Automatic subcutaneous vein blood sampling method based on infrared imaging

Abstract

The invention discloses an automatic subcutaneous vein blood sampling method based on infrared imaging, and relates to the technical field of vein blood sampling, and the method comprises the following steps: obtaining infrared thermal imaging original data of a limb surface skin area to be subjected to blood sampling, and generating a two-dimensional temperature matrix reflecting subcutaneous tissue heat distribution through temperature field reconstruction; the temperature gradient direction and the thermal diffusion rate of the vascular tissue are extracted and input into a preset thermodynamic causal reasoning model, and subcutaneous vein three-dimensional space coordinates are inverted according to a heat transfer physical rule; calculating a vein diameter and a depth estimation value in combination with the temperature value, performing geometric correction on the three-dimensional coordinate to obtain a vein center coordinate, and generating vein position data under a world coordinate system through coordinate system mapping. According to the method, subcutaneous vein three-dimensional distribution can be accurately captured, positioning errors are compensated, the accuracy of vein position data is improved, the problem of positioning deviation in the conventional technology is solved, and more accurate vein positioning is achieved.

Description

Technical Field

[0001] This invention belongs to the field of venous blood collection technology, specifically an automated subcutaneous venous blood collection method based on infrared imaging. Background Technology

[0002] Currently, infrared imaging technology is widely used for vein identification in automated subcutaneous vein blood collection technology. Existing technologies mostly acquire infrared images of the surface of the limb to be blood collected and rely on the difference in image grayscale to identify the approximate location of veins. Some technologies perform simple filtering and noise reduction on the raw infrared data, which can only output two-dimensional projection information of veins on the body surface and cannot accurately reflect the three-dimensional spatial distribution of veins under the skin.

[0003] Existing technologies do not fully consider the differences in thermal conductivity between vascular tissue and surrounding subcutaneous tissue. They cannot accurately capture key parameters such as the depth and diameter of veins by relying solely on image grayscale features, nor can they deduce the three-dimensional spatial coordinates of veins. Furthermore, they do not perform targeted geometric correction on the identified vein locations, resulting in a deviation in the mapping between the infrared image coordinate system and the physical spatial coordinate system, leading to insufficient accuracy in vein localization.

[0004] Existing technologies struggle to accurately obtain the three-dimensional spatial location of subcutaneous veins using infrared imaging data, and cannot optimize vein location accuracy through reasonable correction methods. This results in insufficient positioning accuracy during automated blood collection, failing to meet the precise positioning requirements of automated blood collection and affecting blood collection efficiency and safety. Summary of the Invention

[0005] This invention aims to solve at least one of the technical problems existing in the prior art;

[0006] Therefore, this invention proposes an automated subcutaneous venous blood collection method based on infrared imaging, comprising:

[0007] Acquire raw infrared thermal imaging data of the target area, which is the surface skin area of ​​the limb to be blood collected;

[0008] The raw infrared thermal imaging data is processed to reconstruct the temperature field, generating a two-dimensional temperature matrix that reflects the thermal distribution of subcutaneous tissue.

[0009] Based on the two-dimensional temperature matrix, the thermal conduction characteristics of vascular tissue are extracted, including the temperature gradient direction and the thermal diffusion rate.

[0010] The temperature gradient direction and the thermal diffusion rate are input into a preset thermodynamic causal reasoning model, which, based on the physical laws of heat transfer from the core region of the body to the body surface, inversely derives the three-dimensional spatial coordinates of the subcutaneous veins.

[0011] Based on the three-dimensional spatial coordinates and the temperature values ​​in the two-dimensional temperature matrix, the estimated diameter and depth of the subcutaneous vein are calculated.

[0012] Based on the estimated diameter and the estimated depth, the three-dimensional spatial coordinates of the subcutaneous vein are geometrically corrected to obtain the corrected coordinates of the vein center.

[0013] Using the corrected vein center coordinates, a mapping transformation is performed between the infrared image coordinate system and the physical space coordinate system to generate vein location data in the world coordinate system.

[0014] Further, the raw infrared thermal imaging data undergoes temperature field reconstruction processing to generate a two-dimensional temperature matrix reflecting the thermal distribution of subcutaneous tissue, including:

[0015] Non-uniformity correction is performed on the raw infrared thermal imaging data to eliminate fixed pattern noise introduced by the detector's own characteristics in the raw infrared thermal imaging data;

[0016] An adaptive Gaussian filtering algorithm is used to smooth and denoise the corrected raw infrared thermal imaging data while preserving the temperature abrupt changes at the edge of blood vessels.

[0017] According to Planck's blackbody radiation law, the smoothed and denoised grayscale image data is converted into absolute temperature values ​​to obtain the initial temperature distribution map.

[0018] Detect abnormally high or low temperature pixels in the initial temperature distribution map and replace them with the median temperature of the surrounding neighboring pixels to complete the temperature field repair.

[0019] The repaired initial temperature distribution map is resampled to a preset resolution to generate the two-dimensional temperature matrix.

[0020] Furthermore, based on the two-dimensional temperature matrix, the heat conduction characteristics of vascular tissue are extracted, including:

[0021] Calculate the first-order temperature derivatives of the two-dimensional temperature matrix in the horizontal and vertical directions to obtain the temperature gradient vector field;

[0022] The temperature gradient vector field is vector synthesized to calculate the heat flow direction at each pixel.

[0023] The two-dimensional temperature matrix is ​​convolved using the Laplacian operator to calculate the thermal diffusivity at each pixel, which reflects the ease with which heat diffuses at the pixel.

[0024] Select a continuous pixel region where the heat flow direction is consistent and the heat diffusivity is lower than a preset threshold, and mark the continuous pixel region as a suspected blood vessel region;

[0025] The average temperature gradient direction and average thermal diffusion rate of all pixels within the suspected vascular region are extracted as the thermal conduction characteristics of the vascular tissue.

[0026] Furthermore, the thermodynamic causal reasoning model, based on the physical laws governing heat transfer from the core region of the body to the body surface, inversely derives the three-dimensional spatial coordinates of the subcutaneous veins, including:

[0027] In the aforementioned thermodynamic causal reasoning model, a constant heat source boundary condition representing the core temperature of the human body is set.

[0028] Based on the Fourier heat conduction equation, a nonlinear mapping relationship between surface temperature and subcutaneous tissue depth in the two-dimensional temperature matrix is ​​established.

[0029] Using numerical methods, the nonlinear mapping relationship is iteratively solved to calculate the theoretical temperature values ​​at different depths below the suspected blood vessel region.

[0030] The calculated theoretical temperature value is compared with the actual measured temperature value obtained from the two-dimensional temperature matrix, and the residual between the theoretical temperature value and the actual measured temperature value is calculated.

