Emergency operation examination intelligent evaluation system based on light weight visual identification

Abstract

The application discloses an emergency practical operation examination intelligent evaluation system based on light-weight visual identification, relates to the technical field of artificial intelligence visual identification, and comprises the following steps: acquiring a practical operation video collected by a single ordinary camera device, dividing a center area and a peripheral area of a chest compression area, constructing a center apparent return sequence and a peripheral related return sequence, extracting inside-out asynchronous rebounding strength and peripheral final state return progress, calculating rebounding integrity coefficients and effective compression values of a single compression release period, and outputting a final score result in combination with compression rhythm. The application can improve the accuracy of rebounding integrity and effective compression determination, and reduce evaluation cost.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence visual recognition technology, and in particular to an intelligent assessment system for emergency practical examinations based on lightweight visual recognition. Background Technology

[0002] In CPR practical exams, the assessment typically includes key operational quality indicators such as compression location, rhythm, depth, and sufficient rebound. To meet the needs of grassroots testing sites, vocational schools, and training institutions for low-cost, easily deployable, and rapidly implementable intelligent assessment, the use of single-channel ordinary cameras and lightweight visual recognition models to automatically analyze emergency practice processes has become an important development direction for intelligent practical exams. However, in actual testing, the chest compression area of ​​the training mannequin is not a single rigid structure. After repeated compressions, a slack can easily form between the chest skin and the internal compression support structure. This results in a phenomenon where the central surface area recovers first during a single compression and release, while the peripheral area recovers later. At this point, the central chest area in the video may appear to have completed its return to its original position, but the internal effective return is not yet fully complete. This can easily cause intelligent assessments based on ordinary video to misjudge the apparent return as complete rebound, leading to a distortion in the judgment of the completeness and effectiveness of compression release.

[0003] Existing technologies have two main shortcomings in addressing the aforementioned problems. Firstly, some solutions rely on pressure sensors, depth sensors, or dedicated feedback mannequins to directly acquire compression depth and rebound status. While this improves detection accuracy, it is costly, complex to deploy, and poorly adaptable, making it difficult to meet the needs of widespread application in ordinary practical examination settings. Secondly, while some video analysis-based solutions lower the equipment barrier, they typically judge rebound status based solely on surface undulations in the central chest area, changes in body posture, or displacement information in a single area. They lack the ability to analyze the temporal differences between apparent recovery in the central chest area and the recovery of peripheral areas, making it difficult to identify situations where surface recovery precedes internal recovery due to excess chest skin laxity. Consequently, they are prone to systematically overestimating the adequacy of rebound, leading to insufficient objectivity and reliability in the final evaluation results. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing practical assessment methods based on ordinary videos, which are difficult to accurately determine the integrity of rebound and are prone to distortion of effective pressing evaluation. The invention proposes an intelligent assessment system for emergency practical examinations based on lightweight visual recognition.

[0005] To address the problems existing in the prior art, the present invention adopts the following technical solution: An intelligent assessment system for emergency practical examinations based on lightweight visual recognition includes: The region segmentation module is used to acquire the practical video captured by a single-channel ordinary camera device, extract positioning points and contour boundaries from the image frames of the practical video through a lightweight visual recognition model, and divide the pressing area into a central area and a peripheral area. The displacement construction module is used to calculate the optical flow field between adjacent image frames, and obtain the normal displacement of the pixel by combining the normal projection direction. Based on the normal displacement in the central area and the peripheral area, the center apparent retrieval sequence and the peripheral entanglement retrieval sequence are constructed respectively. The intensity extraction module is used to determine the release interval of each compression release cycle based on the central apparent return sequence. Within the release interval, the module performs time series analysis on the central apparent return sequence and the peripheral entanglement return sequence to extract the intensity of the asynchronous rebound sign and the peripheral terminal return progress. The press correction module is used to calculate the rebound integrity coefficient based on the asynchronous rebound intensity and the peripheral final state return progress, and to generate the effective press value of a single press release cycle by combining the effective displacement of the center apparent return sequence and the peripheral entangled return sequence. The intelligent evaluation module is used to generate the final score result based on the rebound integrity coefficient and effective pressure value of each press-release cycle, combined with the press rhythm.

