Postoperative infant restraint comfort image perception quantification analysis method and system

Abstract

The application discloses a postoperative infant restraint comfort degree image perception quantitative analysis method and system, and relates to the technical field of image processing.The method comprises the following steps: acquiring video images before and after the restraint of an infant, identifying a target region such as a restrained limb, and establishing an individual behavior reference template based on the posture motion characteristics before the restraint; comparing the behavior characteristics after the restraint to screen abnormal regions, combining the abnormal regions with the space, motion direction and action repetition relationship of the restrained limb to calculate a restraint correlation behavior index; when the index exceeds a threshold, analyzing the behavior change after nursing intervention, and further generating an abnormal behavior type, a restraint position and a nursing review suggestion.The application can improve the reliability of restraint-related abnormal behavior judgment and nursing review suggestion generation.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to a method and system for image perception and quantitative analysis of postoperative constraint comfort in infants and young children. Background Technology

[0002] During postoperative recovery, infants and young children often require appropriate restraint of their upper limbs, lower limbs, or other body parts to prevent them from scratching wounds, pulling out IV lines, or tugging on monitoring equipment. Because infants and young children have limited language skills, their discomfort from restraint is usually expressed through behaviors such as limb twisting, localized struggling, repetitive swinging, or changes in posture.

[0003] To facilitate observation of postoperative restraint in infants and young children, existing technologies employ video images for objective analysis of their behavior after restraint. However, this typically involves nursing staff manually observing the child's range of motion, frequency of struggling, and limb posture after restraint, or using image recognition models to detect limb movement characteristics after restraint, and judging whether the child exhibits abnormal behavior based on pre-set uniform standards. However, different infants and young children vary significantly in age, body size, postoperative pain levels, anesthesia recovery status, and daily activity habits. Even under the same restraint conditions, their natural range of motion and behavioral manifestations may differ. Using uniform behavioral standards or analyzing only a single period after restraint makes it difficult to obtain judgment criteria adapted to the individual behavioral characteristics of the target child.

[0004] In particular, infants and young children may experience pain, anesthesia recovery, emotional agitation, and unconscious limb movements simultaneously after surgery. These behaviors may manifest as increased range of motion or repetitive movements on imaging. Therefore, current technology can typically only detect whether the child has experienced significant motor changes, but it is difficult to further determine whether these changes are caused by the specific restraint location. This makes it difficult to provide reliable evidence for nursing staff to determine whether to adjust the restraints or implement targeted nursing care. Summary of the Invention

[0005] To address the problem that existing images of postoperative constraints for infants and young children lack a close correlation between motion changes and specific constraint locations, this invention provides a method and system for quantitative analysis of postoperative constraint comfort in infants and young children based on image perception.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows:

[0007] In the first aspect, this application discloses a method for image-based quantitative analysis of postoperative restraint comfort in infants and young children, including the following steps:

[0008] Acquire the first video image data of the target object before the constraint operation is performed and the second video image data after the constraint operation is performed;

[0009] Target region identification is performed on the first video image data and the second video image data respectively. Then, the posture motion features are extracted from the identification results of the first video image data to establish an individual behavior reference template of the target object. The current behavior features corresponding to the individual behavior reference template are extracted from the identification results of the second video image data. The target region includes the constrained limb region.

[0010] Based on the first degree of difference calculated from the individual behavior reference template and the current behavior characteristics, abnormal areas are screened out. The spatial relationship between the abnormal areas and the restricted limb areas, as well as the movement direction relationship and the repetition relationship of the abnormal areas, are obtained and weighted to obtain the constraint-related behavior index.

[0011] If the constraint-related behavior index exceeds the threshold, the third video image data of the same abnormal area after the nursing staff performs the nursing intervention operation is obtained. The feature is extracted and the difference between the data and the individual behavior reference template is calculated to obtain the second degree of difference. The change in the degree of abnormality is calculated by combining the first degree of difference.

[0012] Behavioral analysis results are generated based on changes in the degree of abnormality and the constraint-related behavioral index. The behavioral analysis results include the type of abnormal behavior, the location of the associated constraint, and nursing review suggestions.

[0013] Secondly, this application discloses an image perception and quantitative analysis system for postoperative restraint comfort in infants and young children, including an image data acquisition module, a behavior feature extraction module, a restraint-related behavior index calculation module, an abnormality degree change calculation module, and a suggestion output module.

[0014] The image data acquisition module is used to acquire the first video image data of the target object before the constraint operation is performed and the second video image data after the constraint operation is performed.

[0015] The behavior feature extraction module is used to identify target regions in the first video image data and the second video image data respectively. Then, it extracts posture and motion features from the identification results of the first video image data to establish an individual behavior reference template for the target object, and extracts the current behavior features corresponding to the individual behavior reference template from the identification results of the second video image data. The target region includes the constrained limb region.

[0016] The constraint-related behavior index calculation module is used to filter out abnormal areas based on the first degree of difference calculated from the difference between the individual behavior reference template and the current behavior characteristics, obtain the spatial relationship between the abnormal area and the constrained limb area, as well as the movement direction relationship and the action repetition relationship of the abnormal area, and perform weighted calculation to obtain the constraint-related behavior index.

[0017] The abnormality degree change calculation module is used to determine whether the constraint-related behavior index exceeds the threshold. If it does, it obtains the third video image data of the same abnormal area after the nursing staff performs the nursing intervention operation, extracts its features and calculates the difference with the individual behavior reference template to obtain the second difference degree. Combined with the first difference degree, it calculates the abnormality degree change.

[0018] The output module is recommended to generate behavioral analysis results based on changes in the degree of abnormality and the constraint-related behavioral index. The behavioral analysis results include the type of abnormal behavior, the location of the associated constraint, and nursing review suggestions.

[0019] Thirdly, this application discloses an image perception and quantitative analysis device for postoperative restraint comfort of infants and young children, including an image acquisition device, a memory, a processor, and a display component.

[0020] The image acquisition device is used to acquire first video image data of the target object before the restraint operation is performed, second video image data after the restraint operation is performed, and third video image data after the nursing intervention operation.

[0021] The memory stores a computer program as well as first video image data, second video image data, and third video image data;

[0022] The processor is connected to the image acquisition device, memory and display components, and is configured to execute computer programs to implement the steps of the aforementioned method for image perception and quantitative analysis of postoperative constraint comfort in infants and young children.

[0023] The display component is used to show the behavioral analysis results.

[0024] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0025] This application establishes an individual behavioral reference template for target infants before restraint, and compares the current post-restraint post-restraint posture and movement characteristics with the target's own behavioral baseline, thereby reducing the impact of individual differences on abnormal behavior identification. By screening abnormal areas and comprehensively considering the spatial relationship, movement direction relationship, and repetition relationship between the abnormal areas and the restrained limbs to calculate the restraint-related behavior index, it can distinguish between general postoperative activities and abnormal behaviors that may be caused by restraint operations. By acquiring third-party video image data after nursing intervention when the restraint-related behavior index exceeds the threshold, and comparing the degree of abnormality before and after nursing intervention, the effectiveness of nursing intervention can be objectively verified. Furthermore, by combining the restraint-related behavior index and changes in the degree of abnormality, it generates abnormal behavior types, associated restraint locations, and nursing review suggestions, thereby achieving individualized identification, associated location, quantitative evaluation, and intervention feedback for postoperative restraint-related abnormal behaviors in infants, providing more accurate and actionable auxiliary information for nursing staff to review and adjust restraint status. Attached Figure Description

[0026] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0027] Figure 1 This is a flowchart of the image perception and quantitative analysis method for postoperative constraint comfort of infants and young children as described in this invention;

[0028] Figure 2 For based on Figure 1 A schematic diagram of key point recognition;

[0029] Figure 3 For based on Figure 1 A diagram illustrating the determination of the repetitive relationship of actions;

[0030] Figure 4 This is a structural block diagram of the image perception and quantitative analysis system for postoperative restraint comfort of infants and young children, as described in Example 2.