[0031] By adjusting the depth and radius parameters of the subcutaneous vein, the residual is minimized. The corresponding depth and radius parameters are then the estimated depth and diameter of the subcutaneous vein, thereby determining the three-dimensional spatial coordinates.

[0032] The thermodynamic causal reasoning model is constructed in the following way:

[0033] Obtain a training dataset, which includes multiple sets of labeled infrared thermal imaging sample data and their paired three-dimensional spatial location data of real subcutaneous veins measured by medical imaging equipment;

[0034] The model architecture is defined as follows: the thermodynamic causal reasoning model adopts a nonlinear regression model based on a multilayer neural network. Its input layer is the heat conduction characteristics of the vascular tissue, including the temperature gradient direction and heat diffusion rate. The output layer is the predicted vein depth parameters and radius parameters. At least one hidden layer is set between the input layer and the output layer, and a nonlinear activation function is applied in the hidden layer.

[0035] A physical constraint loss function is established, which consists of two parts: the first part is the mean square error loss between the predicted output and the true labeled values ​​in the training dataset, and the second part is the physical consistency loss based on Fourier's heat conduction law. The physical consistency loss penalizes the deviation between the theoretical body surface temperature distribution caused by the predicted output of the model and the actual temperature distribution observed by the infrared image.

[0036] The thermodynamic causal inference model is iteratively trained using the training dataset. In each iteration, sample data is input into the model to obtain the predicted output, and the total loss is calculated according to the physical constraint loss function. Then, the weight parameters of the thermodynamic causal inference model are updated through the backpropagation algorithm until the total loss converges to below a preset threshold.

[0037] After the model training is completed, the performance of the thermodynamic causal inference model is evaluated on an independent validation dataset to verify whether the average error between the vein depth and radius it inverts and the true value is lower than the preset accuracy threshold.

[0038] The evaluated thermodynamic causal reasoning model is solidified and deployed as the pre-set thermodynamic causal reasoning model.

[0039] Furthermore, based on the three-dimensional spatial coordinates and the temperature values ​​in the two-dimensional temperature matrix, the estimated diameter and depth of the subcutaneous vein are calculated, including:

[0040] Using the three-dimensional spatial coordinates as the center, extract a cross-sectional temperature profile curve that passes through the center of the three-dimensional spatial coordinates from the two-dimensional temperature matrix;

[0041] Analyze the half-width at half-maximum (WHM) of the temperature profile curve of the cross section, where WHM refers to the width when the temperature drops to half of the peak value;

[0042] The physical diameter of the subcutaneous vein is calculated based on the half-width at half-maximum and the pre-calibrated attenuation coefficient of infrared light in the tissue, and is used as the estimated diameter value.

[0043] The temperature difference between the surface temperature value at the three-dimensional spatial coordinates and the average temperature value of the surrounding adipose tissue region is measured.

[0044] Based on the temperature difference and the typical temperature difference range between arterial and venous blood in the human body, combined with the heat conduction model, the path length of heat transfer is deduced and used as the estimated value of the depth.

[0045] Further, based on the estimated diameter and the estimated depth, the three-dimensional spatial coordinates of the subcutaneous vein are geometrically corrected to obtain the corrected vein center coordinates, including:

[0046] Determine whether the estimated diameter value is less than a preset minimum puncture diameter threshold;

[0047] If the value is less than 1, the currently identified blood vessel is determined to be a capillary or a small vein, and it is removed from the candidate list.

[0048] If it is not less than, then adjust the vertical component of the three-dimensional spatial coordinates according to the depth estimate to correct the projection error caused by the unevenness of the skin surface;

[0049] Using the estimated diameter as a constraint, an elliptic curve is refitted in a local neighborhood, the elliptic curve representing the cross-sectional projection of the subcutaneous vein.

[0050] The coordinates of the center of the elliptic curve are updated to the corrected coordinates of the vein center.

[0051] Furthermore, using the corrected vein center coordinates, a mapping transformation is performed between the infrared image coordinate system and the physical space coordinate system to generate vein location data in the world coordinate system, including:

[0052] Read the set of three-dimensional coordinate points of the calibration plate installed below the infrared camera in the world coordinate system;

[0053] Extract the set of two-dimensional pixel coordinates corresponding to the calibration plate in the infrared image;

[0054] The transformation matrix from the infrared image coordinate system to the world coordinate system is calculated using a perspective transformation algorithm.

[0055] Substitute the corrected vein center coordinates into the transformation matrix to calculate the three-dimensional coordinates of the corrected vein center coordinates in the world coordinate system;

[0056] The three-dimensional coordinate values ​​and their corresponding attitude angles are encapsulated as vein position data in the world coordinate system.

[0057] Furthermore, it also includes:

[0058] A time-series smoothing filter is applied to the vein location data in the world coordinate system to eliminate coordinate jitter caused by limb micro-movements.

[0059] The filtered vein location data in the world coordinate system is registered and superimposed with the pre-acquired visible light image data of the target area;

[0060] Based on the registered and superimposed data, output the positioning result containing the spatial location, direction angle and subcutaneous depth information of the subcutaneous vein;

[0061] The step of performing temporal smoothing filtering on the vein position data in the world coordinate system to eliminate coordinate jitter caused by limb micro-movements includes:

[0062] Create a fixed-length data buffer to store vein location data in the world coordinate system over the most recent sampling periods;

[0063] Calculate the Euclidean distance between the vein position data in the world coordinate system at the current moment and the vein position data in the world coordinate system at the previous moment;

[0064] If the Euclidean distance is greater than the preset maximum displacement threshold, it is determined that a large limb movement has occurred, the data buffer is cleared and repositioning is performed;

[0065] If the Euclidean distance is not greater than the preset maximum displacement threshold, then all historical coordinate points are extracted from the data buffer.

[0066] Using the Kalman filter algorithm, combined with the historical coordinate points and the current measurement value, an optimal coordinate estimate is predicted, and the current vein position data in the world coordinate system is replaced with the optimal coordinate estimate.

[0067] Further, the filtered vein location data in the world coordinate system is registered and superimposed with the pre-acquired visible light image data of the target area, including:

[0068] Feature point detection is performed on the visible light image data to extract blood vessel texture feature points;

[0069] The same feature point detection is performed on the infrared image data to extract temperature edge feature points;

[0070] A feature matching algorithm is used to find matching feature point pairs between the visible light image data and the infrared image data.

[0071] Based on the feature point pairs, the affine transformation parameters between the visible light image data and the infrared image data are calculated;

[0072] Using the affine transformation parameters, the filtered vein location data in the world coordinate system is mapped to the coordinate system of the visible light image, thereby achieving visual overlay of the vein location data and the visible light image data.

