visible light images, ensuring a relatively consistent brightness distribution under varying lighting conditions. Denoising visible light images can be achieved using methods such as Gaussian filtering and median filtering. This embodiment uses Gaussian filtering to remove image noise.

[0065] Step 3: Use the visible light and thermal imaging video matching algorithm to map the pixel coordinates of the blood vessel density area after morphological operation to the corresponding position of the visible light image and use it as the ROI area; the implementation process is as follows: Figure 2 As shown, including:

[0066] Step 3.1: Camera calibration. Use a checkerboard or dot grid to calibrate the internal parameters of the visible light camera and thermal imaging camera, respectively. Then, use a rigid spatial alignment algorithm to spatially align the two cameras, obtaining the rotation matrix R and translation vector T between them. The initial homography matrix H0 is constructed based on the rotation matrix R and translation vector T. In this embodiment, the camera internal parameters include focal length, principal point position, and lens distortion internal parameters.

[0067] Step 3.2, resolution difference calculation, obtains the resolution of the thermal imaging camera and the visible light camera, extracts the width and height parameters of the two images respectively, and then calculates the scaling coefficients Sx and Sy of the thermal imaging image relative to the visible light image in the horizontal and vertical coordinates. The specific calculation formula is as follows:

[0068]

[0069] Among them, s x represents the horizontal scaling factor of the thermal image relative to the visible light image, s y W represents the vertical scaling factor of the thermal image relative to the visible light image.v Indicates the width of the visible light image, Represents the width of the thermal imaging image after the rotation matrix R, H v represents the height of the visible light image, Represents the height of the thermal imaging image after rotation by the rotation matrix R.

[0070] Step 3.3, coordinate normalization and scale mapping, normalize the pixel coordinates of the thermal imaging image and the visible light image to the range of [0, 1] respectively. According to the scaling coefficients Sx and Sy calculated in step S3.2, map the normalized thermal imaging image to the normalized visible light image to achieve preliminary geometric alignment of cross-modal images.

[0071] The thermal imaging image pixel normalization formula is as follows:

[0072]

[0073] Among them, xi represents the horizontal coordinate value of the normalized thermal imaging image, u i represents the horizontal pixel position of the original thermal imaging image, yi represents the vertical coordinate value of the normalized thermal imaging image, and v i Indicates the vertical pixel position of the original thermal imaging image.

[0074] The formula for normalizing visible light image pixels is as follows:

[0075]

[0076] x v Indicates the horizontal coordinate value of the normalized visible light image, y v Indicates the vertical coordinate value of the normalized visible light image, u v Indicates the horizontal pixel position of the original visible light image, v v Indicates the vertical pixel position of the original visible light image, W v Indicates the width of the visible light image, H v Indicates the height of the visible light image.

[0077] The formula for mapping to the visible light normalized plane according to the scaling factor is:

[0078] x v =s x x i ;

[0079] y v =s y y i ;

[0080] Among them, x vIndicates the corresponding horizontal coordinate value of the normalized visible light image, y v Indicates the corresponding vertical coordinate value of the normalized visible light image.

[0081] Step 3.4, construct an objective function, and based on the objective function, use the Levenberg–Marquardt or ICP algorithm to iteratively update the homography matrix H0 and the scaling coefficient; the objective function is:

[0082]

[0083] in, Indicates applying the homography matrix H to the kth pixel in the thermal imaging image The mapping coordinates obtained by the normalized coordinate vector (xi, yi) are the predicted positions of the thermal imaging pixels in the visible light image coordinate system after homography transformation; Represents the normalized coordinate vector (x v ,y v );Symbol ∑ k Indicates the summation operation of all corresponding pixels, k is the index of the pixel point, and traverses all pixel pairs involved in the calculation; λ represents the regularization parameter, which is used to balance the weight between the data fitting term and the regularization term. The data fitting term is the pixel coordinate error term, and the regularization term is the scaling coefficient penalty term, so as to prevent the model from overfitting.

[0084] In step 3.5, the updated homography matrix H0 is used to extract the pixel points of the vascular region from the thermal image and project them into the visible light image using the coordinate transformation matrix. Then, based on the spatial distribution of the pixel points in the vascular region, the minimum enclosing rectangle is calculated to represent the geometric boundary of the region, thus completing the mapping from the thermal image to the visible light image.