[0031] Figure 5 This is a structural block diagram of the image perception and quantitative analysis device for postoperative restraint comfort of infants and young children, as described in Example 3. Detailed Implementation

[0032] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0033] During the postoperative recovery period, infants and young children often require protective restraints on their limbs due to anesthesia recovery, wound pain, and indwelling tubes to prevent them from unconsciously pulling out tubes, scratching wounds, or falling out of bed. However, the restraints themselves can also cause discomfort to the child. For example, if the restraints are too tight, there is insufficient cushioning, or the child is in an improper position, the child may express this discomfort by struggling or twisting.

[0034] In existing technologies, the determination of whether a child is experiencing discomfort due to restraint primarily relies on manual rounds and subjective observation by nursing staff. Specifically, for example, nursing staff need to visually observe the child's limb movements and facial expressions, relying on experience to judge whether the child is in a state of discomfort. Therefore, the following problems may arise: First, infants and young children cannot express discomfort verbally, and the natural agitation during anesthesia recovery and resistance movements caused by restraint are highly similar in appearance, making it difficult to distinguish them visually. This can easily lead to misjudging normal recovery activities as restraint discomfort, or conversely, overlooking genuine restraint discomfort. Second, even if abnormal movements are observed, it is difficult to objectively confirm whether the abnormal movements are indeed caused by restraint, or by other factors such as pain, hunger, or emotions, making it difficult to provide targeted nursing interventions. Third, manual observation involves intermittent rounds, making it impossible to continuously track the child's condition. Furthermore, different nursing staff have different judgment standards, leading to strong subjectivity and a high risk of missed or misjudgment.

[0035] This application uses the infant's natural behavioral state before restraint as an individualized reference benchmark. By comparing posture and movement characteristics before and after restraint, it identifies abnormal behavioral areas that appear after restraint. Furthermore, by combining the spatial location, direction of movement, and repetition of actions between the abnormal areas and the restrained limbs, it quantifies the degree of correlation between the abnormal behavior and the restraint operation. When the correlation is high, the effectiveness of the nursing intervention and the causes of the abnormal behavior are further verified by comparing changes in abnormal behavior before and after nursing intervention. Finally, it generates abnormal behavior types, associated restraint locations, and nursing review recommendations. This forms a closed-loop analytical approach of "individual benchmark establishment—restraint correlation identification—nursing intervention verification—analysis result output."

[0036] After introducing the basic concept of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0037] Example 1

[0038] like Figure 1 The image presented describes a method for image-based quantitative analysis of postoperative restraint comfort in infants and young children, including the following steps:

[0039] S101. Obtain the first video image data of the target object before the constraint operation is performed and the second video image data after the constraint operation is performed.

[0040] Specifically, during the postoperative recovery phase of infants and young children, a continuous video image sequence can be acquired using an image acquisition device positioned beside the bed. This continuous video image sequence consists of multiple video frames arranged in chronological order. It should be understood that the image acquisition device can be a visible light camera or an infrared camera, and the embodiments of the present invention are not limited thereto.

[0041] To facilitate subsequent analysis of behavioral changes over time, the continuous video image sequence can be divided into time windows, that is, the continuous video sequence can be divided into several video segments according to a preset number of consecutive sampling frames. It should be understood that the length of the time window can be set according to the frame rate of the image acquisition device. For example, at a frame rate of 25 frames / second, the preset time window can be set to 1 to 3 seconds, corresponding to 25 to 75 consecutive sampling frames, to balance the integrity of motion capture and the real-time performance of computation. It should also be understood that to control the amount of data to be processed, representative frames can be extracted at fixed intervals within each time window (e.g., one frame every few frames) for subsequent feature extraction and analysis, or a lower analysis frame rate can be used, thereby reducing the computational load while ensuring the integrity of motion capture. The video image data acquired before the constraint operation is implemented is the first video image data, and the video image data acquired after the constraint operation is implemented is the second video image data.

[0042] S102. Target region recognition is performed on the first video image data and the second video image data respectively. Then, posture motion features are extracted from the recognition results of the first video image data to establish an individual behavior reference template for the target object. Current behavior features corresponding to the individual behavior reference template are extracted from the recognition results of the second video image data. The target region includes the constrained limb region.

[0043] Specifically, target regions can be identified from video images using image detection algorithms. Optionally, target regions include infant body regions, facial regions, restrained limb regions, and restraint device regions. It should be understood that restrained limb regions refer to the regions where limbs restricted by restraint devices (such as wrists and ankles) are located. Furthermore, posture motion features are extracted from the recognition results of the first video image data to establish an individual behavior reference template for the target object. Among them, posture features can be represented by the positions of human key points, that is, several predefined human key point positions (such as wrist, elbow, shoulder, ankle, knee, etc. joints) can be extracted from the image using existing human posture estimation algorithms (such as human key point detection models based on convolutional neural networks). Based on the changes in key point positions between adjacent frames, the motion trajectory of each key point is obtained. Further, based on the motion trajectory, the average motion amplitude of the region corresponding to each key point can be calculated, which is used to represent the average displacement of the key point per unit time. It is the average Euclidean distance of the displacement of key points in adjacent frames within a preset time window, and the calculation formula is:

[0044]

[0045] In the formula, This represents the average motion amplitude of the region corresponding to the i-th key point. This indicates the total number of frame intervals between adjacent frames within a preset time window. This represents the coordinate position of the i-th keypoint in frame t.

[0046] like Figure 2 As shown, P1~P 11 The key points of the human body can be seen in the figure, including the target object's facial area, body area, restrained limb area, and restraint device area.

[0047] Therefore, the individual behavior reference template for the target object can be composed of pre-constraint posture features, motion features, facial region features, and body state features. Correspondingly, the current behavioral features corresponding to the individual behavior reference template are extracted from the recognition results of the second video image data, namely, the post-constraint posture features, motion features, facial region features, and body state features. It should be understood that the individual behavior reference template is statistically obtained from a set of feature samples (e.g., frame-by-frame or time-window-by-time feature vectors) extracted during a relatively calm period before constraints. For each feature dimension, the template stores the mean and standard deviation of that dimension on the pre-constraint samples. The means of each dimension together constitute the reference feature vector, and the standard deviation of each dimension is used to characterize the normal fluctuation range of the child's own behavior. Accordingly, the "standard deviation of the features corresponding to the reference template" mentioned later is the standard deviation of each dimension on the pre-constraint samples, and setting the difference threshold to 2 to 3 times the standard deviation of the corresponding dimension has a clear basis.

[0048] In this way, by establishing an individual behavioral reference template for each child before implementing restraint procedures, rather than using a uniform group standard, subsequent judgments of behavioral changes are based on the child's natural state. This approach can accommodate individual differences in age, body size, and activity level among children, thereby improving the accuracy of abnormal behavior identification. It should be understood that this individual behavioral reference template serves as the initial value for the child's individualized behavioral baseline. During postoperative recovery, it can be gradually updated according to S105 as the child's behavior stabilizes, continuously representing the normal level of their current stage.

[0049] It should be understood that, in order for the individual behavior reference template to represent the child's natural state, the first video image data before restraint should be collected during a period when the child is relatively calm. If the child is crying or obviously agitated before restraint, relatively calm segments can be selected, or multiple segments of pre-restraint data can be combined to establish an individual behavior reference template, so as to reduce the impact of baseline bias on subsequent abnormal judgments.

[0050] Specifically, the average motion amplitude of the aforementioned key points can be used. Objectively identify relatively calm periods: Time windows where the mean of the average movement amplitude at key points within a preset time window is lower than a preset calm threshold (e.g., the lower quantile of the distribution of this mean within a period before restraint, or an empirically set low movement amplitude threshold) are defined as relatively calm periods. Video image data from these periods are used to establish individual behavioral reference templates. To ensure statistical sufficiency, the cumulative duration of relatively calm data used to establish the template should not be less than a preset minimum duration (e.g., not less than 10 to 30 seconds). If the relatively calm data collected before restraint is insufficient for this minimum duration, the collection time before restraint can be extended, or supplementary data can be collected during a relatively stable period in the early postoperative restraint state, combining multiple data segments to establish the template. Alternatively, a baseline template for the entire patient population can be used initially, and then switched to an individual template after obtaining sufficient individual calm data. Furthermore, pre-restraint data collection should ideally be performed when the child is in a state of wakefulness or arousal similar to the subsequent comparison period to avoid baseline shifts due to differences between sedation and wakefulness.