[0073] Furthermore, the step of outputting a positioning result containing the spatial location, orientation angle, and subcutaneous depth information of the subcutaneous vein based on the registered and superimposed data includes:

[0074] From the registered and superimposed data, the specific values ​​of the corrected vein center coordinates in the world coordinate system are read as the spatial location of the subcutaneous vein.

[0075] Along the course of the main vein where the corrected vein center coordinates are located, select several consecutive sampling points;

[0076] Calculate the angle between the line connecting two adjacent sampling points and the horizontal axis, and take the average of all angles to obtain the direction angle;

[0077] The depth estimate used to generate the corrected vein center coordinates is read as the subcutaneous depth information;

[0078] The spatial location, orientation angle, and subcutaneous depth information are packaged into a structured data packet, and the data packet is sent to a designated display terminal for display.

[0079] Compared with the prior art, the beneficial effects of the present invention are:

[0080] The temperature gradient direction and heat diffusion rate of vascular tissue are extracted and input into a pre-set thermodynamic causal reasoning model. This model follows the physical laws of heat transfer from the core region of the body to the surface, retrieving the three-dimensional spatial coordinates of subcutaneous veins. Compared to conventional infrared imaging techniques that can only acquire two-dimensional projection information of veins, this method can accurately capture the spatial distribution of veins under the skin. It overcomes the limitation of conventional techniques that cannot obtain the three-dimensional position of veins from infrared data, accurately reflecting the spatial direction and positional relationship of veins, reducing the deviation in vein location identification, and making vein positioning more consistent with the actual spatial state of subcutaneous veins.

[0081] Based on the inverted three-dimensional spatial coordinates of the subcutaneous vein and the temperature values ​​in the two-dimensional temperature matrix, the estimated diameter and depth of the subcutaneous vein are calculated. Then, based on these estimated diameter and depth values, the three-dimensional spatial coordinates of the subcutaneous vein are geometrically corrected to obtain the corrected vein center coordinates. Conventional techniques do not perform geometric correction on the identified vein location, which can easily lead to inaccurate vein localization due to deviations in diameter and depth estimation. This technique, however, can compensate for errors in the diameter and depth estimation process through geometric correction, optimizing the accuracy of the vein center coordinates. This makes the vein location more closely match the actual location of the subcutaneous vein, further improving the accuracy of vein location data and providing reliable positional support for precise puncture in automated blood collection. Attached Figure Description

[0082] Figure 1 This is a flowchart illustrating the steps of an automated subcutaneous venous blood collection method based on infrared imaging as described in this invention.

[0083] Figure 2A flowchart for extracting the thermal conductivity characteristics of vascular tissue;

[0084] Figure 3 A comparison chart of theoretical and actual measured temperatures at different subcutaneous depths;

[0085] Figure 4 Heatmap of estimated subcutaneous vein depth;

[0086] Figure 5 This is a time-series curve of the original vein coordinates. Detailed Implementation

[0087] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0088] See Figure 1 The method acquires raw infrared thermal imaging data of a target area, which is the surface skin area of ​​the limb to be blood collected. The raw infrared thermal imaging data is then processed and analyzed to output vein location data in the world coordinate system, which can be used to guide automated blood collection equipment. The method first reconstructs the temperature field of the raw infrared thermal imaging data to generate a two-dimensional temperature matrix reflecting the heat distribution of subcutaneous tissue. Based on this matrix, the heat conduction characteristics of the vascular tissue are extracted, including the temperature gradient direction and heat diffusion rate. Then, the temperature gradient direction and heat diffusion rate are input into a pre-set thermodynamic causal inference model. This model, based on the physical laws of heat transfer from the core region of the body to the body surface, inversely derives the three-dimensional spatial coordinates of the subcutaneous vein. Next, based on the three-dimensional spatial coordinates and the temperature values ​​in the two-dimensional temperature matrix, the estimated diameter and depth of the subcutaneous vein are calculated. Then, based on the estimated diameter and depth, the three-dimensional spatial coordinates of the subcutaneous vein are geometrically corrected to obtain the corrected vein center coordinates. Finally, using the corrected vein center coordinates, a mapping transformation is performed between the infrared image coordinate system and the physical space coordinate system to generate vein location data in the world coordinate system.

[0089] In one embodiment of the present invention, the first step in reconstructing the temperature field of the raw infrared thermal imaging data is to perform non-uniformity correction. When an infrared thermal imaging device detects infrared radiation, the response characteristics of each pixel in the detector have inherent inconsistencies. This inconsistency manifests as fixed pattern noise in the acquired raw infrared thermal imaging data. This noise appears as regular or irregular spatial fixed brightness variations in the image, which seriously interferes with the subsequent accurate analysis of temperature distribution. The non-uniformity correction process is based on the pre-calibrated gain coefficient and bias coefficient of each pixel, and performs a linear transformation on the output value of each pixel, thereby numerically eliminating the fixed pattern noise introduced by the detector's own characteristics, and obtaining the pre-corrected raw infrared thermal imaging data.

[0090] An adaptive Gaussian filtering algorithm is used to smooth and denoise the corrected infrared thermal imaging raw data. The core of the adaptive Gaussian filtering algorithm is that the size and standard deviation of its filtering kernel function can be dynamically adjusted according to the noise level and feature structure of the local area of ​​the image. In specific implementation, the algorithm first calculates the gray-level variance of the neighborhood of each pixel in the image. For flat areas with small variance, a larger-scale filtering kernel is used to enhance the noise suppression effect. For edge areas with large variance, such as temperature change feature areas where blood vessel edges are located, a smaller-scale filtering kernel is switched to avoid blurring details. This process can effectively filter out random noise while preserving the temperature change features of blood vessel edges to the maximum extent, thus preserving the necessary contrast information for subsequent feature extraction.

[0091] According to Planck's law of blackbody radiation, the smoothed and denoised grayscale image data is converted into absolute temperature values. In practice, the raw infrared thermal imaging data is essentially the grayscale value received by the detector, which is proportional to the radiation energy of the object's surface. Planck's law of blackbody radiation describes the precise relationship between blackbody radiation intensity, wavelength, and temperature. The conversion formula involves the detector's spectral response function, environmental radiation compensation, and correction parameters for the object's emissivity. The conversion formula is expressed as follows:

[0092]

[0093] in: This represents the absolute temperature value to be determined. This indicates the center operating wavelength of the infrared detector. and These are the first radiation constant and the second radiation constant, respectively. Indicates the skin at wavelength Emission rate at that location It is the maximum grayscale value output by the detector at that wavelength. It is the equivalent optical path length compensation factor for the target distance detector. The calibration coefficient is determined by both the detector gain and bias. By applying this transformation relationship to the grayscale value of each pixel in the image, an initial temperature distribution map reflecting the absolute temperature of each point on the surface of the tested skin is generated.