[0085] Step 4: Calculate the arithmetic mean of the red R, green G, and blue B channels of all pixels in the ROI area, and repeat the process at different time points t to obtain three light intensity change curves R(t), G(t), and B(t) that change with time. R(t) represents the change of the average pixel value of the red channel over time, G(t) represents the change of the average pixel value of the green channel over time, and B(t) represents the change of the average pixel value of the blue channel over time. The specific conversion formula of R(t) is:

[0086]

[0087] Where R(x,y,t) represents the red channel value of the pixel at coordinate (x,y) at time t. N is the number of pixels in the ROI area. It means that all pixels within the ROI are summed and averaged; similarly, G(t) and B(t) can be obtained.

[0088] The weighted fusion method is used to obtain the fused light intensity change curve I(t). The calculation formula is as follows:

[0089] I(t)=wr·R(t)+wg·G(t)+wb·B(t)

[0090] Where wr, wg, and wb are the weight coefficients of R(t), G(t), and B(t), respectively, and satisfy wr + wg + wb = 1. The weight distribution can be determined based on the analysis of a large amount of experimental data.

[0091] Step 5: Calculate the heart rate and blood pressure based on the light intensity change curve I(t), calculate the blood oxygen level using R(t) and B(t), and monitor the health status of the wild animal based on the calculation results.

[0092] Heart rate calculation:

[0093] I(t) is filtered using a Butterworth passband filter; its transfer function is:

[0094]

[0095] Among them, f c is the center frequency, W is the passband width, and n is the filter order.

[0096] Combined with wavelet threshold denoising, let the wavelet coefficient be d and the threshold be λ, then the soft threshold processing is:

[0097] d′=sign(d)·max(|d|-λ,0);

[0098] Finally, the median filter output is:

[0099] y[i]=median(x[ik],x[i-k+1],…,x[i+k]);

[0100] Where k is the window radius.

[0101] Based on the periodic characteristics of the heart rate signal, the time domain signal is transformed into a frequency domain signal using fast Fourier transform or wavelet transform. The frequency value corresponding to the main frequency is determined by analyzing the spectral characteristics of the frequency domain signal, and the heart rate value is converted based on this.

[0102] Blood pressure calculation:

[0103] Two positions within the ROI area are selected based on the path of blood flow. The direction of blood flow through the face is generally from top to bottom, so the "timing delay" of the two waveforms can be observed. Therefore, the two positions selected in this embodiment are divided into the facial foreground of the wild animal, corresponding to the upper and lower cheek edges of the animal. By analyzing the time difference between the pulse wave arriving at these two positions, the direction of pulse wave propagation is determined; the measurement path is planned according to the direction of pulse wave propagation, and the propagation time PTT of the pulse wave on the selected path is calculated. The formula SBP is then used to estimate the systolic pressure value to complete the blood pressure calculation; the formula SBP = a·PTT -b +c, where a, b, and c are empirical parameters obtained by experimental calibration.

[0104] Blood oxygen calculation:

[0105] First, perform window processing on the light intensity signal curves R(t) and B(t), calculate their standard deviation AC and average value DC respectively, and substitute them into the blood oxygen saturation formula to calculate the blood oxygen value; the blood oxygen saturation formula is:

[0106] Where SpO2 represents blood oxygen saturation, A and B are constants obtained by fitting experimental or clinical data, and A = 125 and B = 26 are taken as empirical parameters; AC blue Indicates the standard deviation of the blue channel, DC red Represents the mean of the red channel, AC red Indicates the standard deviation of the red channel, DC blue Represents the mean of the blue channel.

[0107] To achieve real-time updates, this embodiment slides the window every 1 to 2 seconds, calculates AC / DC, and then substitutes it into the formula to output SpO2.

[0108] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by one of ordinary skill in the art without departing from the spirit and technical principles disclosed herein are intended to be covered by the claims of the present invention.

Claims

1. A non-contact wildlife health monitoring method that combines thermal imaging with visible light, characterized in that: The following steps are involved: Step 1: Collect visible light video and thermal imaging video of the same area; Step 2: Process the thermal imaging video frame by frame to identify areas with dense blood vessels, and perform noise reduction and illumination normalization on the visible light video frame by frame. Step 3: Use the visible light and thermal imaging video matching algorithm to map the pixel coordinates of the vascular density area to the corresponding position in the visible light image and use it as the ROI area; Step 4: Calculate the arithmetic mean of the R, G, and B channels of all pixels in the ROI area, and repeat this process at different time points t to obtain three time-varying light intensity curves R(t), G(t), and B(t). Use a weighted fusion method to assign different weights to R(t), G(t), and B(t) according to the degree of reflection of different physiological signals in each channel, and then perform weighted summation to obtain the fused light intensity change curve I(t). Step 5: Calculate the heart rate and blood pressure based on the light intensity change curve I(t) obtained in step 4, and calculate the blood oxygen level using R(t) and B(t); monitor the health status of the wild animal based on the calculation results.