[0051] S103. Based on the first degree of difference calculated from the individual behavior reference template and the current behavior characteristics, abnormal areas are screened out. The spatial relationship between the abnormal areas and the restricted limb areas, the movement direction relationship of the abnormal areas, and the repetition relationship of the actions are obtained and weighted to obtain the constraint-related behavior index.

[0052] It should be understood that restlessness is common in infants and young children during the postoperative anesthesia recovery period. These normal recovery activities are visually similar to resistance movements caused by restraint. Judging solely by the magnitude of the movements can easily lead to misinterpreting normal recovery restlessness as restraint discomfort. Therefore, this step analyzes the correlation between abnormal movements and the restrained area in three dimensions: spatial dimension, direction of movement, and repetitiveness of movement. This helps identify truly restraint-related abnormal behaviors from ordinary recovery activities, as detailed below:

[0053] Candidate analysis regions are constructed based on the constrained limb region and its adjacent limb regions. The first degree of difference between the current behavioral characteristics of the candidate analysis regions and the individual behavioral reference template is calculated.

[0054] Specifically, considering that discomfort from constraints often manifests not only in the constrained limb itself but also in adjacent limbs (for example, when the wrist is constrained, the forearm, upper arm, and even the shoulder may experience coordinated struggle), candidate analysis regions are constructed using the constrained limb region and its adjacent limb regions. The first degree of difference characterizes the degree of deviation of the current behavioral characteristics of the candidate analysis region after constraint from the individual's behavioral reference template. It can be obtained by measuring the difference between the posture, movement, and other features of the candidate analysis region and the corresponding features of the template (such as the distance between feature vectors). The greater the difference, the higher the first degree of difference.

[0055] Optionally, the first degree of difference can be represented by the Euclidean distance between the current behavior feature vector of the candidate analysis region and the corresponding feature vector of the individual behavior reference template, calculated as follows:

[0056]

[0057] In the formula, Indicates the first degree of difference. This represents the j-th component of the current behavior feature vector of the candidate analysis region. This represents the j-th component of the feature vector corresponding to the individual behavior reference template. This represents the dimension of the feature vector. It's important to note that features such as pose, motion, facial regions, and body states have different dimensions and numerical ranges (e.g., pixel coordinates can reach hundreds, while action repetition rates range from 0 to 1). Directly calculating the Euclidean distance without processing would allow large-dimensional features to dominate the difference results. Therefore, before calculating the first degree of difference, each feature dimension is standardized (i.e., (f−μ) / σ) or normalized to [0,1] based on the mean and standard deviation of the samples before constraints, making each feature dimensionless and comparable. The feature components in the above formula are the standardized or normalized feature components. Furthermore, weights can be assigned to each dimension according to feature type, and the first degree of difference can be calculated using weighted Euclidean distance. When features such as facial regions are missing due to occlusion or recognition failure, the corresponding dimension is removed from the distance calculation, and the weights of the remaining dimensions are renormalized to avoid artificially increasing or decreasing the first degree of difference for missing dimensions.

[0058] Determine whether the degree of first difference in each candidate analysis region exceeds the threshold; if so, it is determined to be an abnormal region.

[0059] Specifically, the first degree of difference for each candidate analysis region is compared with a preset degree of difference threshold. If the difference exceeds the threshold, the candidate analysis region is determined to be an abnormal region. Optionally, the degree of difference threshold can be adaptively determined based on the fluctuation range of the individual behavior reference template before the child is restrained, for example, set to 2 to 3 times the standard deviation of the corresponding feature of the reference template, so as to avoid misjudging the child's normal movement fluctuations as abnormal.

[0060] Obtain the center position of the abnormal area and the center position of the constrained limb area, and calculate the spatial distance between the two center positions as the spatial relationship;

[0061] Specifically, spatial relationship refers to the spatial distance between the location of the abnormal action and the constraint location. The smaller the spatial distance, the closer the abnormal action is to the constraint location, and the more likely it is to be related to the constraint. The spatial distance can be the Euclidean distance between the center of the abnormal region and the center of the constrained limb region. When there are multiple frames, the average spatial distance of each frame within a preset time window can be taken, calculated using the following formula:

[0062]

[0063] In the formula, This indicates spatial relationships (i.e., spatial distance), and T represents the number of video frames within a preset time window. This indicates the center location of the abnormal region in frame t. This indicates the center position of the constrained limb region in frame t. This indicates the calculation of the Euclidean norm of a vector (i.e., the straight-line distance between the two centers).

[0064] The center position of the abnormal area and the center position of the constrained limb area can be taken as the average value of the coordinates of all key points in the corresponding area (i.e., the centroid of the key points), or as the geometric center of the bounding rectangle of the corresponding area.

[0065] The system obtains the pre-marked key point positions in the abnormal region, obtains the inter-frame motion vector based on the changes in key point positions in adjacent frames, calculates the average motion vector based on the mean of the inter-frame motion vectors within a preset time window, determines the reference direction vector based on the key points and fixed points in the constrained limb region, and calculates the similarity between the average motion vector and the reference direction vector as the motion direction relationship.

[0066] Specifically, the motion direction relationship is used to characterize the degree of consistency between the motion direction of abnormal movements and the constraint direction. It should be understood that when a child struggles due to discomfort from restraint, their limb movement direction is usually towards the direction of breaking free from the restraint, i.e., it tends to be consistent with the constraint direction. First, based on the positional changes of pre-marked key points in the abnormal region across adjacent frames, inter-frame motion vectors are obtained; the average motion vector is obtained by averaging the inter-frame motion vectors within a preset time window; second, a reference direction vector is determined based on the line connecting the key points of the restrained limb region and the fixed point of the restraint device, which represents the constraint direction; finally, the similarity between the average motion vector and the reference direction vector is calculated as the motion direction relationship. Optionally, the similarity can be cosine similarity, which is the ratio of the dot product of two vectors to the product of their magnitudes, calculated using the following formula:

[0067]

[0068] In the formula, Indicates the relationship between the directions of motion. Represents the average motion vector. Represents the reference direction vector. This represents the dot product of the average motion vector and the reference direction vector, with the denominator representing the product of the magnitudes of the two vectors. The value of the motion direction relationship ranges from -1 to 1; the closer the value is to 1, the more consistent the motion direction of the abnormal action is with the constraint direction.

[0069] The reference direction vector is determined as follows: the fixed point of the restraint device is the relatively fixed anchor point of the restraint device such as restraint belt or guardrail in the image, and its position in the image can be marked in advance; the reference direction vector starts from the fixed point of the restraint device and ends at the key point of the restrained limb, pointing to the expected direction of movement of the limb when the child breaks free from the restraint, and this direction is the restraint restriction direction.

[0070] It should be understood that the actual direction of the struggle is somewhat dispersed, and the expected struggle direction differs under different constraint methods and body positions. Therefore, the reference direction vector can be determined separately according to the constraint method and body position: for example, when using wrist / ankle straps for constraint, the line connecting the fixed point to the constrained joint is used as the reference direction; when using guardrails or full-body constraints, the normal direction from the constrained limb to the constraint boundary is used as the reference direction. Corresponding reference direction determination rules can be pre-configured for different constraint methods. To reduce the limitations of a single direction, the movement direction relationship can also be determined using one of the following alternative or supplementary methods: First, set an expected direction cone (angle range) with the reference direction as the axis, and use the degree to which the average movement vector falls within this cone as the movement direction relationship; Second, directly calculate the proportion of the projection component (i.e., the struggle component) of the average movement vector in the direction "away from the fixed point of the constraint device" to its modulus as the movement direction relationship. All of the above methods use "whether the movement tends to break free from the constraint" as the criterion, and can be flexibly selected according to the constraint method and body position.

