A new-born baby nursing table with image acquisition and transmission functions

Abstract

The present application relates to the technical field of medical monitoring, in particular to a neonatal nursing table with image acquisition and transmission functions. Through the respiratory monitoring system, the mechanical and optical characteristic time sequence of each region in the video is synchronously analyzed, the trajectory closure degree is calculated in the phase space to accurately identify the neonatal respiratory activity region, and differential video compression coding is implemented on the respiratory region and the background region according to the same, so that the weak respiratory details are clearly restored at the remote monitoring end without increasing the transmission bandwidth, and false respiratory pause misjudgment is effectively avoided.

Description

Technical Field

[0001] This invention relates to the field of medical monitoring technology, specifically to a neonatal care station with image acquisition and transmission functions. Background Technology

[0002] Neonatal intensive care units (NICUs) are generally equipped with video monitoring systems for non-contact, 24 / 7 observation of infants. In specific scenarios, such as jaundice treatment, newborns need to be exposed to blue light of a specific wavelength for extended periods. This single-spectrum illumination significantly reduces the contrast of the infant's skin texture, making traditional image gradient-based feature extraction algorithms ineffective. More importantly, the respiratory movements of newborns are extremely weak, and the pixel displacement they cause in the video image is often less than one pixel.

[0003] Existing video coding standards (such as H.264 or H.265 / HEVC) generally employ a rate-distortion optimization (RDO)-based bitrate control strategy. Under this strategy, the video encoder tends to classify low-amplitude texture variations in the image as non-critical visual redundancy or sensor noise, thus compressing or filtering these weak signals at a high rate (smoothing) during quantization. This coding mechanism leads to the loss of detail in the infant's chest and abdomen displayed on the monitoring terminal, presenting a static state similar to apnea, which can easily cause misinterpretation by medical personnel.

[0004] Furthermore, in low-light or monochromatic lighting environments, image sensors generate significant Gaussian white noise or thermal noise. Traditional motion detection algorithms (such as frame difference or optical flow methods) struggle to effectively distinguish weak biological signs from random noise under low signal-to-noise ratio conditions because their signal amplitudes are often on the same order of magnitude. Simply increasing the overall video transmission bitrate can alleviate the loss of detail, but it significantly increases the bandwidth burden on the hospital's internal network and cannot fundamentally solve the noise interference problem. Therefore, existing technologies are insufficient to effectively locate weak respiratory signs and guide the video encoder to retain critical details in low-bandwidth, low-contrast environments. Summary of the Invention

[0005] To address the technical challenge of effectively capturing weak respiratory signs and guiding the video encoder to preserve critical details in low-bandwidth, low-contrast environments, this invention aims to provide a neonatal care table with image acquisition and transmission capabilities. The specific technical solution adopted is as follows:

[0006] This invention proposes a neonatal care table with image acquisition and transmission functions, comprising a care table body, on which a video monitoring device is installed, including a respiratory monitoring system with image acquisition and transmission functions, used to monitor and analyze the respiratory movement images of the newborn. The respiratory monitoring system with image acquisition and transmission functions includes:

[0007] Data acquisition module: used to acquire the mechanical and optical feature values ​​of each feature region unit in each frame of respiratory motion image of a newborn, and to construct the mechanical feature time sequence and optical feature time sequence of each feature region unit corresponding to each frame of respiratory motion image;

[0008] Feature analysis module: used to calculate the degree of closure of the time-series trajectory formed in phase space based on the time-series trajectories of the mechanical feature time-series and the optical feature time-series, as a closure index;

[0009] Decision and segmentation module: used to select respiratory seed point regions from all feature region units based on the closure index of all feature region units, and perform region growth on the respiratory seed point regions to obtain respiratory feature regions and background regions;

[0010] Feedback control module: Used to compress each frame of image based on the breathing feature region and the background region using preset differentiated compression transformation rules to generate a video stream.

[0011] Furthermore, the mechanical characteristic values ​​include:

[0012] The motion estimation device is used to obtain the motion vectors of the subcutaneous muscles and soft tissues within the current feature region unit, and the vertical component of the motion vector is used as the mechanical feature value.

[0013] Furthermore, the optical characteristic values ​​include:

[0014] Read the arithmetic mean of the grayscale values ​​of all pixels in the current feature region unit under the blue light channel, and use the arithmetic mean as the optical feature value.

[0015] Furthermore, the method for constructing the mechanical feature time series and the optical feature time series includes:

[0016] The first preset number of breathing motion images preceding each frame of breathing motion image are used as the corresponding historical motion images. The mechanical feature values ​​of each feature region unit in each frame of breathing motion image and all corresponding historical motion images are arranged in chronological order to determine the initial sequence of mechanical features. Based on the principle of obtaining the initial sequence of mechanical features, the initial sequence of optical features is determined according to the optical feature values. The initial sequences of mechanical features and optical features are preprocessed to determine the temporal sequence of mechanical features and the temporal sequence of optical features.

[0017] Furthermore, the preprocessing process includes:

[0018] The mechanical feature values ​​in the initial mechanical feature sequence are normalized to a specific interval to obtain a normalized mechanical feature time series sequence; the optical feature values ​​in the initial optical feature sequence are normalized to a specific interval to obtain a normalized optical feature time series sequence; the normalized mechanical feature time series sequence is smoothed using a time-domain smoothing filter to obtain a smoothed mechanical feature time series sequence; the normalized optical feature time series sequence and the smoothed mechanical feature time series sequence are then decentered.