[0094] In some embodiments, detecting abnormally high or low temperature pixels in the initial temperature distribution map is a necessary follow-up step. These abnormal points may originate from transient sensor noise, environmental reflections, or foreign objects on the skin surface. Detection methods are typically based on the neighborhood statistical characteristics of pixels. For example, calculating the median temperature and standard deviation within a window centered on each pixel. If the difference between the pixel's temperature value and the median exceeds a preset multiple of the standard deviation, it is determined to be an abnormal pixel. It can be understood that the temperature value of a marked abnormal point will be replaced with the median temperature of all non-abnormal points within the window centered on that point. This process completes the temperature field repair, ensuring the spatial consistency of the temperature data.

[0095] In some embodiments, after the temperature field repair is completed, the initial temperature distribution map needs to be resampled. The purpose of resampling is to unify the initial temperature distribution map to a preset spatial resolution. The preset resolution is usually matched with the scale of subsequent processing, for example, it is set to correspond to a specific number of pixels per millimeter. The resampling process can use bilinear interpolation or cubic spline interpolation algorithms. Its goal is to generate a two-dimensional temperature matrix with standardized spatial coordinates and accurate temperature values. Each element of this two-dimensional temperature matrix stores the absolute temperature value of the corresponding physical location obtained after the above series of processing, which reflects the heat distribution of subcutaneous tissue.

[0096] In one embodiment of the present invention, see [reference] Figure 2The initial step in extracting the heat conduction characteristics of vascular tissue based on a two-dimensional temperature matrix is ​​to calculate the first-order temperature derivatives of the two-dimensional temperature matrix in the horizontal and vertical directions. The two-dimensional temperature matrix is ​​an array of temperature values ​​stored as a pixel grid. Calculating the first-order derivative in the horizontal direction involves using the central difference method to calculate the temperature difference between each pixel and its left and right adjacent pixels for each row of the matrix, thus obtaining the horizontal temperature change rate. Similarly, calculating the first-order derivative in the vertical direction involves using the same method to calculate the temperature difference between each pixel and its upper and lower adjacent pixels for each column of the matrix, obtaining the vertical temperature change rate. These two rate components constitute the temperature gradient vector of each pixel, and the collection of temperature gradient vectors from all pixels constitutes the temperature gradient vector field. After obtaining the temperature gradient vector field, it is necessary to synthesize the temperature gradient vectors at each pixel. Since the horizontal and vertical components of the vectors are known, the heat flow direction at each pixel can be obtained by calculating the vector's direction angle. The heat flow direction indicates the direction of the fastest temperature decrease in the two-dimensional plane, which is usually related to the direction of heat conduction from subcutaneous blood vessels to the body surface. The synthesized heat flow direction of each pixel constitutes a direction field with the same dimension as the two-dimensional temperature matrix.

[0097] Another parallel process for extracting thermal conductivity features is to perform convolution operations on the two-dimensional temperature matrix using the Laplacian operator. The Laplacian operator is a second-order differential operator. On a discrete pixel grid, a specific convolution kernel can be used to operate on the two-dimensional temperature matrix. This kernel is typically a 3x3 matrix, with a positive value at the center and negative values ​​at the four adjacent positions (top, bottom, left, and right). After the Laplacian convolution operation, each value in the output matrix represents the sum of the second derivatives of the temperature at that pixel. This value is called the thermal diffusivity, which reflects the ease with which heat diffuses at a pixel. The absolute value of the thermal diffusivity is larger at the edges of heat sources or where temperature changes are drastic. In practice, calculating the thermal diffusivity of each pixel is a quantitative indicator of its thermal conductivity. The thermal diffusivity characteristics of vascular tissue differ from those of surrounding tissues such as fat and muscle, providing a basis for distinguishing different tissues.

[0098] After obtaining the heat flow direction and thermal diffusivity of each pixel, the algorithm begins region labeling. The labeling goal is to select continuous pixel regions with consistent heat flow directions and thermal diffusivity below a preset threshold. The preset threshold is an empirical value obtained from statistical analysis of the thermal properties of a large number of normal human tissues. The algorithm scans the entire two-dimensional temperature matrix and performs region growing on pixels that meet the conditions. Pixels that are adjacent to each other, have an angle between their heat flow directions less than a set tolerance, and all have thermal diffusivity below the threshold are merged into a single connected region. This connected region is labeled as a suspected vascular region. Blood vessels, as flowing heat sources, typically have low thermal diffusivity, and their surface heat flow directions tend to be consistent. After labeling, the average temperature gradient direction and average thermal diffusivity of all pixels within the suspected vascular region are extracted. The average temperature gradient direction is obtained by averaging the heat flow direction vectors of all pixels within the region, and the average thermal diffusivity is obtained by averaging the absolute values ​​of the thermal diffusivity of all pixels within the region. These two average values ​​together constitute the thermal conduction characteristics used to characterize vascular tissue. These characteristics physically describe the heat conduction mode and rate of the vascular region. In some embodiments, the calculation of the temperature gradient direction can be performed using a weighted average, where the weight can be the temperature gradient magnitude of each pixel. This can enhance the contribution of the main heat flow direction in the average calculation. The resulting heat conduction characteristics of the vascular tissue are a set of scalars that can be input into the subsequent thermodynamic causal inference model for inverting the three-dimensional structure of the vein.

[0099] In one embodiment of the present invention, the thermodynamic causal reasoning model infers the three-dimensional spatial coordinates of subcutaneous veins based on the physical laws of heat transfer from the core region of the body to the body surface. In the thermodynamic causal reasoning model, a constant heat source boundary condition representing the core temperature of the human body is set, with the core temperature set at 37.0 degrees Celsius, which serves as the constant temperature boundary of the model in deep tissues. Based on the Fourier heat conduction equation, a nonlinear mapping relationship between surface temperature and subcutaneous tissue depth is established in the two-dimensional temperature matrix. The Fourier heat conduction equation describes the basic laws of heat transfer in a medium. In a non-homogeneous medium composed of multiple layers of biological tissues such as skin, subcutaneous fat, and muscle, the heat conduction process is complex. Numerical methods are used to iteratively solve the nonlinear mapping relationship. The finite difference method can be used to discretize the subcutaneous tissue into multiple thin layers in the depth direction and calculate the theoretical temperature value at each discrete depth below the suspected vascular area. The calculated theoretical temperature value is compared with the actual measured temperature value obtained from the two-dimensional temperature matrix, and the residual between the theoretical temperature value and the actual measured temperature value is calculated. By adjusting the depth and radius parameters of the subcutaneous vein, the residual is minimized. The corresponding depth and radius parameters are the estimated depth and diameter values ​​of the subcutaneous vein, and thus the three-dimensional spatial coordinates are determined. The depth parameter defines the vertical distance from the center of the vein to the skin surface, and the radius parameter defines the cross-sectional radius of the vein.