[0071] It should be noted that the above method based on average motion vectors may cause the average motion vector to approach zero in reciprocating struggle scenarios due to the mutual cancellation of outward and inward movements. This can lead to the denominator of the cosine similarity approaching zero, unstable motion direction relationships, or even inability to calculate them. Reciprocating struggle is a typical manifestation of constraint resistance behavior. Therefore, the motion direction relationship is preferably calculated in a phase-by-phase or frame-by-frame manner: the reciprocating motion is divided into outward and inward phases based on the zero-crossing point of the key point's motion velocity. Only the motion vector of the outward (breakaway) phase and the reference direction vector are used to calculate the directional consistency, or the maximum value of the directional consistency among the phases is taken; alternatively, the proportion of frames in which the motion vector falls within the expected direction cone with the reference direction vector as the axis within a preset time window is used as the motion direction relationship. When the magnitude of the average motion vector is lower than a preset threshold, the above frame-by-frame or phase-by-phase measurement is used instead, or the judgment is based primarily on the repetition relationship of the action, to avoid instability caused by the average vector approaching zero.

[0072] Obtain the sequence of key points in the abnormal area within a preset time window, identify action segments based on this sequence, perform similarity analysis on the action segments, and determine the action repetition relationship by the ratio of the number of action segments of the same type to the total number of action segments within the preset time window.

[0073] Specifically, the action repetition relation is used to characterize the repetitiveness of abnormal actions. It should be understood that resistance behavior caused by restraint often manifests as repetitive, rhythmic struggling movements, while normal awakening activities are more random and irregular. First, based on the sequence of key point positions in the abnormal area within a preset time window, several action segments are identified, for example, by dividing action segments according to the repetitive changes in key point positions. Second, the identified action segments are analyzed for similarity, for example, by clustering similarity based on the trajectory shape, amplitude, and other characteristics of the action segments, and counting the number of action segments belonging to the same category. Finally, the ratio of the number of action segments belonging to the same category within the preset time window to the total number of action segments is determined as the action repetition relation. The value of the action repetition relation ranges from 0 to 1; the higher the ratio, the more repetitive the abnormal action.

[0074] To ensure the statistical validity of action repetition relationships, the aforementioned ratio is only calculated and used when the total number of action segments identified within a preset time window is not less than a preset minimum number of segments (e.g., not less than 3). When the total number of action segments is less than this minimum number of segments, the evidence of repetition within that time window is deemed insufficient, and the action repetition relationship is set to 0 or excluded from the calculation of the constraint-related behavior index. Alternatively, several adjacent time windows may be merged, the time window used for repetition analysis may be extended, and then the calculation may be performed to avoid accidental high repetition rate misjudgments due to insufficient segments. Optionally, action repetition relationships can also be determined using a periodicity measure of keypoint motion. For example, autocorrelation analysis can be performed on the keypoint position sequence, and the intensity of its significant periodic component can be used as a repetition characterization, thus reliably reflecting the degree of action repetition even when segment division is unstable.

[0075] like Figure 3 As shown, the identification and class analysis of the above-mentioned action segments can be carried out in the following manner: When identifying action segments, the continuous sequence of key point positions can be divided into several action segments based on the zero-crossing point of the key point's motion velocity (i.e., the moment when the motion direction reverses) or the local minimum value of the motion amplitude. Each action segment corresponds to a relatively complete reciprocating motion. When performing class analysis, the features of each action segment can be extracted (such as the shape descriptor of the motion trajectory, motion amplitude, duration, and main motion direction, etc.), the feature similarity between each pair of action segments can be calculated (e.g., using Euclidean distance or dynamic time warping distance as a metric), and then similar action segments can be grouped into the same class through clustering algorithms (e.g., K-means clustering or hierarchical clustering).

[0076] The constraint-related behavior index is obtained by normalizing spatial relationships, movement direction relationships, and action repetition relationships and then weighting them.

[0077] Specifically, since spatial relationships, movement direction relationships, and action repetition relationships have different dimensions and value ranges, they need to be normalized to a unified value range [0,1]. Spatial relationships represent spatial distance; the smaller the distance, the stronger the association. Therefore, the spatial distance is reverse-normalized to obtain the spatial proximity D∈[0,1] (for example, D=(D...). max -D c ) / (D max -D min ), or D=1 / (1+D c The larger the value of D, the closer the objects are in space and the stronger the constraint relationship. The motion direction relationship is represented by the cosine similarity Q∈[−1,1], which is linearly mapped to the normalized motion direction relationship Q=(Q+1) / 2∈[0,1]. The action repetition relationship R b Its value is [0,1]. Then, the normalized spatial proximity D, normalized motion direction relationship Q, and action repetition relationship R are analyzed. b We perform a weighted summation to obtain the constrained association behavior index, calculated using the following formula:

[0078]

[0079] In the formula, The index represents the constraint association behavior index, and D represents the spatial proximity obtained after the spatial distance is reverse normalized (∈[0,1], the smaller the spatial distance, the larger its value). This represents the normalized relationship of motion direction (∈[0,1]). This indicates a repetition relationship of actions (∈[0,1]). , , The weight parameters corresponding to spatial relationships, movement direction relationships, and action repetition relationships are respectively, all of which are greater than 0 and the sum of the three is 1, thus constraining the correlation behavior index C. b The value range is [0,1].

[0080] Optionally, weight parameters , , All three are greater than 0 and their sum equals 1. It should be understood that the weights can be flexibly matched according to different clinical scenarios: in scenarios where limb struggle is the primary manifestation, the weights of spatial relationships and movement direction relationships can be increased, for example, by taking... 0.4 0.4 The value is 0.2; in scenarios where repeated scratching is the main characteristic, the weight of the repetitive action relationship can be increased, for example, by taking... 0.3 0.3 It is 0.4. Optionally, , , The values ​​can be 0.2 to 0.5, 0.2 to 0.5, and 0.2 to 0.5, respectively.

[0081] It should be understood that when multiple abnormal regions are obtained after screening, the constraint-related behavior index can be calculated for each abnormal region separately; the abnormal region with the highest constraint-related behavior index is taken as the main abnormal region and participates in the subsequent nursing intervention verification and behavior analysis results generation of S104 and S105; or S104 and S105 can be executed for each abnormal region and the corresponding behavior analysis results can be output respectively.

[0082] Before calculating the constraint-related behavior index, a reliability screening of abnormal regions is performed to exclude those that do not meet the reliability criteria. The remaining abnormal regions are then used in the calculation of the constraint-related behavior index. The reliability criteria include:

[0083] The normalized spatial relationship between the abnormal area and the constrained limb area is not greater than the preset spatial distance threshold, and the movement direction relationship reaches the preset direction consistency threshold or the movement repetition relationship of the abnormal area reaches the preset movement repetition threshold.

[0084] Specifically, to avoid accidental actions unrelated to constraints (such as changes in images caused by caregivers touching the child or the child's blanket sliding) being mistakenly included in the calculation, abnormal areas are first screened for reliability before calculating the constraint-related behavior index.

[0085] It should be noted that the reliability screening and the constraint-related behavior index have different responsibilities and distinct threshold levels: the former is a binary gating system used to eliminate occasional actions unrelated to the constraint (such as caregivers touching or blankets sliding) regardless of the magnitude of the movement, and serves as an admission criterion for "whether to include in the analysis"; the latter provides a continuous correlation strength between the screened abnormal regions and grades them according to the correlation index threshold. To avoid contradictory thresholds, the directional consistency threshold and action repetition threshold of the reliability screening should be consistent with the discrimination level used for the corresponding normalized components in the constraint-related behavior index. Optionally, the preset spatial distance threshold can be set to 0.2 to 0.4 after normalization, the preset directional consistency threshold can be set to 0.6 to 0.8, and the preset action repetition threshold can be set to 0.5 to 0.7. Only abnormal regions that simultaneously meet the criteria of being "spatially close to the constraint location" and having "consistent movement direction or sufficiently repetitive action" are retained for the calculation of the constraint-related behavior index.