[0019] Furthermore, the method for calculating the closure index includes:

[0020] Using the optical feature time sequence as the horizontal axis and the mechanical feature time sequence as the vertical axis, a phase space coordinate system is constructed. The coordinate values ​​of each frame in the phase space coordinate system are substituted into the shoelace formula to obtain the directed area enclosed by the signal trajectory corresponding to each feature region unit. The directed area is used as the closure index.

[0021] Furthermore, the method for filtering the respiratory feature region and the background region includes:

[0022] Region growth is performed using all respiratory seed point regions as initial seed points. Feature region units within the preset neighborhood of a seed point that meet the region growth conditions are used as new seed points to continue region growth until no new seed points can be found. The region growth conditions include: when the absolute value of the closure index of the feature region unit within the preset neighborhood of the seed point is greater than a preset decision threshold, and the closure index of the neighboring feature region unit has the same positive or negative sign as the closure index of the respiratory seed point, the feature region unit in that neighborhood is used as a new seed point. All feature region units containing seed points are considered as respiratory feature regions, and other feature region units are considered as background regions.

[0023] Furthermore, the method for setting the preset decision threshold includes:

[0024] The median of the absolute values ​​of the closure indices of all feature region units is calculated, and the median is weighted by a preset first multiple to obtain a first threshold. The maximum value of the first threshold and the preset minimum threshold is selected as the initial decision threshold. The preset decision threshold is determined by multiplying the preset second multiple and the initial decision threshold.

[0025] Furthermore, the compression transformation rules include:

[0026] Based on the division results of the breathing feature region and background region of all feature region units, a compression quality adjustment map is set. The compression quality adjustment map generated in the current frame is superimposed and analyzed with the basic quantization parameters of the current frame to obtain the transformation coefficients of all feature region units in the current frame. The image of the current frame is compressed according to the transformation coefficients.

[0027] Furthermore, the method for generating the basic quantization parameters includes:

[0028] The second number of pre-set breathing motion images preceding each frame are taken as the corresponding historical motion images; the second number of frames following the current frame are taken as the corresponding frames of the current frame. The bitrate control algorithm built into the video encoder is used to analyze the historical motion images of the corresponding frames and calculate the basic quantization parameters for the current frame.

[0029] The present invention has the following beneficial effects:

[0030] This invention achieves multi-dimensional acquisition of respiratory signals in a low-contrast blue light environment by simultaneously acquiring mechanical and optical dual-modal features through a data acquisition module. The feature analysis module robustly identifies respiratory movements with phase lag characteristics from noise using detrending and centering processing and closure index calculation. The decision and segmentation module generates precise region indicator masks through threshold judgment and region growing, ensuring the coherence and accuracy of the respiratory region. The feedback control module constructs a compression quality adjustment map based on the region indicator mask and overlays basic quantization parameters, achieving detail enhancement in the respiratory region and noise suppression in the background region. Ultimately, this invention effectively locks weak respiratory signs and guides the video encoder to retain key details under low contrast and without increasing bandwidth. Attached Figure Description

[0031] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a structural block diagram of a respiratory monitoring system with image acquisition and transmission functions provided in one embodiment of the present invention. Detailed Implementation

[0033] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a neonatal care table with image acquisition and transmission functions proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0034] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0035] The following description, in conjunction with the accompanying drawings, details a specific solution for a neonatal care station with image acquisition and transmission functions provided by the present invention.

[0036] This invention provides a neonatal care table with image acquisition and transmission functions, including a care table body, on which a video monitoring device is installed, including a respiratory monitoring system with image acquisition and transmission functions, for monitoring and analyzing the respiratory motion images of newborns.

[0037] It should be noted that the respiratory monitoring system with image acquisition and transmission function is equipped with a high-definition RGB image sensor, an image signal processor (ISP), and a video encoder (such as a hardware encoding core that supports the H.264 or H.265 standard).

[0038] Please see Figure 1 The diagram illustrates a structural block diagram of a respiratory monitoring system with image acquisition and transmission functions according to an embodiment of the present invention. The system includes:

[0039] Data acquisition module 101: used to acquire the mechanical and optical feature values ​​of each feature region unit in each frame of respiratory motion image of a newborn, and to construct the mechanical feature time sequence and optical feature time sequence of each feature region unit corresponding to each frame of respiratory motion image.

[0040] Because the amplitude of a newborn's breathing fluctuations is extremely small, its displacement in an image is often less than one pixel. These weak vital signs are not uniformly distributed throughout the body, but are highly concentrated in localized areas such as the chest and abdomen. Analyzing the entire image frame would result in a sharp drop in signal-to-noise ratio due to interference from numerous irrelevant background areas, making it impossible to detect sub-pixel-level subtle changes. Therefore, this invention decomposes the image into several feature region units, capturing these localized, subtle vital activities within these units. Furthermore, considering that in the context of neonatal jaundice phototherapy, single-wavelength blue light irradiation significantly reduces the contrast of skin surface texture, and the grayscale changes in skin caused by breathing are extremely small; and that the respiratory movements of newborns are extremely weak, causing chest and abdominal displacements of less than one pixel, in low-light, high-noise environments, this sub-pixel-level displacement signal is on the same order of magnitude as the random noise generated by the image sensor (such as Gaussian white noise or thermal noise). Therefore, this invention collects the mechanical and optical characteristic values ​​of the feature region units and constructs a mechanical and optical characteristic time-series sequence for each feature region unit corresponding to each frame of respiratory motion images for comprehensive analysis of the temporal changes in neonatal breathing.