[0006] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention divides the chest compression area in a practical video into a central and peripheral region, and constructs a central apparent return sequence and a peripheral entanglement return sequence respectively. This allows for the differentiation and characterization of the superficial chest recovery and effective internal return transmission process during compression and release. Based on this, by extracting the intensity of asynchronous rebound signs and the peripheral terminal return progress, the rebound state of a single compression and release cycle is specifically corrected, thereby avoiding directly judging the apparent return of the central chest region as complete rebound. This enables more accurate identification of the rebound appearance distortion caused by excess chest skin laxity, improving the authenticity and stability of rebound integrity judgment.

[0007] 2. This invention further combines the commonly supported effective displacement amount with the rebound integrity coefficient to generate an effective pressing value for a single pressing and releasing cycle. It also combines the pressing rhythm to comprehensively score the entire practical operation process, so that the final evaluation result simultaneously reflects the pressing depth, rebound sufficiency, and continuous operation quality. This method does not rely on dedicated depth sensing equipment and high-cost detection devices. It can realize the automatic evaluation of emergency practical examinations based only on a single ordinary camera device and lightweight visual recognition processing. This reduces the system deployment cost and the threshold for use, while improving the objectivity, consistency, and scalability of the scoring results. Attached Figure Description

[0008] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a functional module diagram of an intelligent assessment system for emergency practical examinations based on lightweight visual recognition, provided in an embodiment of the present invention. Figure 2 This is a flowchart illustrating an intelligent assessment method for emergency practical examinations based on lightweight visual recognition, provided in an embodiment of the present invention. Detailed Implementation

[0009] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0010] Example: This example provides an intelligent assessment system for emergency practical examinations based on lightweight visual recognition. See [link to example]. Figure 1 Specifically, including: The region segmentation module is used to acquire the practical video captured by a single-channel ordinary camera device, extract positioning points and contour boundaries from the image frames of the practical video through a lightweight visual recognition model, and divide the pressing area into a central area and a peripheral area. The displacement construction module is used to calculate the optical flow field between adjacent image frames, and obtain the normal displacement of the pixel by combining the normal projection direction. Based on the normal displacement in the central area and the peripheral area, the center apparent retrieval sequence and the peripheral entanglement retrieval sequence are constructed respectively. The intensity extraction module is used to determine the release interval of each compression release cycle based on the central apparent return sequence. Within the release interval, the module performs time series analysis on the central apparent return sequence and the peripheral entanglement return sequence to extract the intensity of the asynchronous rebound sign and the peripheral terminal return progress. The press correction module is used to calculate the rebound integrity coefficient based on the asynchronous rebound intensity and the peripheral final state return progress, and to generate the effective press value of a single press release cycle by combining the effective displacement of the center apparent return sequence and the peripheral entangled return sequence. The intelligent evaluation module is used to generate the final score result based on the rebound integrity coefficient and effective pressure value of each press-release cycle, combined with the press rhythm.

[0011] In an embodiment of the present invention, a practical video captured by a single-channel ordinary camera is acquired. A lightweight visual recognition model is used to extract positioning points and contour boundaries from the image frames of the practical video, and a central area and a peripheral area are divided within the pressing area. This includes: Extract the left and right shoulder positioning points, the center point of the pressed palm, and the outer contour boundary of the chest from the image frames of the practical video; The horizontal reference vector in front of the chest is determined based on the positioning points of the left and right shoulders, and the normal projection direction in front of the chest is constructed based on the horizontal reference vector in front of the chest. Calculate the minimum distance from the center point of the pressing palm to the outer contour boundary of the chest, and use it as the current pressing action scale; Using the center point of the pressing palm as the center, divide the chest compression area into a central area and a peripheral area according to the current compression scale; This embodiment uses a single-channel ordinary USB camera as the video acquisition terminal. The camera is fixed at a 45-degree angle above and to the side of the chest of the CPR training mannequin. The vertical height of the camera from the plane of the mannequin's chest is set to 1.2 to 1.5 meters. The acquisition resolution of the camera is set to 1920×1080, and the acquisition frame rate is set to 30 frames per second. This is used to acquire a continuous video sequence of the entire CPR practice process of the examinee. The lightweight visual recognition model used in this embodiment is a combination architecture of the YOLOv8n lightweight object detection network and the MobileNetV2 lightweight human pose estimation network. The model pre-training process is completed using a human pose and mannequin contour dataset containing CPR practice scenarios. The number of training iterations is set to 300 rounds, and the initial learning rate is set to 0.001. After training, the model achieves a key point and contour localization accuracy of over 98% in the target scene, which is used to extract the required localization points and contour boundary information from video image frames.