[0086] Taking a specific scenario as an example, a postoperative child's right wrist is restrained, and the image acquisition device captures obvious movement in the right wrist area. Calculations show that the normalized spatial distance between the center of this abnormal area and the center of the restrained limb area is 0.15 (less than the preset spatial distance threshold of 0.3), the cosine similarity between its average motion vector and the restraint direction is 0.82 (reaching the preset direction consistency threshold of 0.7), and the proportion of similar struggling action segments within a preset time window is 0.65 (reaching the preset action repetition threshold of 0.6), thus meeting the reliability conditions and being retained; 0.4 0.4 The constraint-related behavior index was calculated with a weighting of 0.2, resulting in a high index. Based on this, it was determined that there was suspected abnormal behavior related to the constraint in the right wrist area.

[0087] In this way, by associating and weighting abnormal actions with the constrained area across three dimensions—spatial relationship, direction of movement, and repetition of action—these three dimensions complement and corroborate each other: spatial relationship dictates that the location of the abnormal action must be close to the constrained area; direction of movement dictates that the action must point in the direction of breaking free from the constraint; and repetition of action dictates that the action must possess the repetitiveness characteristic of constraint resistance. The reliability screening requires that the abnormal area be spatially close to the constrained location and satisfy at least one of "consistent direction of movement" or "repetition of action," thus excluding occasional actions. Based on this, the constraint-related behavior index further weights and fuses the spatial, direction of movement, and repetition of action dimensions. While any single dimension may be occasionally satisfied by ordinary awakening activities, the probability of all three dimensions jointly exhibiting constraint-related characteristics is extremely low. Therefore, this step can effectively distinguish restraint-induced resistance from ordinary agitation during anesthesia recovery, overcoming the defect that normal recovery activities are easily misjudged as restraint discomfort based solely on the amplitude of the movement, thus improving the accuracy and specificity of restraint-related abnormal behavior identification. Furthermore, by combining reliability screening to exclude occasional movements unrelated to restraint, the false positive rate is further reduced, enabling the restraint-related behavior index to more accurately reflect the strength of the correlation between abnormal behavior and restraint.

[0088] S104. Determine whether the constraint-related behavior index exceeds the threshold. If so, obtain the third video image data of the same abnormal area after the nursing staff performs the nursing intervention operation, extract its features and calculate the difference with the individual behavior reference template to obtain the second degree of difference, and combine it with the first degree of difference to calculate the change in the degree of abnormality.

[0089] Specifically, the constraint-related behavior index is compared with a preset correlation index threshold. This correlation index threshold is the low correlation threshold (e.g., 0.5) in S105 below, and both are the same threshold. If the constraint-related behavior index reaches this threshold, it is determined that there is a suspected constraint-related abnormal behavior, and a nursing intervention response analysis is triggered. If the constraint-related behavior index is lower than this threshold, a nursing intervention response is not triggered, and the output is as described in scenario five below. It should be understood that the nursing intervention operations performed by nursing staff may include one or more of the following: loosening restraints, adding cushioning, adjusting body position, soothing, and re-fixation. In order to ensure that the change in the degree of abnormality before and after the nursing intervention can point to the specific cause of constraint, preferably, the nursing intervention performed after triggering the nursing intervention response is mainly an adjustment operation targeting the constraint state, such as loosening restraints, adjusting the tightness of restraints, adding cushioning, adjusting the placement of the restrained limbs, or re-fixation. When the nursing intervention also includes non-restraint factors such as soothing and feeding, the change in the degree of abnormality is used to characterize the remedialability of the abnormal behavior. It is necessary to combine the constraint-related behavior index to comprehensively judge the source of the abnormality, rather than solely relying on this to conclude that the abnormality is caused by constraint.

[0090] To determine the acquisition period of the third video image data used for analysis after nursing intervention and to distinguish between constraint adjustment interventions and non-constraint interventions, nursing intervention event information is received. The nursing intervention event information includes the type of nursing intervention operation, start time, and end time. The acquisition period of the third video image data is determined based on the end time of the nursing intervention operation and in conjunction with a preset post-intervention observation delay. The nursing intervention is classified into constraint adjustment intervention or non-constraint intervention based on the type of nursing intervention operation. The nursing intervention event information can be entered by nursing staff through an interactive interface or obtained from the nursing operation records of the nursing information system.

[0091] The threshold for the correlation index can be set according to the normalized range of the constraint-related behavior index. It can be a boundary value that can distinguish between normal awakening activities and restraint resistance behaviors, such as 0.5. Alternatively, it can be adaptively determined by the constraint-related behavior index statistic (e.g., the mean plus a set standard deviation) of a sample collected early after restraint is applied to the child (at which point the restraint has been established, the direction of restraint, the area of ​​the restrained limb, and the fixing point of the restraint device have all been determined, and the constraint-related behavior index can be calculated). Or it can be determined by the constraint-related behavior index statistic of a historical case group to accommodate individual differences.

[0092] After the nursing staff performs the nursing intervention, third video image data of the same area corresponding to the aforementioned abnormal area (i.e., the candidate analysis area identified as the abnormal area in step S103) is acquired. Feature extraction is performed on the third video image data, and a second degree of difference between it and the individual behavior reference template is calculated. The second degree of difference is obtained for the same area as the first degree of difference, using the same features and calculation method as the first degree of difference, and is used to characterize the degree of deviation of the behavioral characteristics of that area from the individual behavior reference template after the nursing intervention. Furthermore, the change in the degree of abnormality is calculated by combining the first degree of difference (the degree of deviation before the nursing intervention) and the second degree of difference (the degree of deviation after the nursing intervention). The change in the degree of abnormality is used to characterize the degree of improvement or deterioration of abnormal behavior before and after the nursing intervention. The calculation formula is as follows:

[0093] ;

[0094] in, Indicates a change in the degree of abnormality. Indicates the first degree of difference. Indicates the second degree of difference. This represents the zero-prevention constant. Optionally, the zero-prevention constant... To prevent the denominator from being zero, a positive number much smaller than the first degree of difference can be taken, for example, 0.000001 to 0.001.

[0095] It should be understood that when the second degree of difference is less than the first degree of difference, the change in abnormality is positive, indicating that the abnormal behavior has weakened after nursing intervention; when the second degree of difference is greater than the first degree of difference, the change in abnormality is negative, indicating that the abnormal behavior has strengthened after nursing intervention; the larger the absolute value of the change in abnormality, the more significant the change before and after nursing intervention. The range of the change in abnormality is usually between -1 and 1. To ensure that the change in abnormality falls within the above range and is compatible with the subsequent linear mapping based on (change in abnormality + 1) / 2, when the result calculated by the above formula exceeds [-1, 1], the value is taken according to the boundary, that is, the change in abnormality is truncated to [-1, 1].

[0096] In this way, by proactively introducing nursing interventions after detecting suspected constraint-related abnormal behaviors, and using the change in the degree of difference before and after the intervention as a verification signal, static anomaly identification is upgraded to dynamic correlation verification: the constraint-related behavior index solves the problem of "whether the abnormal action is related to the constraint in terms of space, direction, and repetition," while the change in the degree of abnormality further provides verification evidence for "whether the abnormal action is related to the constraint": if the degree of abnormality decreases significantly under targeted nursing interventions that only adjust the constraint state (such as loosening the constraint or adjusting the tightness of the constraint), since the intervention does not change other factors such as soothing or feeding, the abnormal behavior can be confirmed to be related to the constraint with a high degree of confidence; conversely, if the nursing intervention also includes non-constraint factors, the decrease in the degree of abnormality only indicates that the abnormality can be alleviated, and cannot be solely attributed to the constraint. In this case, the constraint-related behavior index is the primary indicator, and the change in the degree of abnormality is secondary, for a comprehensive judgment. This overcomes the deficiency of existing technologies that can only identify abnormalities but cannot confirm the source of the abnormality. In addition, temporal or non-restraint factors such as natural anesthesia recovery, the onset of analgesics, the approach or reassurance of nursing staff, feeding, natural changes in posture, differences in observation time periods, and changes in video occlusion may also affect the change in the degree of abnormality before and after nursing intervention. Therefore, changes in the degree of abnormality should only serve as evidence of the correlation between abnormal behavior and restraint adjustment, rather than the sole basis for determining causality. To reduce the influence of temporal confounding, the observation period after nursing intervention should be temporally close to the period before intervention. If necessary, the natural trend of the child's changes should be considered in a comprehensive judgment, and the decrease in the degree of abnormality should not be simply equated with restraint as the sole cause.