2. The non-contact wildlife health monitoring method using thermal imaging and visible light fusion according to claim 1 is characterized in that: The following steps are involved: Step 2.1: Extract facial foreground targets. Each frame of the thermal imaging video is grayscaled to generate a grayscale image. The grayscale image is then converted into a binary image using the Otsu binarization algorithm and morphologically manipulated to obtain the binary foreground target. The binary foreground target is then masked with the original thermal image to obtain the final facial foreground target of the wild animal. Step 2.2: Calculate the temperature gradient. Treat each frame of the thermal imaging video as a two-dimensional matrix, where each matrix element corresponds to a pixel in the image, and its value represents the temperature value detected at that pixel. Use the Sobel operator to calculate the horizontal and vertical temperature gradients of this matrix and synthesize them to obtain a gradient magnitude image. In step 2.3, after normalizing the gradient amplitude image, the regions where the pixels are higher than the set threshold are extracted and morphological operations are performed on them, and finally the regions with dense blood vessel distribution are obtained.

3. The non-contact wildlife health monitoring method combining thermal imaging and visible light according to claim 1 is characterized in that: The specific process of using the visible light and thermal imaging video matching algorithm in step 3 to map the pixel coordinates of the blood vessel density area to the corresponding position of the visible light image and use it as the ROI area includes: Step 3.1: Use a checkerboard or dot board to calibrate the internal parameters of the visible light camera and thermal imaging camera, respectively. Then, perform spatial registration using a rigid spatial alignment algorithm to obtain the rotation matrix R and translation vector T between the two. The initial homography matrix H0 is constructed based on the rotation matrix R and translation vector T. Step 3.2: Obtain the resolution of the thermal imaging camera and the visible light camera, extract the width and height parameters of the two images respectively, and then calculate the scaling coefficients Sx and Sy of the thermal imaging image relative to the visible light image in the horizontal and vertical coordinates; Step 3.3: Normalize the pixels of the thermal image and the visible light image respectively. Based on the resolution ratios Sx and Sy calculated in step 3.2, map the normalized thermal image to the normalized visible light image. Step 3.4, construct the objective function, and based on the objective function, use the Levenberg–Marquardt or ICP algorithm to iteratively update the homography matrix H0 and the scaling coefficient; In step 3.5, the updated homography matrix H0 is used to extract the pixel points of the vascular region from the thermal image and project them into the visible light image using the coordinate transformation matrix. Then, based on the spatial distribution of the pixel points in the vascular region, the minimum enclosing rectangle is calculated to represent the geometric boundary of the region, thus completing the mapping from the thermal image to the visible light image.

4. The non-contact wildlife health monitoring method using thermal imaging and visible light fusion according to claim 1 is characterized in that: In step 2, the illumination normalization method for the visible light image is histogram equalization or adaptive histogram equalization, and the noise reduction method for the visible light image is Gaussian filtering or median filtering.

5. The non-contact wildlife health monitoring method combining thermal imaging and visible light according to claim 1 is characterized in that: In the process of extracting the light intensity curves R(t), G(t) and B(t), step 4 also uses a low-pass filter or a band-pass filter to remove high-frequency noise, and uses an image chromaticity information analysis method to suppress the influence of light intensity fluctuations.

6. A non-contact wildlife health monitoring method combining thermal imaging and visible light according to any one of claims 1 to 5, characterized in that: The specific process of performing heart rate calculation, blood pressure calculation and blood sample calculation based on the obtained physiological signal data in step 5 includes: Heart rate calculation: The light intensity variation curve I(t) is filtered using a Butterworth passband filter. Based on the periodic characteristics of the heart rate signal, the time domain signal is transformed into a frequency domain signal using a fast Fourier transform or a wavelet transform. The frequency value corresponding to the main frequency is determined by analyzing the spectral characteristics of the frequency domain signal, and the heart rate value is converted from this frequency value. Blood pressure calculation: Two locations within the ROI are selected based on the blood flow path. By analyzing the time difference between the light intensity change curve I(t) reaching these two locations, the pulse wave propagation time PTT along the planned path is calculated. The systolic pressure value is estimated using the formula SBP to complete the blood pressure calculation; the formula SBP = a·PTT -b +c, where a, b, and c are empirical parameters obtained by experimental calibration; Blood oxygen calculation: First, perform window processing on the light intensity signal curves R(t) and B(t), calculate their standard deviation AC and average value DC respectively; and substitute these into the blood oxygen saturation formula to obtain the blood oxygen value.

Images