[0012] Specifically, the acquired video sequence is decoded frame by frame to obtain consecutive image frames, denoted as the t-th frame. t is a positive integer, and the initial frame t=1 corresponds to the start time of the actual pressing action. Each frame image... Input the pre-trained lightweight visual recognition model and extract the candidate's left shoulder localization point from the model output. Right shoulder positioning point Press the center of your palm and the set of simulated human chest outer contour boundaries Left shoulder positioning point Positioning point of right shoulder All measurements use the pixel center coordinates of key points on the human shoulder, pressing the center point of the palm. The pixel center coordinates of the palm area that contacts the simulated human's chest during the pressing action are used to define the outer contour boundary of the simulated human's chest. It is an ordered set of continuous pixel coordinates of the outer edge of the visible area in front of a human chest.

[0013] Based on the positioning point of the left shoulder and right shoulder positioning point Determine the transverse reference vector in front of the chest The calculation formula is: in It is a two-dimensional planar vector. The x-component is the difference between the x-coordinate of the right shoulder positioning point and the x-coordinate of the left shoulder positioning point. The y-component is the difference between the y-coordinates of the right shoulder positioning point and the left shoulder positioning point. The reason for using the left and right shoulder positioning points to construct the lateral reference vector is that the line connecting the left and right shoulders is consistent with the lateral direction of the simulated person's chest, which can eliminate the influence of coordinate system tilt caused by the candidate's position deviation and the camera equipment angle deviation, and provide a reference direction that matches the pressing plane for subsequent normal projection.

[0014] Construct the normal projection direction of the chest based on the transverse reference vector of the chest. The calculation formula is: in The horizontal reference vector in front of the chest The y-component, The horizontal reference vector in front of the chest The x-component, The horizontal reference vector in front of the chest The length of the mold, The normal vector is the unit normal vector. The reason for constructing the normal projection direction in this way is that the normal vector is perpendicular to the line connecting the left and right shoulders, which corresponds to the depth direction of the simulated chest compression action. This can project the optical flow displacement in the two-dimensional plane onto the normal dimension consistent with the compression action, eliminating the interference of invalid displacement in the horizontal and vertical directions in the plane, and accurately representing the depth fluctuations of the chest area during the compression process.

[0015] Calculate the minimum distance from the center point of the pressing palm to the outer contour boundary of the chest, and use this distance as the current pressure scale. The calculation formula is: Where b is the set of the outer contour boundaries of the chest. Any pixel coordinate point in the data, Press the center point of the palm The Euclidean distance to boundary point b, min represents the minimum value among all distances. The reason for using the minimum distance from the center point of the pressing palm to the chest boundary as the scale of the pressing action is that this distance can adaptively represent the effective range of the pressing action in the current frame, regardless of the distance of the camera device or the scaling ratio of the image, ensuring that the division between the central area and the peripheral area always matches the size of the actual pressing area.

[0016] Using the center point of the pressing palm as the center, divide the chest compression area into a central area and a peripheral area according to the current compression scale. The delineation formula is as follows Surrounding area The delineation formula is as follows Where p is the image frame Any pixel in From pixel p to the center of the pressing palm The Euclidean distance. With the center point of the pressed palm as the center, and... A circular area with a radius of half the radius was designated as the central region. This central region was used to characterize the apparent recovery process of the pectoral skin surface during the compression-release process. Half of the inner radius The annular region with an outer radius is designated as the peripheral region, which is used to characterize the recovery process of the internal compression structure of the simulated human through the periphery of the chest skin. The reason for adopting this dual-region division is that the central region corresponds to the direct point of application of the compression action and can most directly reflect the apparent undulation changes of the chest skin surface, while the peripheral region corresponds to the boundary between the chest skin and the internal compression structure, which can effectively capture the hysteretic displacement changes caused by the rebound of the internal structure, providing two independent sets of observation areas for the subsequent extraction of asynchronous rebound features.