[0097] S105. Generate behavioral analysis results based on changes in the degree of abnormality and the constraint-related behavioral index. The behavioral analysis results include the type of abnormal behavior, the location of the associated constraint, and nursing review suggestions. Specifically, this includes:

[0098] The type of abnormal behavior is determined based on the relationship between the direction of movement and the repetition of actions in the abnormal area;

[0099] Specifically, different types of abnormal behavior exhibit different characteristics in terms of the relationship between movement direction and the repetition of actions. For example, when both the relationship between movement direction and the repetition of actions are high, the abnormal behavior type can be identified as restraint resistance struggle; when the relationship between the repetition of actions is high and the relationship between movement direction is low, the abnormal behavior type can be identified as repetitive scratching; and so on.

[0100] The location of associated constraints is determined by location mapping based on the spatial relationship between the abnormal area and the constrained limb area;

[0101] A mapping relationship between image coordinate regions and human body constraint parts can be pre-established. For example, the image can be divided into several sub-regions corresponding to constraint parts such as the left wrist, right wrist, left ankle, and right ankle. Based on the sub-region into which the center of the abnormal region falls, or based on the minimum spatial distance between the center of the abnormal region and the center of each constrained limb region, the abnormal behavior is mapped to the corresponding constraint part as the associated constraint position. It should be noted that when there are multiple constrained parts, step S103 constructs candidate analysis regions and calculates spatial relationship and other features for each constrained part. The associated constraint position in this step is the constrained part that is closest to the abnormal region (with the strongest spatial relationship). When there is only a single constrained part, the associated constraint position is that constrained part. The position mapping in this step is only used to mark its sub-region in the image, and the two do not form a cycle.

[0102] The constraint-related behavior index is compared with the preset association threshold range to obtain the corresponding association level. The preset review rules are matched with the changes in the degree of abnormality to generate nursing review suggestions.

[0103] Specifically, the preset correlation threshold range is divided into three levels—high correlation, medium correlation, and low correlation—based on preset high correlation and low correlation thresholds. Combined with changes in the degree of abnormality, nursing review suggestions can be generated according to the following preset review rules:

[0104] Scenario 1 (High correlation and significant reduction after nursing intervention): When the constraint-related behavior index reaches the high correlation threshold and the change in the degree of abnormality reaches the preset significant improvement threshold, it is determined that the target abnormal behavior is highly correlated with the location of the constraint, and the degree of abnormality has significantly decreased after nursing intervention. Based on this, a nursing review suggestion of "recommending review and maintaining the current adjustment status" is generated.

[0105] Scenario 2 (Moderate correlation and weakening after nursing intervention): When the constraint-related behavior index is between the low and high correlation thresholds, and the change in the degree of abnormality reaches the preset general improvement threshold, it is determined that the target abnormal behavior may be related to the location of the constraint. After nursing intervention, the degree of abnormality has decreased. Based on this, nursing review suggestions for continued observation and re-verification of the constraint status are generated.

[0106] Scenario 3 (High correlation but no significant change after nursing): When the constraint-related behavior index reaches the high correlation threshold, but the change in the degree of abnormality is between the negative enhancement threshold and the general improvement threshold (i.e., the degree of abnormality has not changed significantly after nursing), considering the possible reasons such as insufficient nursing adjustment, no actual change in the constraint position, too short observation time after nursing, or other synchronous discomfort factors of the child, the source of constraint is not directly denied. Based on this, nursing review suggestions for re-examining the constraint position, tightness, cushioning materials, or body position are generated.

[0107] Scenario 4 (Post-Nursing Abnormal Enhancement): When the change in the degree of abnormality is less than or equal to the negative enhancement threshold (i.e., the degree of abnormality increases after nursing intervention), regardless of whether the constraint-related behavior index is at a high or medium level, it is determined that the target abnormal behavior has increased after nursing intervention and the behavioral response of the current nursing operation does not meet expectations. Based on this, nursing review suggestions are generated to prioritize the review of nursing intervention operations, related constraint positions, limb placement status, and video occlusion or recognition errors.

[0108] Scenario 5 (Low Constraint Correlation): When the constraint correlation behavior index is below the low correlation threshold (i.e., the triggering condition of S104 is not met), the nursing intervention response is not triggered. It is directly determined that the current image behavioral evidence is insufficient to support a clear correlation between the target abnormal behavior and the constraint location, and a prompt to continue routine observation is generated accordingly.

[0109] Scenario 6 (High correlation and generally weakened after nursing): When the constraint-related behavior index reaches the high correlation threshold and the change in the degree of abnormality is between the general improvement threshold and the significant improvement threshold, it is determined that the target abnormal behavior is highly correlated with the location of the constraint, and the degree of abnormality has decreased after nursing intervention but the improvement is limited. Based on this, nursing review suggestions are generated to maintain the current adjustment and continue to review the constraint status and observe subsequent changes.

[0110] Scenario 7 (Moderate correlation and no significant change after nursing): When the constraint-related behavior index is between the low correlation threshold and the high correlation threshold, and the change in the degree of abnormality is between the negative enhancement threshold and the general improvement threshold (i.e., the degree of abnormality has not changed significantly after nursing), the correlation between the target abnormal behavior and the constraint is still uncertain and the intervention has not shown significant effect. Based on this, nursing review suggestions are generated to continue to observe, review the constraint status, and investigate other potential discomfort factors such as pain, hunger, and emotions.

[0111] Optionally, the significant improvement threshold, general improvement threshold, and enhancement threshold can be set according to the range of values ​​of the abnormality degree. For example, the significant improvement threshold can be 0.5 to 0.7, the general improvement threshold can be 0.2 to 0.3, and the enhancement threshold can be 0.2 to 0.3. The high correlation threshold and low correlation threshold can be set according to the range of the constraint correlation behavior index after normalization, for example, 0.7 to 0.8 and 0.4 to 0.5 respectively.

[0112] The behavioral analysis results are formed by combining the type of abnormal behavior, the location of the associated constraint, and nursing review suggestions. These results can be output via a display component (e.g., a screen). Restraint comfort is characterized by the constraint-related behavior index and changes in the degree of abnormality: a higher constraint-related behavior index and a more severe change in the degree of abnormality (i.e., smaller or even negative values) indicate more significant constraint-related discomfort and lower restraint comfort; conversely, a lower constraint-related behavior index and a more improved change in the degree of abnormality after nursing intervention indicate higher restraint comfort. Optionally, the constraint-related behavior index, combined with changes in the degree of abnormality, can be mapped to a restraint comfort level (e.g., comfortable, average, poor) according to a preset range, and output as part of the behavioral analysis results via the display component for cross-referencing with clinical comfort assessments (e.g., FLACC, COMFORT-B scales).

[0113] The behavioral analysis results also include the confidence level, which is calculated by weighting the changes in the degree of anomaly and the constraint-related behavioral index.

[0114] Specifically, the confidence level is used to characterize the credibility of the judgment that "abnormal behavior is related to constraints" in the results of this behavioral analysis. It can be obtained by weighting the constraint-related behavior index and the direction of change in the degree of abnormality after nursing intervention. The higher the constraint-related behavior index and the more the degree of abnormality tends to improve after nursing intervention (the change in the degree of abnormality is positive and larger), the more credible the judgment is, and the higher the confidence level is. Conversely, when the degree of abnormality increases instead of decreasing after nursing intervention (the change in the degree of abnormality is negative), it indicates that the intervention response is inconsistent with the expectation that "abnormality is caused by constraints", and the confidence level should be reduced.