[0041] In this embodiment of the invention, a high-definition RGB image sensor is used to acquire each frame of the newborn's respiratory motion image. Each macroblock or coding tree unit (CTU) is used as a feature region unit, and an address index is set for each feature region unit. The mechanical and optical feature values ​​of each frame of the newborn's respiratory motion image are collected on the feature region unit.

[0042] It should be noted that when acquiring the mechanical and optical feature values ​​of each frame of breathing motion image, the system uses hardware timestamps (PTS) and feature region unit indexes to ensure that the mechanical and optical feature values ​​strictly correspond to the same physical time and the same spatial region, thus ensuring that subsequent analysis is based on the mechanical and optical dual responses of the same feature region unit in the same video frame.

[0043] Preferably, in some possible implementations of the embodiments of the present invention, the mechanical characteristic values ​​include:

[0044] The chest and abdominal fluctuations caused by respiration physically manifest as reciprocating mechanical displacement of biological tissue in the direction of gravity. To capture this weak displacement signal, this embodiment of the invention uses a motion estimation device to acquire the motion vectors of subcutaneous muscles and soft tissues within the current feature region unit and extracts the vertical component of these motion vectors. This data directly quantifies the instantaneous displacement of subcutaneous muscles and soft tissues in the vertical direction. Unlike the conventional residual absolute value (SAD), the vertical motion vector has directionality (positive values ​​represent downward displacement, negative values ​​represent upward displacement, or vice versa), and its sign change strictly corresponds to the inspiratory and expiratory phases of respiration, thus preserving complete phase information.

[0045] It should be noted that, under normal configuration, the motion estimation unit is mainly used to find the best matching block to eliminate temporal redundancy. This system configures the motion estimation module to sub-pixel accuracy mode (i.e., 1 / 4 pixel accuracy) via the driver interface. In this mode, the encoder performs multiphase filtering interpolation on each frame, thereby calculating a displacement vector smaller than one physical pixel spacing.

[0046] Preferably, in some possible implementations of the embodiments of the present invention, the optical characteristic values ​​include:

[0047] Since blue light therapy is commonly used to treat neonatal jaundice, skin texture details (color / edges) are severely lost, making it difficult to rely solely on texture tracking. However, the chest and abdominal movements caused by breathing alter the curvature and angle of the skin surface. According to Lambert's cosine law, changes in the angle of incidence and the angle of reflection directly lead to regular fluctuations in the surface reflection brightness received by the sensor. Therefore, the brightness value of the feature region unit is selected as the optical feature value.

[0048] Specifically, the system reads the grayscale values ​​of all pixels within the current feature region unit's coverage area in the blue light channel, calculates their arithmetic mean, and uses the arithmetic mean as the optical feature value.

[0049] Preferably, in some possible implementations of the embodiments of the present invention, in order to capture the phase lag characteristics caused by the viscoelasticity of biological tissues, the system cannot rely solely on the instantaneous value at the current moment, but must analyze the signal evolution trajectory over a continuous time period. Therefore, it is necessary to construct a mechanical feature time series and an optical feature time series based on the mechanical feature values ​​and optical feature values, respectively. The method for constructing the mechanical feature time series and the optical feature time series includes:

[0050] The first preset number of breathing motion images preceding each frame of breathing motion image are used as the corresponding historical motion images. The mechanical feature values ​​of each feature region unit in each frame of breathing motion image and all corresponding historical motion images are arranged in chronological order to determine the initial sequence of mechanical features. Based on the principle of obtaining the initial sequence of mechanical features, the initial sequence of optical features is determined according to the optical feature values. The initial sequences of mechanical features and optical features are preprocessed to determine the temporal sequence of mechanical features and the temporal sequence of optical features.

[0051] In one specific implementation of this invention, considering that the respiratory rate of a newborn is about 40-60 breaths per minute, and the video frame rate acquired by the high-definition RGB image sensor used in this embodiment is 30fps, the first number of frames L is set to 50 frames, which can completely cover each respiratory cycle of the newborn.

[0052] It should be noted that in the initial stage of system startup (i.e., when the cumulative number of video frames is less than the first number of frames), the number of historical motion images is insufficient, and the system will temporarily output invalid mechanical or optical feature values ​​(such as 0).

[0053] Preferably, in some possible implementations of the embodiments of the present invention, the preprocessing process includes:

[0054] (1) Normalization

[0055] Since mechanical eigenvalues ​​are displacements with positive and negative values ​​(units of 1 / 4 pixel), while optical eigenvalues ​​are non-negative mean brightness values ​​(units of gray levels), their dimensions, value ranges, and physical units are inconsistent. To construct a proportionally scaled phase space coordinate system in subsequent steps, the system performs the following normalization process:

[0056] The system reads the search range parameters of the motion estimation device. To preserve the sign of the mechanical eigenvalues, they need to be normalized to... For intervals, the system uses the following formula to calculate feature region units. Normalized mechanical eigenvalues ​​in frame t :

[0057]

[0058] in This is a sub-pixel precision multiple (in this embodiment of the invention, it is set to 1 / 4 pixel precision). ); The search range parameter is obtained by reading the encoder configuration file (set to 64 in this embodiment, but can be adjusted according to the actual application scenario). For feature region units The mechanical characteristic value in frame t; This represents the maximum displacement that the motion estimation device can theoretically detect under the current encoder hardware configuration; by analyzing the ratio of the mechanical characteristic value to this maximum displacement, the mechanical characteristic value is normalized to... Interval.