[0017] After completing the initial division of the central and peripheral regions, pixel and physical distance calibration is performed. The calibration process is as follows: Using the maximum horizontal width of the simulated human chest's outer contour boundary set as a reference, the actual physical size corresponding to this width is the standard horizontal width of the simulated human chest, 25 centimeters. The number of pixels corresponding to this horizontal width is denoted as... The formula for calculating the conversion coefficient k between pixels and physical distance is: Where k is the actual physical distance corresponding to a single pixel, in millimeters per pixel, and 250 is the number of millimeters corresponding to the standard horizontal width of a simulated human chest. This represents the number of pixels corresponding to that width. Using this conversion coefficient, the calculated normal displacement in pixels within the image can be converted into actual physical distance, thereby determining the corresponding numerical equivalent of the actual pressing depth. The effective amplitude threshold corresponding to a 5mm actual depth is converted to pixel equivalents using the formula 5 divided by k. The pixel equivalent range corresponding to a pressing depth of 5cm to 6cm is 50 divided by k to 60 divided by k. This calibration method is used because it can adaptively eliminate size differences caused by camera installation distance and image scaling, ensuring a consistent pressing depth determination standard across different deployment scenarios.

[0018] Alternative implementation methods: In addition to the combined architecture of YOLOv8n lightweight object detection network and MobileNetV2 lightweight human pose estimation network, lightweight visual recognition models can also be implemented using single lightweight neural network models with key point output capabilities, such as YOLOv5n and YOLOv10-n. The model input size is set to 640×640, and the training parameters are consistent with the aforementioned combined architecture. After training, the same accuracy of localization point and contour boundary extraction can be achieved.

[0019] In embodiments of the present invention, the optical flow field between adjacent image frames is calculated, and the normal displacement of the pixel is obtained by combining the normal projection direction. A center apparent retrieval sequence and a peripheral entanglement retrieval sequence are constructed based on the normal displacement in the central region and peripheral region, respectively, including: Calculate the dense optical flow field between adjacent practical video image frames; The dense optical flow field is projected and accumulated along the normal projection direction in front of the chest to obtain the normal displacement of the pixel. The median normal displacement of pixels in the central and peripheral regions is taken to obtain the apparent repositioning sequence in the center and the entangled repositioning sequence in the peripheral region. Specifically, based on the chest-front normal projection direction and the division results of the central and peripheral regions, dense optical flow calculations are performed on adjacent practical video image frames, with adjacent frames denoted as the (t-1)th frame. With the image of frame t t is a positive integer greater than or equal to 2, and the initial frame t=1 corresponds to the start time of the pressing action. The Farneback dense optical flow algorithm is used to calculate the dense optical flow field between two frames. The number of pyramid layers in the algorithm is set to 3, the pyramid scaling ratio is set to 0.5, the number of iterations per pyramid level is set to 10, the optical flow calculation window size is set to 15×15 pixels, the neighborhood polynomial expansion number is set to 5, and the Gaussian smoothing standard deviation of the polynomial expansion is set to 1.1. The reason for using this algorithm and parameter settings is that the Farneback algorithm can achieve dense optical flow calculation of the entire pixel based on polynomial expansion without feature point matching, which is suitable for the lightweight deployment requirements of this invention. The above parameter settings can ensure the accuracy of optical flow calculation while taking into account real-time processing speed, and effectively suppress the optical flow mismatch problem caused by palm occlusion and slight screen shaking during the pressing process. In the calculated dense optical flow field, the image frame The optical flow vector corresponding to any pixel p within the matrix is ​​denoted as Optical flow vector It is a two-dimensional planar vector that represents the planar displacement change of pixel p from frame (t-1) to frame t.