[0115] Alternatively, the formula for calculating the confidence level can be:

[0116]

[0117] In the formula, Indicates the confidence level of the judgment. This represents the constraint association behavior index (its value is [0,1]). The term represents the consistency of the intervention response obtained by linear mapping of the change in the degree of abnormality (∈[0,1], where the value is greater than 0.5 when the change in the degree of abnormality is positive (i.e., when there is improvement) and increases with the increase of improvement, and less than 0.5 when there is aggravation (i.e., when there is aggravation) and decreases with the increase of aggravation). and These are the weights corresponding to the constraint association behavior index and the change in the degree of abnormality, respectively, and their sum equals 1; optionally, It can be taken as 0.5 to 0.7. You can take a value between 0.3 and 0.5 (for example, take...). 0.6 (The confidence level is 0.4), and the calculated confidence level can be further expressed as a percentage.

[0118] Taking a specific scenario as an example, a child with a right wrist restraint was released by a caregiver. Third video image data was collected, and the abnormality level was calculated to be 0.62 (reaching the significant improvement threshold of 0.6). The restraint-related behavior index was 0.90. Previously, the child's restraint-related behavior index had already reached the high correlation threshold, meeting the criteria for scenario one. 0.6 The confidence level is 0.4, and the judgment confidence level Conf = 0.6 × 0.90 + 0.4 × (0.62 + 1) / 2 ≈ 0.86, which is 86%. Based on this, the behavioral analysis results are: the abnormal behavior type is abnormal behavior associated with right wrist constraint, the associated constraint location is the right wrist area, the judgment confidence level is 86%, and the nursing review suggestion is to review the tightness of the constraint and maintain the current adjustment state.

[0119] After generating behavioral analysis results based on changes in the degree of anomaly and the constraint-related behavioral index, it is further determined whether the second degree of difference is lower than the preset stable threshold and whether the change in the degree of anomaly reaches the preset improvement threshold. If so, the feature extraction results corresponding to the third video image data are used as template correction samples, and the individual behavior reference template is updated gradually through weighted gradation.

[0120] Specifically, when the child's condition stabilizes after nursing intervention (i.e., the second degree of difference is below the preset stabilization threshold) and the degree of abnormality truly improves (i.e., the change in the degree of abnormality reaches the preset improvement threshold), it indicates that the child's recovery has stabilized in the current stage after nursing adjustments, and their behavioral characteristics can be used as an individualized behavioral baseline sample for the child in the current stage. It should be noted that the updated individual behavioral reference template is not a fixed, unconstrained natural state before constraint, but rather a dynamically updated "individualized current behavioral baseline" that progresses with the postoperative recovery process. Its initial value is taken from the natural behavioral state before constraint (see S102), and it is progressively modified with stable behavior after nursing intervention when the above stabilization and improvement conditions are met. This is used to characterize the child's current normal level as natural changes such as anesthesia recovery and recovery of activity levels occur. To avoid the baseline gradually drifting due to repeated introduction of stable states under constraint and weakening sensitivity to constraint discomfort, the update ratio is... The smaller value is selected, and updates are only triggered when the child's condition is stable and shows definite improvement. The feature extraction results corresponding to the third video image data are used as template correction samples, and the individual behavior reference template is updated in a weighted, progressive manner. The update method can be expressed as:

[0121]

[0122] In the formula, This indicates the updated individual behavior reference template. This refers to the individual behavior reference template before the update. This represents the feature extraction result (i.e., template correction sample) corresponding to the third video image data. This represents the update ratio, and satisfies the condition that it is greater than 0 and less than 1. Optionally, the update ratio... It can be set to 0.1 to 0.3. The larger the value, the greater the influence of the new sample on the template.

[0123] It should be understood that through gradual updates, individual behavioral reference templates can be gradually adapted to the natural changes in children's recovery from anesthesia, age differences, emotional changes, and recovery of activity levels, thereby maintaining the accuracy of abnormal behavior identification at different stages of postoperative recovery.

[0124] In this way, by combining the correlation strength reflected by the constraint-related behavior index with the intervention response reflected by the change in the degree of abnormality in a two-dimensional discriminant approach, multiple scenarios are identified and corresponding nursing review suggestions are matched. The two work together: the correlation strength is used to determine whether the abnormal behavior is related to the constraint and the degree of correlation, while the intervention response is used to verify the correlation and evaluate the nursing effect. Especially in scenario three (high correlation but no significant change after nursing) and scenario four (increased after nursing), the system does not simply deny the source of the constraint because the nursing is ineffective or the abnormality worsens. Instead, it prompts a review of the nursing operation itself, the location of the constraint, and even the identification error, thereby avoiding omissions and misjudgments that may be caused by a single-dimensional judgment. Combined with the progressive template update after nursing intervention, the system can dynamically adapt to the child's status, forming a complete closed loop of "identification-verification-feedback-update", which significantly improves the safety, pertinence, and intelligence of postoperative constraint nursing.

[0125] Therefore, the image perception and quantitative analysis method for postoperative restraint comfort of infants and young children in this invention establishes individual behavioral reference templates for each child and constructs a restraint-related behavioral index from three dimensions: space, direction of movement, and repetitiveness of movement. This method can objectively and continuously identify abnormal behaviors caused by restraint from normal awakening activities. Furthermore, by introducing nursing interventions and verifying the source of abnormality by observing changes in the degree of abnormality before and after the intervention, this method overcomes the shortcomings of existing technologies that rely solely on subjective human observation, are difficult to distinguish between normal awakening and restraint discomfort, and cannot confirm the source of abnormality. This greatly improves the safety and effectiveness of postoperative restraint care for infants and young children.

[0126] Example 2

[0127] like Figure 4 As shown in this embodiment, an image perception and quantitative analysis system for postoperative restraint comfort in infants and young children is introduced, which includes:

[0128] The image data acquisition module 201 is used to acquire the first video image data of the target object before the constraint operation is performed and the second video image data after the constraint operation is performed.

[0129] The behavior feature extraction module 202 is used to perform target region recognition on the first video image data and the second video image data respectively, and then extract posture motion features from the recognition result of the first video image data to establish an individual behavior reference template of the target object, and extract the current behavior features corresponding to the individual behavior reference template from the recognition result of the second video image data; the target region includes the constrained limb region;

[0130] The constraint-related behavior index calculation module 203 is used to filter out abnormal areas based on the first degree of difference calculated from the difference between the individual behavior reference template and the current behavior characteristics, obtain the spatial relationship between the abnormal area and the constrained limb area, as well as the movement direction relationship and the action repetition relationship of the abnormal area, and perform weighted calculation to obtain the constraint-related behavior index.

[0131] The abnormality degree change calculation module 204 is used to determine whether the constraint-related behavior index exceeds the threshold. If it does, it obtains the third video image data of the same abnormal area after the nursing staff performs the nursing intervention operation, extracts its features and calculates the difference with the individual behavior reference template to obtain the second difference degree, and calculates the abnormality degree change by combining the first difference degree.

[0132] It is recommended that output module 205 be used to generate behavioral analysis results based on changes in the degree of abnormality and the constraint-related behavioral index. The behavioral analysis results include the type of abnormal behavior, the location of the associated constraint, and nursing review suggestions.

[0133] It should be understood that the above-mentioned image data acquisition module 201, behavior feature extraction module 202, constraint association behavior index calculation module 203, anomaly degree change calculation module 204, and suggestion output module 205 are used to implement steps S101 to S105 of the method in Embodiment 1, respectively. The specific implementation method, parameter settings, and technical effects of each module are the same as the corresponding steps in Embodiment 1. For the sake of brevity, they will not be described again here.

[0134] Example 3

[0135] like Figure 5 As shown in the figure, this embodiment introduces an image perception and quantitative analysis device for postoperative restraint comfort in infants and young children, which includes:

[0136] Image acquisition device 301 is used to acquire first video image data of the target object before the constraint operation is performed, second video image data after the constraint operation is performed, and third video image data after the nursing intervention operation.

[0137] The memory 302 stores a computer program as well as first video image data, second video image data and third video image data;

[0138] The processor 303 is connected to the image acquisition device, the memory and the display components, and is configured to execute a computer program to implement the steps of the image perception and quantitative analysis method for postoperative restraint comfort of infants and young children described in Example 1.