[0059] The system reads the color bit depth of the image sensor. In order to normalize the optical eigenvalues ​​to For intervals, the system uses the following formula to calculate feature region units. Normalized optical eigenvalues ​​in frame t :

[0060]

[0061] in, For feature region units The optical characteristic value in the t-th frame; For the color bit depth of the image sensor; It is a very small positive number (in this embodiment, it is set to be...). This is to prevent computational crashes under extreme abnormal configurations (such as errors in bit depth parameter reading leading to a zero denominator). Because the range of optical eigenvalues ​​is... Color bit depth calculation via image sensor This indicates the number of binary bits required to store the brightness of one more pixel, thereby normalizing the optical feature values ​​to... Interval.

[0062] In one specific implementation of this invention, the color bit depth of the image sensor is set to 8 bits. This indicates that 255 bits are needed to store the brightness of one more pixel.

[0063] (2) Time-domain smoothing filter

[0064] Since the mechanical feature values ​​are obtained by the motion estimation device through a block matching algorithm, and are affected by image noise and sub-pixel interpolation errors, the normalized mechanical feature time series may contain high-frequency jitter or abrupt changes. To extract the true tissue displacement trend, the system needs to perform temporal smoothing filtering on the mechanical feature time series. The temporal smoothing filtering process is as follows:

[0065] Let the time sequence of mechanical characteristics in the queue be... The system uses a 3-point moving average filter to smooth the sequence, resulting in the processed sequence. :

[0066]

[0067] For the beginning and end boundary points of the sequence, the boundary values ​​can be replicated for calculation. This step effectively suppresses quantization noise caused by non-biological signs, making the waveform of the mechanical characteristics smoother and closer to the real respiratory mechanical waves.

[0068] It should be noted that other time-domain smoothing filtering methods, such as Gaussian filtering, may also be used in other implementations of the present invention, which are not limited or elaborated here.

[0069] (3) Decentralization

[0070] To eliminate the DC drift component caused by slow changes in ambient light or overall translation of the infant's position, the system must convert the signal to the AC domain centered at the origin. Therefore, it is necessary to perform a decentering operation on the normalized optical characteristic time series and the smoothed mechanical characteristic time series. The specific operation is as follows:

[0071] The system extracts the normalized optical feature time series, using the elements in this series... For example, the decentralized optical eigenvalues ​​are calculated using the following formula:

[0072]

[0073] in, It is the arithmetic mean of the elements in the normalized optical feature time series; represents the centered optical feature value of element m in the normalized optical feature time series; is the optical characteristic value of element m in the normalized optical characteristic time series.

[0074] Based on the principle of decentralization of optical feature time series, a decentralization operation is performed on the smoothed mechanical feature time series.

[0075] After the above preprocessing, the initial sequences of mechanical features and optical features are used to obtain the time-series sequences of mechanical features and optical features.

[0076] Feature analysis module 102: used to calculate the degree of closure of the time-series trajectory formed in phase space based on the time-series trajectory formed by the mechanical feature time-series sequence and the optical feature time-series sequence, as a closure index.

[0077] Under the extreme conditions of neonatal phototherapy, the magnitudes of mechanical displacement and optical brightness changes caused by respiration are extremely small, and are at the same level as the sensor noise. On their individual time-series waveforms, the respiratory signal is completely submerged in noise, making analysis difficult. Considering the viscoelasticity of biological tissues and the regularity of respiration, based on the Lissajous figure principle: in phase space, the composite trajectory of two signals with the same frequency and a phase difference will form a closed elliptical loop, i.e., a hysteresis loop. Its rotation direction (clockwise or counterclockwise) depends on the sign of the phase difference, while the size of the enclosed area depends on the signal amplitude and the magnitude of the phase difference. Conversely, if it is random noise, there is no fixed phase-locked relationship between the two signals, and the trajectory will exhibit disordered Brownian motion near the origin, failing to form a stable closed loop. Therefore, this embodiment of the invention calculates the degree of closure of the time-series trajectory formed in phase space based on the mechanical and optical characteristic time-series sequences, using this as a closure index. The closure index characterizes the degree of similarity between a feature region unit and a respiratory feature region; the larger the closure index, the stronger and more periodically stable the respiratory activity in the feature region, and the higher the similarity with the respiratory feature region.