[0020] The dense optical flow field is projected along the chest-facing normal projection direction and accumulated to obtain the normal displacement of the pixel. The formula for calculating the normal displacement is: in Let p be the normal displacement of pixel p in frame t-1, and let t be the normal displacement of all pixels at the initial time t=1. All are set to 0. Optical flow vector Direction of normal projection of the chest dot product operation, The unit normal vector is used. The reason for using this projection and cumulative calculation is that the dot product operation can project the optical flow displacement in the two-dimensional plane onto the normal dimension consistent with the depth direction of the pressing action, filtering out the interference of planar lateral and longitudinal displacements that are unrelated to the pressing and rebound, and retaining only the normal displacement component that can characterize the undulation changes in the chest area. The cumulative calculation can convert the inter-frame displacement of each frame into the total displacement from the moment the pressing begins, completely characterizing the continuous depth change process of the chest area during the pressing process, and accurately reflecting the full-cycle displacement change of the pressing compression and release rebound.

[0021] By taking the median of the normal displacement of pixels in the central and peripheral regions respectively, the apparent retracement sequence of the center and the entanglement retracement sequence of the periphery are obtained. The value of the t-th frame of the apparent retracement sequence of the center is... The calculation formula is The value of the t-th frame of the peripheral entanglement repositioning sequence The calculation formula is Where median is the median calculation operator. The central region is defined in the t-th frame of the image. The median is defined as the peripheral region in the t-th frame of the image. The median is used instead of the mean for calculation because it effectively resists interference from outliers such as hand occlusion during pressing, local pixel optical flow mismatch, and image noise. This avoids the displacement deviation of individual abnormal pixels affecting the statistical results of the entire region, ensuring the stability and robustness of the retrieval sequence. The final apparent retrieval sequence is... This is a one-dimensional temporal sequence that varies continuously with frame number, used to characterize the apparent recovery process of the pectoral skin surface in the central chest region during compression and release, and the peripheral entrapment sequence. To and One-dimensional time series of the same length are used to characterize the hysteretic recovery process of the anterior peripheral region caused by the involvement of internal compression structures.

[0022] It should be noted that the central apparent repositioning sequence refers to a sequence formed chronologically by summarizing the displacement changes of each pixel within the central area of ​​the chest compression region during a single compression-release operation, along the chest normal projection direction. This sequence characterizes the gradual recovery of the chest's central surface morphology after compression and release. It reflects the surface contour recovery trend at the compression center and can describe the speed and degree of apparent repositioning of the chest skin during the release phase, but it primarily corresponds to the recovery state of the outer visible area.

[0023] It should be noted that the peripheral traction and return sequence refers to a sequence formed chronologically by summarizing the displacement changes of each pixel within the peripheral area surrounding the chest compression zone during a single compression and release operation, along the chest normal projection direction. This sequence characterizes the delayed recovery process of the chest periphery under the traction of the internal compression structure after the center of the compression is released. This sequence reflects the traction and retraction trend of the chest periphery caused by the return of the internal support structure, demonstrating the recovery process as effective internal return is transmitted to the outer region. Therefore, it can be used to distinguish the asynchronous rebound state where the central chest surface recovers first while the periphery has not yet recovered synchronously.

[0024] Optional implementation methods: In addition to the Farneback dense optical flow algorithm, the local block matching displacement estimation algorithm can also be used to calculate pixel displacement. The block matching window is set to 16×16 pixels, the search range is set to 32×32 pixels, and the matching criterion is the absolute error criterion. After obtaining the planar displacement vector of the pixel through block matching, the subsequent normal projection and cumulative calculation are then performed to achieve the same effect of extracting the normal displacement.