[0100] In some embodiments, the construction process of the thermodynamic causal inference model is systematic. First, a training dataset is acquired, comprising multiple sets of labeled infrared thermal imaging sample data and paired data of the actual three-dimensional spatial location of subcutaneous veins measured by medical imaging equipment, such as ultrasound imaging equipment capable of accurately measuring the depth and diameter of veins. Next, the model architecture is defined. The thermodynamic causal inference model employs a nonlinear regression model based on a multi-layer neural network. Its input layer consists of the thermal conduction characteristics of vascular tissue, including the temperature gradient direction and thermal diffusion rate. The output layer contains the predicted vein depth and radius parameters. At least one hidden layer is set between the input and output layers, and a nonlinear activation function, such as the ReLU function, is applied in the hidden layer. Finally, a physical constraint loss function is established, consisting of two parts: the first part is the mean squared error loss between the predicted output and the true labeled values ​​in the training dataset; the second part is the physical consistency loss based on Fourier's law of heat conduction. The physical consistency loss penalizes the deviation between the theoretical body surface temperature distribution caused by the model's predicted output and the actual temperature distribution observed in the infrared image. The thermodynamic causal inference model is iteratively trained using a training dataset. In each iteration, sample data is input into the model to obtain the predicted output, and the total loss is calculated based on the physical constraint loss function. Then, the weight parameters of the thermodynamic causal inference model are updated using the backpropagation algorithm until the total loss converges to below a preset threshold. After the model training is complete, the performance of the thermodynamic causal inference model is evaluated on an independent validation dataset to verify whether the average error between the inferred vein depth and radius and the true value is lower than a preset accuracy threshold. The preset accuracy threshold can be set to a depth error of less than 0.5 mm and a radius error of less than 0.2 mm. The evaluated thermodynamic causal inference model is then solidified and deployed as a pre-built thermodynamic causal inference model. Solidification means saving the trained model weights and structure as a file for direct loading and use during the inference phase.

[0101] After calculating the three-dimensional spatial coordinates of the subcutaneous vein, the estimated diameter and depth of the subcutaneous vein are calculated based on the three-dimensional spatial coordinates and the temperature values ​​in the two-dimensional temperature matrix. A cross-sectional temperature profile curve is extracted from the two-dimensional temperature matrix, centered on the three-dimensional spatial coordinates and passing through this center. This temperature profile curve is a straight line perpendicular to the estimated vein direction, capturing one-dimensional temperature distribution data from the two-dimensional temperature matrix. The half-width at half-maximum (HWHM) of the cross-sectional temperature profile curve is analyzed. HWHM refers to the width at which the temperature drops to half its peak value. The peak temperature is located above the estimated vein center projection point and is typically higher than the adjacent tissues. Based on the HWHM and the pre-calibrated attenuation coefficient of infrared light in the tissue, the physical diameter of the subcutaneous vein is calculated as the estimated diameter. The calculation relationship can be expressed as:

[0102]

[0103] in: This represents the calculated physical diameter of the subcutaneous vein, i.e., the estimated diameter. This represents the half-width at half-maximum (WHM) measured from the cross-sectional temperature profile. This represents the attenuation coefficient of pre-calibrated infrared light in human tissue, obtained through in vitro tissue model experiments. The temperature difference between the surface temperature at a three-dimensional coordinate and the average temperature of the surrounding adipose tissue region is measured. The surrounding adipose tissue region is typically selected as a homogeneous adipose tissue area within a certain range around the vein projection point. Based on the temperature difference and the typical temperature difference range between arterial and venous blood in the human body, combined with a heat conduction model, the path length of heat transfer is deduced as a depth estimate. The typical temperature difference range between arterial and venous blood in the human body is a known physiological parameter. Combined with the thermal conductivity of biological tissue, the depth can be solved using a one-dimensional steady-state heat conduction formula. The heat conduction model assumes that heat is conducted vertically from venous blood (considered a linear heat source) to the skin surface. Given the skin surface temperature, venous blood temperature, and tissue thermal conductivity, a mathematical relationship between depth and temperature difference can be derived. Solving this relationship yields the depth estimate, which can be mutually verified or integrated with the depth parameters previously obtained through a thermodynamic causal reasoning model.

[0104] See Figure 3 This is a comparison chart of theoretical and actual measured temperatures at different subcutaneous depths. The temperature increases with subcutaneous depth, consistent with the physical law of heat transfer from the core region to the body surface. The temperature increases logarithmically with depth, approaching the core temperature of 37°C, reflecting the attenuation characteristics of thermal conduction in biological tissues. The overall trend of the actual measured temperature and the theoretically calculated temperature are highly consistent, indicating that the raw infrared thermal imaging data, after correction, denoising, and temperature conversion, accurately reflects the thermal distribution characteristics of subcutaneous tissue. The residuals of the two curves in the chart represent the key physical constraint loss for model training. By minimizing these residuals, the depth and diameter parameters of subcutaneous veins can be derived. By analyzing the temperature deviations at different depths, the temperature field reconstruction algorithm can be optimized to improve the accuracy of vein localization.

[0105] In one embodiment of the present invention, the three-dimensional spatial coordinates of the subcutaneous vein are geometrically corrected based on the estimated diameter and depth values ​​to obtain the corrected vein center coordinates. It is then determined whether the estimated diameter is less than a preset minimum puncture diameter threshold, which is set according to the specifications of the blood collection needle, for example, 1.0 mm. If the estimated diameter is less than the preset minimum puncture diameter threshold, the currently identified blood vessel is determined to be a capillary or a small vein, and it is removed from the candidate list. It can be understood that if it is not less than the preset minimum puncture diameter threshold, the vertical component of the three-dimensional spatial coordinates is adjusted based on the estimated depth value to correct the projection error caused by uneven skin surface. Uneven skin surface means that there is a geometric deviation between the actual position of the vein in three-dimensional space and its two-dimensional projection position on the infrared image. This deviation is most significant in the vertical direction (depth direction). A correction factor is introduced to adjust the vertical component. The correction factor is a function of the estimated depth value and the original depth parameter, and its expression is:

[0106]

[0107] in: This represents the corrected vertical component. Represents the vertical component of the uncorrected, original three-dimensional spatial coordinates. This represents the estimated depth. This represents the depth parameters initially obtained from the thermodynamic causal reasoning model. This is an empirical coefficient used to control the strength of the correction. Using the estimated diameter as a constraint, an elliptic curve is refitted within a local neighborhood. The elliptic curve represents the cross-sectional projection of the subcutaneous vein. The local neighborhood is a rectangular image region centered on the projection point corresponding to the adjusted vertical component. The ellipse fitting algorithm finds an ellipse whose boundary passes through the point with the most significant temperature gradient change within the region, and whose major axis radius is correlated with the estimated diameter. The coordinates of the elliptic curve's center are updated to the corrected vein center coordinates. The corrected vein center coordinates contain the precise position on the two-dimensional image plane and corrected depth information, representing the three-dimensional position of the vein center in the local tissue coordinate system.