[0078] Preferably, in some possible implementations of the embodiments of the present invention, the method for calculating the closure index includes:

[0079] According to the Lissajous figure principle, in phase space, the composite trajectory of two signals with the same frequency and a phase difference will form a closed elliptical loop, i.e., a hysteresis loop. Its rotation direction (clockwise or counterclockwise) depends on the sign of the phase difference, while the size of the enclosed area depends on the signal amplitude and the magnitude of the phase difference. Conversely, if it is random noise, there is no fixed phase-locked relationship between the two signals, and the trajectory will exhibit disordered Brownian motion near the origin, failing to form a stable closed loop. Therefore, the degree of closure of the time series trajectory can be analyzed by examining the directed area of ​​the closed hysteresis loop formed by the time series trajectories of mechanical and optical characteristic time series in phase space, thus calculating the closure index. The calculation method for the closure index is as follows:

[0080] Using the optical feature time sequence as the horizontal axis and the mechanical feature time sequence as the vertical axis, a phase space coordinate system is constructed. The coordinate values ​​of each frame in the phase space coordinate system are substituted into the shoelace formula. By calculating the cross product of adjacent point vectors, the directed area of ​​this polygonal trajectory is obtained. The directed area of ​​the polygonal trajectory is the directed area enclosed by the signal trajectory corresponding to each feature region unit. The directed area is used as the closure index.

[0081] It should be noted that the shoelace formula is a technical method well known to those skilled in the art, and will not be elaborated or limited here.

[0082] As an example, in a specific implementation of this invention, in the feature region unit For each video frame k, extract the mechanical and optical feature values ​​(sequence elements are 1 to L) of the corresponding mechanical and optical feature time sequences; construct a state vector. ,in, Let be the optical feature value of the k-th frame; The mechanical feature value of the k-th frame and the state vector at the next time step. Calculate the area enclosed by the signal trajectories between two adjacent video frames. :

[0083]

[0084] The mathematical logic of this formula lies in the fact that, in two-dimensional plane geometry, the cross product of two vectors is numerically and strictly equal to the directed area of ​​the parallelogram enclosed by the two vectors. By summing up the areas of all infinitesimal elements, the characteristic region unit is obtained. Total directed area :

[0085]

[0086] Where k takes the maximum value When formulas use subscripts and (The last element of the sequence); if we take This exceeds the sequence range, therefore an upper limit is set. Since the signal trajectory sweeps through a triangle formed by the origin, point k, and point k+1 in phase space, the cross product of two vectors is the area of ​​the parallelogram, therefore it needs to be multiplied by... Only by summing the triangles generated from two adjacent video frames can the directed area of ​​the closed hysteresis loop formed by the time sequence trajectory be obtained, and the directed area is used as the closure index.

[0087] It should be noted that if the characteristic region unit has effective respiratory activity, due to the constant phase difference and unidirectional trajectory rotation, most of the infinitesimal elements... Having the same sign, therefore the sum is The absolute value is relatively large.

[0088] If the feature region unit contains only noise, the trajectory will oscillate randomly. The signs alternate randomly, and the positive and negative signs cancel each other out during the accumulation process, making... The result approaches zero.

[0089] Through this integral operation, the algorithm naturally filters out zero-mean random noise, achieving amplification and locking of weak but ordered respiratory signals. Therefore, the calculated closure index... It can be directly used as a feature region unit. The respiratory confidence index is then passed to the subsequent decision-making and segmentation modules.

[0090] Decision and segmentation module 103: used to select respiratory seed point regions from all feature region units based on the closure index of all feature region units, perform region growth on the respiratory seed point regions, and obtain respiratory feature regions and background regions.

[0091] Since the closure index of each feature region unit calculated by the feature analysis module 102 measures the similarity between that feature region unit and the respiratory feature region to a certain extent, a preset initial decision threshold can be used to filter out the respiratory seed point region based on the closure index. Because biological tissues have spatial connectivity, the skin undulations caused by respiration are necessarily continuously distributed, and within the same respiratory cycle, the motion phases of adjacent regions should be close (manifested as the rotation direction of the hysteresis loop being consistent). This invention utilizes this physical characteristic for region growing to complete edge regions that may be missed by the threshold due to slightly weak signals and to remove isolated noise points, thereby filtering out the respiratory feature region and background region from all feature region units.

[0092] Preferably, in some possible implementations of the embodiments of the present invention, the method for filtering the respiratory feature region and the background region includes:

[0093] Using all respiratory seed point regions as initial seed points, region growth is performed. Feature region units within a preset neighborhood of each seed point that meet the region growth conditions are then used as new seed points to continue the process until no new seed points are found. The region growth conditions are defined as follows: a feature region unit within a preset neighborhood of a seed point is only used as a new seed point if its closure index meets the following two conditions:

[0094] (1) The absolute value of the closure index of the feature region unit in the neighborhood is greater than the preset decision threshold, wherein the preset decision threshold should be less than the initial decision threshold, allowing the edge region signal to be weak, but still higher than the background noise;

[0095] (2) The closure index of the characteristic region unit in the neighborhood has the same positive or negative sign as the closure index of the respiratory seed point. This condition requires that the rotation direction of the hysteresis loop in the neighborhood (clockwise or counterclockwise) is consistent with that of the seed point. If the directions are opposite, it means that the phase difference between the two is opposite and they do not belong to the same co-moving biological tissue (which may be the source of interference from the opposite movement). Therefore, the characteristic region unit in the neighborhood is not absorbed and is used as a new seed point.

[0096] Specifically, the absolute value of the closure index measures the intensity of the respiratory signal in a characteristic region to some extent. If the closure index is greater than 0, it indicates that the trajectory rotates counterclockwise, which physically corresponds to the optical characteristic phase leading the mechanical characteristic phase; if the closure index is less than 0, it indicates that the trajectory rotates clockwise, which physically corresponds to the optical characteristic phase lagging behind the mechanical characteristic phase; if the closure index is equal to 0, it indicates that the signal in this region is entirely composed of zero-mean random noise, or that the region is completely still. Due to the randomness of the noise, the positive and negative areas of the trajectory cancel each other out during the accumulation (integration) process, resulting in a net area of ​​zero.