[0025] In an embodiment of the present invention, the release interval of each compression release cycle is determined based on the central apparent return sequence. Within the release interval, time-series analysis is performed on the central apparent return sequence and the peripheral entanglement return sequence to extract the intensity of the asynchronous rebound characteristic and the peripheral terminal return progress, including: Extreme value search is performed on the central apparent return sequence to determine the time of the lowest compression point and the time of the end of the central apparent release in each compression-release cycle; The time interval corresponding to the moment from the lowest point of compression to the end of the apparent release in each compression-release cycle is defined as the release interval. The central apparent retracement sequence and the peripheral entanglement retracement sequence are normalized based on the maximum and minimum values ​​of the central apparent retracement sequence and the peripheral entanglement retracement sequence within the release interval, respectively. Calculate the centroid time of the positive change of the normalized central apparent homing sequence and the peripheral entanglement homing sequence, respectively; The intensity of asynchronous rebound characteristics in the table is calculated; The normalized peripheral entanglement retrieval sequence value at the end of the central apparent release is determined as the peripheral final state retrieval progress. Specifically, based on the center apparent homing sequence synchronized with the actual video frame sequence. and peripheral entanglement repositioning sequence The system automatically identifies the start and end times of the pressing action. The identification logic is as follows: A sliding window variance is calculated on the central apparent return sequence. The sliding window size is set to 10 frames. When the variance of three consecutive windows exceeds a preset action start threshold, the starting frame of that window is determined to be the start time of the pressing action, corresponding to an initial frame t=1. When the variance of 30 consecutive frames is less than a preset action end threshold, that frame is determined to be the end time of the pressing action, and subsequent period division and feature calculation are stopped. The action start threshold and action end threshold are determined using the 1 mm displacement equivalent calibrated in step 1. This identification logic is used because it can automatically identify the start and end intervals of the examinee's pressing action, filtering out invalid images from pre-exam preparation and post-exam cleanup, ensuring the accuracy of the assessment interval, and achieving intelligent assessment without human intervention throughout the entire process.

[0026] A sliding window local extremum search is performed on the apparent return sequence of the chest compressions within the start and end intervals to determine the critical time points of each compression-release cycle. The sliding window size is set to 20 frames, and the sliding step size is set to 1 frame. This window size is chosen because the standard compression rhythm in CPR practice is 100 to 120 compressions per minute, corresponding to 15 to 18 frames per compression cycle. The 20-frame window size can completely cover a single effective compression cycle, avoiding missed or duplicate extreme value detections, while also suppressing spurious extreme value interference from high-frequency noise in the sequence. During the sliding window traversal, local minima and local maxima are identified within the window. Local minima correspond to the moment when the chest region is compressed to its deepest position during compression, and local maxima correspond to the moment when the apparent return of the central chest region to its highest position after compression release. At the same time, an effective press amplitude threshold is set. Only when the difference between the sequence values ​​of adjacent local maxima and local minima is greater than the preset effective amplitude threshold, the set of extreme points is determined to correspond to an effective press action. The effective amplitude threshold is set to the equivalent of the sequence value of 5 mm of actual press depth. The reason for setting this threshold is that it can filter out invalid small amplitude fluctuations in the sequence caused by screen shaking and slight movements, and only retain the extreme points corresponding to the effective press actions that meet the practical requirements, so as to avoid misjudgment of invalid cycles.

[0027] After identifying local extreme points, an invalid extreme point filtering operation is performed. The filtering rules are as follows: the frame interval between adjacent local minimum and local maximum points must not be less than 5 frames, and the frame interval between two adjacent sets of valid extreme points must not be less than 10 frames, corresponding to the minimum number of frames per cycle of the lowest compression rhythm in CPR practice. Extreme points that do not meet these interval requirements are judged as false extreme points and are discarded. The reason for adopting this filtering rule is that it can effectively suppress false extreme points caused by high-frequency noise in the sequence and slight hand shaking, ensuring that each compression cycle corresponds to only one set of valid extreme points, avoiding the problems of repeated detection and missed detection of compression cycles, and ensuring the accuracy of cycle division.

[0028] After sorting all valid extreme points in ascending order of frame number, the frame number of the local minimum point corresponding to the kth valid compression cycle is determined as the time of the compression minimum point. The frame number of the first valid local maximum immediately following the local minimum is determined as the end time of the central apparent release. k is a positive integer representing the compression cycle number. The time interval from the lowest point of compression to the end of the apparent release within each compression-release cycle is defined as the release interval. Release range The expression is in This is the frame number of the lowest compression point in the k-th press-release cycle. This is the sequence number of the center apparent release end frame of the k-th press-release cycle. The reason for using this interval as the rebound analysis interval is that it completely covers the entire release and rebound process from the deepest press-compression position to the complete return of the center apparent position. This can completely eliminate the interference of displacement changes during the press-compression stage on the rebound timing analysis and ensure the accuracy of subsequent asynchronous rebound feature extraction.