[0108] In some embodiments, the corrected vein center coordinates are used to perform a mapping transformation between the infrared image coordinate system and the physical space coordinate system to generate vein position data in the world coordinate system. The three-dimensional coordinate point set of a calibration board mounted below the infrared camera in the world coordinate system is read. The calibration board is a regular pattern board with known three-dimensional coordinates, such as a checkerboard calibration board, whose three-dimensional physical coordinates of each corner point are pre-measured and stored. The two-dimensional pixel coordinate point set corresponding to the calibration board in the infrared image is extracted, and the corner point positions of the calibration board pattern in the infrared image are identified using image processing algorithms. The transformation matrix from the infrared image coordinate system to the world coordinate system is calculated using a perspective transformation algorithm. The perspective transformation describes the projection relationship between image pixel coordinates and three-dimensional world coordinates. The transformation matrix is ​​a 3x3 matrix, obtained by solving the least-squares solution of the following system of equations (see Table 1).

[0109] Table 1: Example Corresponding Points of Coordinate System Mapping

[0110] World coordinate system point (X,Y,Z) Infrared image pixel coordinates (u,v) (0.0,0.0,0.0) (320.5,240.5) (10.0,0.0,0.0) (450.2,241.1) (0.0,10.0,0.0) (321.0,150.8) (10.0,10.0,0.0) (451.5,151.3) (5.0,5.0,0.0) (386.0,196.0)

[0111] In Table 1, each row of data represents a corner point with known 3D physical coordinates on the calibration board and its corresponding 2D pixel coordinates detected in the infrared image. It can be understood that by using multiple pairs of such corresponding points, the elements of the transformation matrix can be solved. Substituting the corrected vein center coordinates into the transformation matrix, the 3D coordinates of the corrected vein center in the world coordinate system are calculated. The corrected vein center coordinates are a 3D point composed of image pixel coordinates (u,v) and a depth value z_c. During mapping, the depth value z_c is used to calculate the scaling factor. The 3D coordinate values ​​and their corresponding attitude angles are encapsulated as vein position data in the world coordinate system. The attitude angles are calculated based on the vein's direction within its local neighborhood, representing the vein's orientation in 3D space.

[0112] See Figure 4 This is a heatmap of estimated subcutaneous vein depth, used to present the spatial distribution of vein depth in subcutaneous tissue. The colors are randomly mottled, without a clear continuous vein orientation, reflecting the non-uniformity of subcutaneous tissue and measurement noise. Depth values ​​are concentrated in the 2.0–2.5 mm range, consistent with the typical depth range of superficial veins in the human body. Yellow areas represent shallower depths, and purple areas represent deeper depths. It visually displays the two-dimensional spatial distribution of subcutaneous vein depth, providing foundational data for subsequent geometric correction and coordinate mapping. It can be used to identify areas of abnormal depth, assisting in selecting suitable veins for puncture. It provides depth dimension input for world coordinate system mapping, generating complete three-dimensional vein location data, which can be registered and overlaid with visible light images, intuitively guiding puncture localization in clinical scenarios.

[0113] In one embodiment of the present invention, temporal smoothing filtering is performed on the vein position data in the world coordinate system to eliminate coordinate jitter caused by limb micro-movements. A fixed-length data buffer is created to store the vein position data in the world coordinate system within the most recent sampling periods. The length of the data buffer can be set according to the sampling frequency, for example, it can be set to store the data of the most recent 10 sampling times. The Euclidean distance between the vein position data in the world coordinate system at the current time and the vein position data in the world coordinate system at the previous time is calculated. The Euclidean distance is the straight-line distance between two points in three-dimensional space, and its calculation formula is:

[0114]

[0115] in: This represents the Euclidean distance between the current and previous timestamps of the vein. This represents the coordinates of the vein's location at time t. This represents the vein position coordinates at the previous time t-1. If the Euclidean distance is greater than the preset maximum displacement threshold, it is determined that a significant limb movement has occurred. The preset maximum displacement threshold can be set to 5 mm. The data buffer is cleared and repositioning is performed. Repositioning means triggering the entire process from acquiring the raw infrared thermal imaging data to be re-executed. If the Euclidean distance is not greater than the preset maximum displacement threshold, all historical coordinate points are extracted from the data buffer. Historical coordinate points are all position coordinate sequences traced back from the current time to the earliest time in the buffer. A Kalman filter algorithm is used to combine historical coordinate points and the measurement value at the current time to predict an optimal coordinate estimate. This optimal coordinate estimate is then used to replace the current vein position data in the world coordinate system. The Kalman filter is a recursive algorithm that combines the system's motion model prediction and the current measurement update to estimate the optimal value of the system state. The process includes two steps: prediction and update. In the prediction step, the state at the current time is predicted based on the state estimate at the previous time and the system's state transition matrix. In the update step, the predicted value is corrected using the measurement value at the current time, resulting in a smoother and more accurate coordinate estimate.

[0116] It is understandable that after completing temporal smoothing filtering, the filtered vein location data in the world coordinate system is registered and superimposed with the pre-acquired visible light image data of the target area. Feature point detection is then performed on the visible light image data to extract vascular texture feature points, which are points with significant local features such as vessel edges and bifurcation points in the visible light image. The same feature point detection is performed on the infrared image data to extract temperature edge feature points, which are edge points formed due to temperature abrupt changes in the infrared image. A feature matching algorithm is used to find matching feature point pairs in the visible light and infrared image data. This algorithm finds corresponding points in the two images by comparing the similarity between feature descriptors. Based on the feature point pairs, the affine transformation parameters between the visible light and infrared image data are calculated. The affine transformation includes translation, rotation, scaling, and shearing. At least three pairs of non-collinear matching points are needed to solve for the transformation matrix. The calculation process involves solving a system of linear equations to obtain the six parameters of the transformation matrix. By using affine transformation parameters, the filtered vein location data in the world coordinate system is mapped to the coordinate system of the visible light image, realizing the visual overlay of vein location data and visible light image data. The mapping process is to transform the two-dimensional projection coordinates of the filtered vein location data through the calculated affine transformation matrix, thereby drawing visual markers such as the location and direction of the veins on the visible light image.