[0097] In one specific implementation of this invention, the system examines the four neighboring (upper, lower, left, and right) feature regions of each seed point. If a feature region meets the region growth conditions, it is determined to be a respiratory activity region.

[0098] Preferably, in some possible implementations of the embodiments of the present invention, the method for setting the preset decision threshold includes:

[0099] To adapt to different lighting conditions and sensor noise levels, the system first calculates an adaptive noise baseline. The system measures the baseline noise level by statistically analyzing the median absolute value of the closure index of all feature region units. This value can effectively eliminate the interference of local outliers and robustly reflect the average noise intensity of the image background. If statistical measures such as mean and standard deviation, which are sensitive to outliers, are used to estimate the noise baseline, the value is easily distorted, leading to an artificially high threshold and thus missing weak breathing signals.

[0100] However, in extreme cases where the infant is not in the frame or is completely still (e.g., during apnea), the entire image consists of background noise, and the median only represents the sensor's thermal noise level. If a multiple relationship is directly used as the initial decision threshold, slightly larger random noise might be misjudged as a breathing signal (i.e., the "avalanche effect"). To prevent such misjudgments, the system sets a preset minimum threshold and then uses a preset first multiple to weight the median, obtaining a first threshold. The maximum value of the first threshold and the preset minimum threshold is selected as the initial decision threshold, ensuring that the initial decision threshold can adaptively fluctuate with noise when the signal-to-noise ratio is high. In environments with extremely low signal-to-noise ratios or complete darkness, the initial decision threshold is clamped at a safe lower limit, i.e., the preset minimum threshold, thus shielding false alarms in purely noisy environments.

[0101] The preset decision threshold is determined by multiplying the preset second multiple by the initial decision threshold. The second multiple should be less than the first multiple, so that the signal in the edge region is weaker, but still needs to be higher than the background noise.

[0102] In one specific implementation of this invention, the preset first multiple is set to 3. This value is intended to resist the influence of possible local interference or non-ideal noise distribution in the image. It is an empirical value based on robust statistics and can be adjusted according to the specific implementation environment. The preset minimum threshold is an empirical constant determined based on the lowest detectable respiratory signal intensity measured by the system under standard test conditions. The preset second multiple is set to 0.5. The preset decision threshold is set to half of the initial decision threshold, which is intended to allow weaker signals in the edge area, but still need to be higher than the background noise. It can be adjusted according to the specific implementation environment.

[0103] Feedback control module 104: Used to compress each frame of image based on the breathing feature region and the background region using a preset differentiated compression transformation rule to generate a video stream.

[0104] Considering that in existing technologies, standard video encoders (such as H.264 / HEVC) employ a global bitrate control strategy based on rate-distortion optimization, applying uniform quantization processing to the entire frame image, this can lead to serious problems in neonatal respiratory monitoring scenarios. The grayscale texture changes of less than one pixel in the neonatal chest and abdomen caused by weak breathing are comparable in magnitude to sensor noise and are therefore judged as visual redundancy by the encoder. During quantization, these changes are filtered out along with the noise, causing the infant's breathing area in the remote monitoring image to appear as a static, uniform color block—a phenomenon known as "false apnea," which can easily lead to clinical misjudgment. Therefore, this invention applies different compression transformation rules to the breathing feature region and background region obtained by the decision and segmentation module 103. This achieves the goal of breaking the balance of global rate-distortion optimization without increasing (or even reducing) the total transmission bandwidth, forcibly preserving the subtle texture details of the breathing area, and fundamentally eliminating the risk of information loss and clinical misjudgment caused by the encoding process.

[0105] It should be noted that, in this embodiment of the invention, after quantization and entropy coding are completed, the generated video stream packet is sent to the remote monitoring terminal via the network communication module. At this time, the feature data intercepted by the data acquisition module 101 and the constructed mechanical feature time series and optical feature time series have completed their mission and do not participate in subsequent transmission, thereby achieving data decoupling between the analysis link and the transmission link. The standard video stream received by the remote terminal already contains the visual effects of "respiratory enhancement" and "background noise reduction," allowing doctors to see clear vital sign details without the need for a special decoder.

[0106] Preferably, in some possible implementations of the embodiments of the present invention, the compression transformation rule includes:

[0107] To preserve the texture fluctuation details of the breathing feature region while suppressing noise in the background region, a compression quality adjustment map needs to be set based on the division results of the breathing feature region and the background region of all feature region units. The compression quality adjustment map generated in the current frame is then overlaid and analyzed with the basic quantization parameters of the current frame. The basic quantization parameters measure the current network bandwidth. By overlaying and analyzing them with the compression quality adjustment map, it is possible to achieve local quality enhancement of key vital sign regions while maintaining the stability of the overall video stream. Finally, the transform coefficients of all feature region units in the current frame are obtained, and the image of the current frame is compressed according to the transform coefficients.