[0139] Display component 304 is used to display the results of behavior analysis.

[0140] It should be understood that the apparatus of this embodiment can correspond to the method in Embodiment 1. The image acquisition device 301 is used to provide the video image data required by the method, the processor 303 executes the computer program stored in the memory 302 to implement steps S101 to S105, and the display component 304 is used to display the behavior analysis results generated by the processor 303. The processor can be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits, off-the-shelf programmable gate arrays, or other programmable logic devices, etc., and the embodiments of the present invention are not limited thereto. Other operations and technical effects of each component are the same as the corresponding steps in Embodiment 1, and will not be repeated here for the sake of brevity.

[0141] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A method for image-based quantitative analysis of postoperative restraint comfort in infants and young children, characterized in that, Includes the following steps: Acquire the first video image data of the target object before the constraint operation is performed and the second video image data after the constraint operation is performed; Target region recognition is performed on the first video image data and the second video image data respectively. Then, posture motion features are extracted from the recognition results of the first video image data to establish an individual behavior reference template of the target object. Current behavior features corresponding to the individual behavior reference template are extracted from the recognition results of the second video image data. The target area includes the area of ​​the restrained limbs; Based on the first degree of difference calculated from the individual behavior reference template and the current behavior characteristics, abnormal areas are screened out. The spatial relationship between the abnormal areas and the restricted limb areas, as well as the movement direction relationship and the repetition relationship of the abnormal areas, are obtained and weighted to obtain the constraint-related behavior index. If the constraint-related behavior index exceeds the threshold, the third video image data of the same abnormal area after the nursing staff performs the nursing intervention operation is obtained. The feature is extracted and the difference between the data and the individual behavior reference template is calculated to obtain the second degree of difference. The change in the degree of abnormality is calculated by combining the first degree of difference. Behavioral analysis results are generated based on changes in the degree of abnormality and the constraint-related behavioral index. The behavioral analysis results include the type of abnormal behavior, the location of the associated constraint, and nursing review suggestions.

2. The method for image perception and quantitative analysis of postoperative restraint comfort in infants and young children according to claim 1, characterized in that, The specific steps for filtering out abnormal regions based on the first degree of difference calculated from the individual behavior reference template and the current behavioral characteristics are as follows: Candidate analysis regions are constructed based on the constrained limb region and its adjacent limb regions. The first degree of difference between the current behavioral characteristics of the candidate analysis regions and the individual behavioral reference template is calculated. Determine whether the degree of first difference in each candidate analysis region exceeds the threshold; if so, it is determined to be an abnormal region.

3. The method for image perception and quantitative analysis of postoperative restraint comfort in infants and young children according to claim 2, characterized in that, The specific steps for obtaining the spatial relationship, movement direction relationship, and action repetition relationship between the abnormal region and the constrained limb region, and then performing a weighted calculation to obtain the constraint-related behavior index, are as follows: Obtain the center position of the abnormal area and the center position of the constrained limb area, and calculate the spatial distance between the two center positions as the spatial relationship; The system obtains the pre-marked key point positions in the abnormal region, obtains the inter-frame motion vector based on the changes in key point positions in adjacent frames, calculates the average motion vector based on the mean of the inter-frame motion vectors within a preset time window, determines the reference direction vector based on the key points and fixed points in the constrained limb region, and calculates the similarity between the average motion vector and the reference direction vector as the motion direction relationship. Obtain the sequence of key points in the abnormal area within a preset time window, identify action segments based on this sequence, perform similarity analysis on the action segments, and determine the action repetition relationship by the ratio of the number of action segments of the same type to the total number of action segments within the preset time window. The constraint-related behavior index is obtained by normalizing and weighting spatial relationships, movement direction relationships, and action repetition relationships.

4. The method for image perception and quantitative analysis of postoperative restraint comfort in infants and young children according to claim 3, characterized in that, Before calculating the constraint-related behavior index, a reliability screening of abnormal regions is performed to exclude those that do not meet the reliability criteria. The remaining abnormal regions are then used in the calculation of the constraint-related behavior index. The reliability criteria include: The normalized spatial relationship between the abnormal area and the constrained limb area is not greater than the preset spatial distance threshold, and the movement direction relationship reaches the preset direction consistency threshold or the movement repetition relationship of the abnormal area reaches the preset movement repetition threshold.

5. The method for image perception and quantitative analysis of postoperative restraint comfort in infants and young children according to claim 1, characterized in that, The formula for calculating the change in the degree of abnormality is: ； in, Indicates a change in the degree of abnormality. Indicates the first degree of difference. Indicates the second degree of difference. This represents the zero constant.

6. The method for image perception and quantitative analysis of postoperative restraint comfort in infants and young children according to claim 1, characterized in that, The behavioral analysis results generated based on changes in the degree of anomaly and the constraint-related behavioral index specifically include: The type of abnormal behavior is determined based on the relationship between the direction of movement and the repetition of actions in the abnormal area; The location of associated constraints is determined by location mapping based on the spatial relationship between the abnormal area and the constrained limb area; The constraint-related behavior index is compared with the preset association threshold range to obtain the corresponding association level. Combined with the changes in the degree of abnormality, the preset review rules are matched to generate nursing review suggestions. The abnormal behavior type, associated constraint location, and nursing review suggestions are combined to form the behavior analysis results.

7. The method for image perception and quantitative analysis of postoperative restraint comfort in infants and young children according to claim 6, characterized in that, The behavioral analysis results also include the confidence level, which is calculated by weighting the changes in the degree of anomaly and the constraint-related behavioral index.

8. The method for image perception and quantitative analysis of postoperative restraint comfort in infants and young children according to claim 1, characterized in that, After generating behavioral analysis results based on changes in the degree of anomaly and the constraint-related behavioral index, it is further determined whether the second degree of difference is lower than the preset stable threshold and whether the change in the degree of anomaly reaches the preset improvement threshold. If so, the feature extraction results corresponding to the third video image data are used as template correction samples, and the individual behavior reference template is updated gradually through weighted gradation.

9. A system for image-based quantitative analysis of postoperative restraint comfort in infants and young children, characterized in that, It includes: The image data acquisition module is used to acquire the first video image data of the target object before the constraint operation is performed and the second video image data after the constraint operation is performed. The behavior feature extraction module is used to identify target regions in the first video image data and the second video image data respectively, and then extract posture motion features from the recognition results of the first video image data to establish an individual behavior reference template of the target object, and extract the current behavior features corresponding to the individual behavior reference template from the recognition results of the second video image data. The target area includes the area of ​​the restrained limbs; The constraint-related behavior index calculation module is used to filter out abnormal areas based on the first degree of difference calculated from the difference between the individual behavior reference template and the current behavior characteristics, obtain the spatial relationship between the abnormal area and the constrained limb area, as well as the movement direction relationship and the action repetition relationship of the abnormal area, and perform weighted calculation to obtain the constraint-related behavior index. The abnormality degree change calculation module is used to determine whether the constraint-related behavior index exceeds the threshold. If it does, it obtains the third video image data of the same abnormal area after the nursing staff performs the nursing intervention operation, extracts its features and calculates the difference with the individual behavior reference template to obtain the second difference degree, and combines the first difference degree to calculate the abnormality degree change. The suggested output module is used to generate behavioral analysis results based on changes in the degree of abnormality and the constraint-related behavioral index. The behavioral analysis results include the type of abnormal behavior, the location of the associated constraint, and nursing review suggestions.

10. A device for image perception and quantitative analysis of postoperative restraint comfort in infants and young children, characterized in that, It includes: An image acquisition device is used to acquire first video image data of the target object before the constraint operation is performed, second video image data after the constraint operation is performed, and third video image data after the nursing intervention operation. The memory stores a computer program as well as the first video image data, the second video image data, and the third video image data. A processor, connected to an image acquisition device, a memory, and a display component, and configured to execute the computer program to implement the steps of the image perception and quantitative analysis method for postoperative constraint comfort of infants and young children as described in any one of claims 1-8; The display component is used to display the results of behavioral analysis.

Images