[0117] In some embodiments, the positioning result, including the spatial location, orientation angle, and subcutaneous depth information of the subcutaneous vein, is output based on the registered and superimposed data. The corrected vein center coordinates in the world coordinate system are read from the registered and superimposed data as the spatial location of the subcutaneous vein. The spatial location is expressed in three-dimensional coordinates, for example (X: 15.3 mm, Y: 22.7 mm, Z: 4.1 mm). Several consecutive sampling points are selected along the vein trunk where the corrected vein center coordinates are located. The sampling points can be three points evenly distributed on the vein trunk. The angle between the line connecting two adjacent sampling points and the horizontal axis is calculated, and the average of all angles is taken to obtain the orientation angle. The horizontal axis is usually defined as the X-axis of the world coordinate system. The angle can be calculated using the vector dot product formula, and then the arithmetic mean of the calculated angles is taken. The depth estimate used to generate the corrected vein center coordinates is read as the subcutaneous depth information. The subcutaneous depth information is a direct reference to the depth estimate, representing the vertical distance from the vein center to the skin surface. The spatial location, orientation angle, and subcutaneous depth information are packaged into a structured data packet. The structured data packet can be defined as a data structure containing three fields. The data packet is then sent to a designated display terminal for display. The display terminal can be a tablet computer or a computer monitor, used to graphically overlay and display the vein location information.

[0118] See Figure 5 This is a time-series curve of the original vein coordinates, used to illustrate the jitter characteristics of the three-dimensional vein coordinates over time without filtering. The X / Y coordinates decrease slowly overall, accompanied by high-frequency small-amplitude jitter, while the Z coordinate remains stable in the 50–53 mm range with even smaller jitter. The jitter originates from limb micro-movements, detector noise, and algorithm measurement errors. This visually demonstrates the time-series jitter problem in the original coordinate data, proving the necessity of the "time-series smoothing filtering" step. Directly using it for puncture guidance would lead to decreased positioning accuracy and insufficient operational stability. The X / Y coordinate jitter is approximately ±2–3 mm, and the Z coordinate jitter is approximately ±1 mm, providing a direct basis for the parameter design of subsequent filtering algorithms. The amplitude of the jitter can be used to determine whether repositioning or activation of limb fixation aids is necessary.

[0119] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. An automated subcutaneous venous blood collection method based on infrared imaging, characterized in that, include: Acquire raw infrared thermal imaging data of the target area, which is the surface skin area of ​​the limb to be blood collected; The raw infrared thermal imaging data is processed to reconstruct the temperature field, generating a two-dimensional temperature matrix that reflects the thermal distribution of subcutaneous tissue. Based on the two-dimensional temperature matrix, the thermal conduction characteristics of vascular tissue are extracted, including the temperature gradient direction and the thermal diffusion rate. The temperature gradient direction and the thermal diffusion rate are input into a preset thermodynamic causal reasoning model, which, based on the physical laws of heat transfer from the core region of the body to the body surface, inversely derives the three-dimensional spatial coordinates of the subcutaneous veins. Based on the three-dimensional spatial coordinates and the temperature values ​​in the two-dimensional temperature matrix, the estimated diameter and depth of the subcutaneous vein are calculated. Based on the estimated diameter and the estimated depth, the three-dimensional spatial coordinates of the subcutaneous vein are geometrically corrected to obtain the corrected coordinates of the vein center. Using the corrected vein center coordinates, a mapping transformation is performed between the infrared image coordinate system and the physical space coordinate system to generate vein location data in the world coordinate system.

2. The automated subcutaneous venous blood collection method based on infrared imaging as described in claim 1, characterized in that, The raw infrared thermal imaging data is processed to reconstruct the temperature field, generating a two-dimensional temperature matrix reflecting the thermal distribution of subcutaneous tissue, including: Non-uniformity correction is performed on the raw infrared thermal imaging data to eliminate fixed pattern noise introduced by the detector's own characteristics in the raw infrared thermal imaging data; An adaptive Gaussian filtering algorithm is used to smooth and denoise the corrected raw infrared thermal imaging data while preserving the temperature abrupt changes at the edge of blood vessels. According to Planck's blackbody radiation law, the smoothed and denoised grayscale image data is converted into absolute temperature values ​​to obtain the initial temperature distribution map. Detect abnormally high or low temperature pixels in the initial temperature distribution map and replace them with the median temperature of the surrounding neighboring pixels to complete the temperature field repair. The repaired initial temperature distribution map is resampled to a preset resolution to generate the two-dimensional temperature matrix.

3. The automated subcutaneous venous blood collection method based on infrared imaging as described in claim 2, characterized in that, Based on the two-dimensional temperature matrix, the heat conduction characteristics of vascular tissue are extracted, including: Calculate the first-order temperature derivatives of the two-dimensional temperature matrix in the horizontal and vertical directions to obtain the temperature gradient vector field; The temperature gradient vector field is vector synthesized to calculate the heat flow direction at each pixel. The two-dimensional temperature matrix is ​​convolved using the Laplacian operator to calculate the thermal diffusivity at each pixel, which reflects the ease with which heat diffuses at the pixel. Select a continuous pixel region where the heat flow direction is consistent and the heat diffusivity is lower than a preset threshold, and mark the continuous pixel region as a suspected blood vessel region; The average temperature gradient direction and average thermal diffusion rate of all pixels within the suspected vascular region are extracted as the thermal conduction characteristics of the vascular tissue.

4. The automated subcutaneous venous blood collection method based on infrared imaging as described in claim 3, characterized in that, The thermodynamic causal reasoning model, based on the physical laws governing heat transfer from the body's core region to the body surface, deduces the three-dimensional spatial coordinates of subcutaneous veins, including: In the aforementioned thermodynamic causal reasoning model, a constant heat source boundary condition representing the core temperature of the human body is set. Based on the Fourier heat conduction equation, a nonlinear mapping relationship between surface temperature and subcutaneous tissue depth in the two-dimensional temperature matrix is ​​established. Using numerical methods, the nonlinear mapping relationship is iteratively solved to calculate the theoretical temperature values ​​at different depths below the suspected blood vessel region. The calculated theoretical temperature value is compared with the actual measured temperature value obtained from the two-dimensional temperature matrix, and the residual between the theoretical temperature value and the actual measured temperature value is calculated. By adjusting the depth and radius parameters of the subcutaneous vein, the residual is minimized. The corresponding depth and radius parameters are then the estimated depth and diameter of the subcutaneous vein, thereby determining the three-dimensional spatial coordinates. The thermodynamic causal reasoning model is constructed in the following way: Obtain a training dataset, which includes multiple sets of labeled infrared thermal imaging sample data and their paired three-dimensional spatial location data of real subcutaneous veins measured by medical imaging equipment; The model architecture is defined as follows: the thermodynamic causal reasoning model adopts a nonlinear regression model based on a multilayer neural network. Its input layer is the heat conduction characteristics of the vascular tissue, including the temperature gradient direction and heat diffusion rate. The output layer is the predicted vein depth parameters and radius parameters. At least one hidden layer is set between the input layer and the output layer, and a nonlinear activation function is applied in the hidden layer. A physical constraint loss function is established, which consists of two parts: the first part is the mean square error loss between the predicted output and the true labeled values ​​in the training dataset, and the second part is the physical consistency loss based on Fourier's heat conduction law. The physical consistency loss penalizes the deviation between the theoretical body surface temperature distribution caused by the predicted output of the model and the actual temperature distribution observed by the infrared image. The thermodynamic causal inference model is iteratively trained using the training dataset. In each iteration, sample data is input into the model to obtain the predicted output, and the total loss is calculated according to the physical constraint loss function. Then, the weight parameters of the thermodynamic causal inference model are updated through the backpropagation algorithm until the total loss converges to below a preset threshold. After the model training is completed, the performance of the thermodynamic causal inference model is evaluated on an independent validation dataset to verify whether the average error between the vein depth and radius it inverts and the true value is lower than the preset accuracy threshold. The evaluated thermodynamic causal reasoning model is solidified and deployed as the pre-set thermodynamic causal reasoning model.