[0108] Specifically, to facilitate the analysis of the segmentation results between the breathing feature region and the background region, in one specific implementation of this invention, a binary mask is constructed for each frame. In the binary mask, the label value of the breathing feature region is set to 1, and the label value of the background region is set to 0. The system sets the compression quality adjustment map according to the binary mask under the following conditions:

[0109] For feature region units in the mask with a label value of 1 (i.e., breathing feature regions), the system sets the corresponding compression quality adjustment value to a significant negative offset value. (In the embodiments of the present invention, the following are taken) In mainstream video coding standards such as H.264 / H.265, decreasing the quantization parameter (QP) by 6 is equivalent to halving the quantization step size, thus doubling the coding accuracy. In low-contrast blue light environments, the grayscale changes in skin caused by a newborn's breathing are extremely small, often spanning only 2 to 3 grayscale levels. Conventional quantization steps easily smooth these tiny grayscale steps into a single color block. By applying an offset of -6, the system forces the encoder to use a finer quantization scale in this area, thereby fully preserving these subtle texture fluctuations and preventing the "pseudo-apnea" phenomenon.

[0110] For feature regions with a label value of 0 in the mask (i.e., background regions), the system sets the corresponding compression quality adjustment value to a gentle positive offset value. (In the embodiments of the present invention, the following are taken) This strategy leverages the human eye's insensitivity to high-frequency noise in static background areas (such as bed sheets or equipment casings). By appropriately increasing the quantization step size, the encoder can more effectively filter out thermal noise generated by the sensor. This operation saves bitrate resources and effectively compensates for the additional bandwidth consumption caused by the improvement in breathing area quality, thereby maintaining the stability of the overall transmission bitrate of the video stream.

[0111] It should be noted that the negative offset value and positive offset value It aims to preserve the texture fluctuation details of the breathing feature area while suppressing noise in the background area, and can be automatically adjusted according to the specific implementation environment.

[0112] Finally, the generated compressed quality adjustment map is superimposed and analyzed with the basic quantization parameters of the current frame to obtain the transformation coefficients of all feature region units in the current frame; and the image of the current frame is compressed according to the transformation coefficients.

[0113] As an example, in a specific implementation of this invention, the formula for calculating the transformation coefficients can be expressed as:

[0114]

[0115] in, Representing feature region units Transformation coefficients; The basic quantization parameters for the current frame; For feature region units The corresponding compression quality adjustment value in the compression quality adjustment map. The calculation formula of the transformation coefficient achieves differentiated compression of the breathing feature region and the background region by superimposing the compression quality adjustment value set by the division result of the breathing feature region and the background region on the basis of the basic quantization parameters, thereby ensuring that the texture fluctuation details of the breathing feature region are preserved while suppressing the noise of the background region without increasing the bandwidth.

[0116] It should be noted that in the breathing feature region, a smaller transform coefficient preserves the texture fluctuation details of the breathing feature region, preventing the phenomenon of "false breathing apnea"; in the background region, a larger transform coefficient quantizes the high-frequency noise figure to zero, saving bitrate resources and effectively compensating for the additional bandwidth consumption brought about by the improvement of breathing region quality.

[0117] Preferably, in some possible implementations of the embodiments of the present invention, the method for obtaining the basic quantization parameters includes:

[0118] The second number of pre-set breathing motion images preceding each frame are taken as the corresponding historical motion images; the second number of frames following the current frame are taken as the corresponding frames of the current frame. The bitrate control algorithm built into the video encoder is used to analyze the historical motion images of the corresponding frames and calculate the basic quantization parameters for the current frame.

[0119] Specifically, to address the timing coordination issue between the analysis algorithm's latency and real-time video stream transmission, this system introduces a look-ahead buffer at the front end of the video encoding pipeline. This buffer is a first-in, first-out (FIFO) frame storage queue, with its length set to the second number of frames. For each video frame... After generating the compressed quality adjustment map for the frame, the system retrieves the corresponding video frame from the head of the pre-analysis buffer. At this point, the frame has been in the buffer for a second number of frame time units, awaiting encoding. The system synchronously sends this video frame and the newly generated compression quality adjustment map into the encoding core of the video encoder. The video encoder then processes the frame... During formal encoding, adaptive quantization mode is enabled. This is done during the processing of the first frame. When defining a feature region unit, the video encoder analyzes the historical motion image of the corresponding frame based on the bitrate control algorithm to calculate the basic quantization parameters for the current frame.

[0120] It should be noted that each video frame acquired by the system is first stored in a pre-analysis buffer. The system's feature analysis module 102 reads the data in the buffer and performs non-destructive feature calculations. Only after the feature analysis module 102 generates the basic quantization parameters for the current frame will the frame be sent to the compression core of the video encoder for final encoding. This mechanism ensures that the basic quantization parameters strictly apply to their corresponding video frames, eliminating control misalignment caused by computational delays.

[0121] In summary, this invention achieves multi-dimensional acquisition of respiratory signals in a low-contrast blue light environment by simultaneously acquiring mechanical and optical dual-modal features through a data acquisition module; the feature analysis module robustly identifies respiratory movements with phase lag characteristics from noise by utilizing detrending and centering processing and closure index calculation; the decision and segmentation module generates accurate region indicator masks through threshold judgment and region growing, ensuring the coherence and accuracy of the respiratory region; and the feedback control module constructs a compression quality adjustment map based on the region indicator mask and superimposes basic quantization parameters, achieving detail enhancement of the respiratory region and noise suppression of the background region. Ultimately, this invention effectively locks weak respiratory signs and guides the video encoder to retain key details under low contrast and without increasing bandwidth.