[0029] Based on the maximum and minimum values ​​of the central apparent retracement sequence and the peripheral entrapment retracement sequence within the release interval, the two sets of sequences are normalized to obtain the normalized central apparent retracement sequence and the peripheral entrapment retracement sequence. The calculation formula is in This represents the original value of the center apparent retracement sequence corresponding to frame t. The apparent reticular sequence centered in the release interval Minimum value within, The apparent reticular sequence centered in the release interval The maximum value within. The normalized periphery entanglement repositioning sequence. The calculation formula is in This represents the original value of the peripheral entanglement retrieval sequence corresponding to frame t. For the peripheral entanglement retracement sequence in the release interval Minimum value within, For the peripheral entanglement retracement sequence in the release interval The maximum value within the range. The reason for adopting this normalization calculation method is that it can uniformly map the rebound sequences with different amplitudes within different pressing cycles to a numerical range of 0 to 1, eliminating the influence of differences in pressing depth and screen scaling ratios in different pressing cycles on the rebound timing analysis. This provides a unified comparison benchmark for the rebound processes in different cycles and regions, ensuring the accuracy and comparability of subsequent calculations of the positive change center of gravity and asynchronous rebound feature extraction. The normalized sequence value 0 corresponds to the lowest compression point at the start of rebound, and the value 1 corresponds to the fully returned state at the end of rebound, which can intuitively represent the completion ratio of the rebound process.

[0030] Release interval based on the k-th press-release cycle Normalized central apparent reticular sequence Regression sequence with peripheral involvement The centroid time of positive change in the two sets of sequences is calculated separately. Positive change refers to the portion of the numerical increment between adjacent frames in the sequence that is greater than 0, corresponding to the effective positive recovery action of the chest skin region during the rebound process. First, the positive value operator is defined as max(z,0), which is used to filter out the negative numerical changes between adjacent frames in the sequence, retaining only the effective positive recovery increment during the rebound process, and avoiding interference from small drops during the rebound process and sequence acquisition noise on the calculation of centroid time.

[0031] The positive change centroid time of the apparent homogeneous sequence of the computation center The calculation formula is: Where t is the frame number within the release interval. Let be the normalized center apparent retracement sequence value of frame t. The normalized center apparent retracement sequence value of the (t-1)th frame. The positive increment of the apparent recovery sequence is the sum of the products of each frame number and its corresponding positive increment in the numerator, and the sum of all positive increments within the release interval. This formula is used to calculate the center of gravity time of the positive change because it uses the positive recovery increment of each frame during the rebound process as a weight, performing a weighted average of the frame numbers. The resulting center of gravity time accurately represents the main occurrence time of the rebound action in the central region. Compared to simple peak or half-value times, it more comprehensively reflects the temporal distribution characteristics of the entire rebound process, effectively capturing the differences in the speed of the rebound process and avoiding calculation errors caused by numerical fluctuations at a single moment.

[0032] Similarly, calculate the positive change centroid time of the peripheral entanglement repositioning sequence. The calculation formula is: in Let be the normalized periphery entanglement repositioning sequence value of frame t. The normalized periphery entanglement repositioning sequence value for frame t-1. The formula represents the positive increment of adjacent frames in the peripheral entanglement retrieval sequence, and the remaining parameters are completely consistent with the formula for calculating the centroid time of the positive change in the central apparent retrieval sequence. This calculation can accurately characterize the main occurrence time of the rebound action in the peripheral region, providing a unified benchmark for subsequent calculations of rebound timing differences.