5. The automated subcutaneous venous blood collection method based on infrared imaging as described in claim 4, characterized in that, Based on the three-dimensional spatial coordinates and the temperature values ​​in the two-dimensional temperature matrix, the estimated diameter and depth of the subcutaneous vein are calculated, including: Using the three-dimensional spatial coordinates as the center, extract a cross-sectional temperature profile curve that passes through the center of the three-dimensional spatial coordinates from the two-dimensional temperature matrix; Analyze the half-width at half-maximum (WHM) of the temperature profile curve of the cross section, where WHM refers to the width when the temperature drops to half of the peak value; The physical diameter of the subcutaneous vein is calculated based on the half-width at half-maximum and the pre-calibrated attenuation coefficient of infrared light in the tissue, and is used as the estimated diameter value. The temperature difference between the surface temperature value at the three-dimensional spatial coordinates and the average temperature value of the surrounding adipose tissue region is measured. Based on the temperature difference and the typical temperature difference range between arterial and venous blood in the human body, combined with the heat conduction model, the path length of heat transfer is deduced and used as the estimated value of the depth.

6. The automated subcutaneous venous blood collection method based on infrared imaging as described in claim 5, characterized in that, Based on the estimated diameter and the estimated depth, the three-dimensional spatial coordinates of the subcutaneous vein are geometrically corrected to obtain the corrected coordinates of the vein center, including: Determine whether the estimated diameter value is less than a preset minimum puncture diameter threshold; If the value is less than 1, the currently identified blood vessel is determined to be a capillary or a small vein, and it is removed from the candidate list. If it is not less than, then adjust the vertical component of the three-dimensional spatial coordinates according to the depth estimate to correct the projection error caused by the unevenness of the skin surface; Using the estimated diameter as a constraint, an elliptic curve is refitted in a local neighborhood, the elliptic curve representing the cross-sectional projection of the subcutaneous vein. The coordinates of the center of the elliptic curve are updated to the corrected coordinates of the vein center.

7. The automated subcutaneous venous blood collection method based on infrared imaging as described in claim 6, characterized in that, Using the corrected vein center coordinates, a mapping transformation is performed between the infrared image coordinate system and the physical space coordinate system to generate vein location data in the world coordinate system, including: Read the set of three-dimensional coordinate points of the calibration plate installed below the infrared camera in the world coordinate system; Extract the set of two-dimensional pixel coordinates corresponding to the calibration plate in the infrared image; The transformation matrix from the infrared image coordinate system to the world coordinate system is calculated using a perspective transformation algorithm. Substitute the corrected vein center coordinates into the transformation matrix to calculate the three-dimensional coordinates of the corrected vein center coordinates in the world coordinate system; The three-dimensional coordinate values ​​and their corresponding attitude angles are encapsulated as vein position data in the world coordinate system.

8. The automated subcutaneous venous blood collection method based on infrared imaging as described in claim 7, characterized in that, Also includes: A time-series smoothing filter is applied to the vein location data in the world coordinate system to eliminate coordinate jitter caused by limb micro-movements. The filtered vein location data in the world coordinate system is registered and superimposed with the pre-acquired visible light image data of the target area; Based on the registered and superimposed data, output the positioning result containing the spatial location, direction angle and subcutaneous depth information of the subcutaneous vein; The step of performing temporal smoothing filtering on the vein position data in the world coordinate system to eliminate coordinate jitter caused by limb micro-movements includes: Create a fixed-length data buffer to store vein location data in the world coordinate system over the most recent sampling periods; Calculate the Euclidean distance between the vein position data in the world coordinate system at the current moment and the vein position data in the world coordinate system at the previous moment; If the Euclidean distance is greater than the preset maximum displacement threshold, it is determined that a large limb movement has occurred, the data buffer is cleared and repositioning is performed; If the Euclidean distance is not greater than the preset maximum displacement threshold, then all historical coordinate points are extracted from the data buffer. Using the Kalman filter algorithm, combined with the historical coordinate points and the current measurement value, an optimal coordinate estimate is predicted, and the current vein position data in the world coordinate system is replaced with the optimal coordinate estimate.

9. The automated subcutaneous venous blood collection method based on infrared imaging as described in claim 8, characterized in that, The filtered vein location data in the world coordinate system is registered and superimposed with the pre-acquired visible light image data of the target area, including: Feature point detection is performed on the visible light image data to extract blood vessel texture feature points; The same feature point detection is performed on the infrared image data to extract temperature edge feature points; A feature matching algorithm is used to find matching feature point pairs between the visible light image data and the infrared image data. Based on the feature point pairs, the affine transformation parameters between the visible light image data and the infrared image data are calculated; Using the affine transformation parameters, the filtered vein location data in the world coordinate system is mapped to the coordinate system of the visible light image, thereby achieving visual overlay of the vein location data and the visible light image data.

10. The automated subcutaneous venous blood collection method based on infrared imaging as described in claim 9, characterized in that, The step of outputting a positioning result containing the spatial location, orientation angle, and subcutaneous depth information of the subcutaneous vein based on the registered and superimposed data includes: From the registered and superimposed data, the specific values ​​of the corrected vein center coordinates in the world coordinate system are read as the spatial location of the subcutaneous vein. Along the course of the main vein where the corrected vein center coordinates are located, select several consecutive sampling points; Calculate the angle between the line connecting two adjacent sampling points and the horizontal axis, and take the average of all angles to obtain the direction angle; The depth estimate used to generate the corrected vein center coordinates is read as the subcutaneous depth information; The spatial location, orientation angle, and subcutaneous depth information are packaged into a structured data packet, and the data packet is sent to a designated display terminal for display.

Images