[0122] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0123] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A neonatal care table with image acquisition and transmission functions, comprising a care table body, wherein a video monitoring device is installed on the care table body, characterized in that, It also includes a respiratory monitoring system with image acquisition and transmission capabilities, used to monitor and analyze respiratory motion images of newborns. The respiratory monitoring system with image acquisition and transmission capabilities includes: Data acquisition module: used to acquire the mechanical and optical feature values ​​of each feature region unit in each frame of respiratory motion image of a newborn, and to construct the mechanical feature time sequence and optical feature time sequence of each feature region unit corresponding to each frame of respiratory motion image; Feature analysis module: used to calculate the degree of closure of the time-series trajectory formed in phase space based on the time-series trajectories of the mechanical feature time-series and the optical feature time-series, as a closure index; Decision and segmentation module: used to select respiratory seed point regions from all feature region units based on the closure index of all feature region units, and perform region growth on the respiratory seed point regions to obtain respiratory feature regions and background regions; Feedback control module: Used to compress each frame of image based on the breathing feature region and the background region using preset differentiated compression transformation rules to generate a video stream.

2. The neonatal care table with image acquisition and transmission function according to claim 1, characterized in that, The mechanical characteristic values ​​include: The motion estimation device is used to obtain the motion vectors of the subcutaneous muscles and soft tissues within the current feature region unit, and the vertical component of the motion vector is used as the mechanical feature value.

3. The neonatal care table with image acquisition and transmission function according to claim 1, characterized in that, The optical characteristic values ​​include: Read the arithmetic mean of the grayscale values ​​of all pixels in the current feature region unit under the blue light channel, and use the arithmetic mean as the optical feature value.

4. The neonatal care table with image acquisition and transmission function according to claim 1, characterized in that, The methods for constructing the mechanical feature time series and the optical feature time series include: The first preset number of breathing motion images preceding each frame of breathing motion image are used as the corresponding historical motion images. The mechanical feature values ​​of each feature region unit in each frame of breathing motion image and all corresponding historical motion images are arranged in chronological order to determine the initial sequence of mechanical features. Based on the principle of obtaining the initial sequence of mechanical features, the initial sequence of optical features is determined according to the optical feature values. The initial sequences of mechanical features and optical features are preprocessed to determine the temporal sequence of mechanical features and the temporal sequence of optical features.

5. The neonatal care table with image acquisition and transmission function according to claim 4, characterized in that, The preprocessing process includes: The mechanical feature values ​​in the initial mechanical feature sequence are normalized to a specific interval to obtain a normalized mechanical feature time series sequence; the optical feature values ​​in the initial optical feature sequence are normalized to a specific interval to obtain a normalized optical feature time series sequence; the normalized mechanical feature time series sequence is smoothed using a time-domain smoothing filter to obtain a smoothed mechanical feature time series sequence; the normalized optical feature time series sequence and the smoothed mechanical feature time series sequence are then decentered.

6. The neonatal care table with image acquisition and transmission function according to claim 5, characterized in that, The method for calculating the closure index includes: Using the optical feature time sequence as the horizontal axis and the mechanical feature time sequence as the vertical axis, a phase space coordinate system is constructed. The coordinate values ​​of each frame in the phase space coordinate system are substituted into the shoelace formula to obtain the directed area enclosed by the signal trajectory corresponding to each feature region unit. The directed area is used as the closure index.

7. The neonatal care table with image acquisition and transmission function according to claim 1, characterized in that, The method for filtering the respiratory feature region and the background region includes: Region growth is performed using all respiratory seed point regions as initial seed points. Feature region units within the preset neighborhood of a seed point that meet the region growth conditions are used as new seed points to continue region growth until no new seed points can be found. The region growth conditions include: when the absolute value of the closure index of the feature region unit within the preset neighborhood of the seed point is greater than a preset decision threshold, and the closure index of the neighboring feature region unit has the same positive or negative sign as the closure index of the respiratory seed point, the feature region unit in that neighborhood is used as a new seed point. All feature region units containing seed points are considered as respiratory feature regions, and other feature region units are considered as background regions.

8. The neonatal care table with image acquisition and transmission function according to claim 7, characterized in that, The method for setting the preset decision threshold includes: The median of the absolute values ​​of the closure indices of all feature region units is calculated, and the median is weighted by a preset first multiple to obtain a first threshold. The maximum value of the first threshold and the preset minimum threshold is selected as the initial decision threshold. The preset decision threshold is determined by multiplying the preset second multiple and the initial decision threshold.

9. The neonatal care table with image acquisition and transmission function according to claim 1, characterized in that, The compression transformation rules include: Based on the division results of the breathing feature region and background region of all feature region units, a compression quality adjustment map is set. The compression quality adjustment map generated in the current frame is superimposed and analyzed with the basic quantization parameters of the current frame to obtain the transformation coefficients of all feature region units in the current frame. The image of the current frame is compressed according to the transformation coefficients.

10. The neonatal care table with image acquisition and transmission function according to claim 9, characterized in that, The method for generating the basic quantization parameters includes: The second number of pre-set breathing motion images preceding each frame are taken as the corresponding historical motion images; the second number of frames following the current frame are taken as the corresponding frames of the current frame. The bitrate control algorithm built into the video encoder is used to analyze the historical motion images of the corresponding frames and calculate the basic quantization parameters for the current frame.

Images