[0033] Based on the difference between the positive change center time of the peripheral entanglement repositioning sequence and the positive change center time of the central apparent repositioning sequence, combined with the cumulative positive difference between the central apparent repositioning sequence and the peripheral entanglement repositioning sequence within the release interval, the intensity of the asynchronous rebound characteristic is calculated. The calculation formula is: in This represents the moment of the positive change center of gravity in the peripheral entanglement sequence. The moment of the positive change centroid of the apparent homing sequence is the time when the homing sequence is at its center. This represents the time lag between the peripheral rebound and the apparent rebound at the center. The total number of frames in the released interval. The normalization coefficient is the interval length. This represents the positive difference between the center apparent rebound sequence value and the peripheral entanglement rebound sequence value within the same frame. The summation term is the cumulative total of all positive differences within the release interval. The temporal lag directly characterizes the degree of lag in peripheral rebound over time, while the cumulative positive difference characterizes the magnitude by which the center apparent rebound progress consistently leads the peripheral rebound throughout the entire release interval. Multiplying these two values ​​comprehensively quantifies the severity of asynchronous rebound from both the temporal lag and progress difference perspectives. The interval length normalization coefficient eliminates the influence of differences in release interval lengths across different press cycles on the feature values, ensuring uniform comparability of feature values ​​across different press cycles. The calculated... The larger the value, the earlier the central apparent repositioning occurs compared to the peripheral repositioning, and the more obvious the asynchronous rebound phenomenon induced by the remaining loose chest skin is.

[0034] The normalized peripheral entanglement retrieval sequence value at the end of the central apparent release is determined as the peripheral final state retrieval progress. The calculation formula is: in This represents the center apparent release end time of the k-th press-release cycle. This represents the normalized peripheral entanglement recovery sequence value corresponding to the end of the central apparent release. The reason for using this value as the peripheral final-state recovery progress is that it directly characterizes the proportion of actual recovery progress completed in the peripheral region when the central apparent rebound reaches a fully recovered state. It intuitively reflects the final-state difference between the central apparent recovery and the peripheral entanglement recovery, providing a direct quantitative basis for subsequent rebound integrity correction.

[0035] It should be noted that the asynchronous rebound sign intensity refers to a value used to characterize the degree of synchronicity between the recovery process of the central superficial layer of the chest compression area and the recovery process of the peripheral layer within the same compression-release cycle. It reflects the temporal misalignment and recovery difference caused by the visible outline of the outer chest skin recovering first, followed by the recovery of the peripheral area driven by the internal compression structures. A larger value indicates that the central apparent recovery is completed earlier than the peripheral traction recovery, and the inconsistency between the outer and internal effective recovery is more pronounced. This indicates a stronger influence of the remaining chest skin relaxation on the appearance of rebound, and a greater likelihood of a situation where the chest surface appears to have completed rebound while the internal effective recovery is not yet fully completed. Therefore, this value can be used to characterize the degree of deviation between apparent recovery and the true rebound state.

[0036] It should be noted that the peripheral recovery progress refers to the degree of recovery completed by the peripheral area along the normal projection direction of the chest when the central chest area reaches the apparent end of the release during a single compression-release cycle. This value characterizes the state of peripheral traction recovery at that moment. This value reflects whether the peripheral area has synchronously completed the recovery process transmitted from the internal compression structure at the moment when the central surface appears to have completed recovery. A higher value indicates that the peripheral area has essentially completed recovery at the end of the apparent release in the center, with relatively consistent surface and internal rebound. A lower value indicates that the peripheral area still exhibits significant lag at that moment, and the effective internal rebound has not been fully transmitted to the peripheral area, thus indicating a strong asynchronous surface and internal rebound phenomenon in the current compression-release cycle.

[0037] In embodiments of the present invention, a rebound integrity coefficient is calculated based on the asynchronous rebound intensity and the peripheral final-state return progress, and an effective press value for a single press-release cycle is generated by combining the effective displacement of the central apparent return sequence and the peripheral entanglement return sequence, including: The rebound integrity coefficient is calculated based on the asynchronous rebound intensity and the peripheral final state return progress. Based on the change in the central apparent retracement sequence and the change in the peripheral entrapment retracement sequence within the release interval, the smaller of the two values ​​is taken as the effective displacement supported by both. Multiply the effective displacement by the rebound integrity coefficient to obtain the effective pressure value for a single press-release cycle; Specifically, the rebound integrity coefficient is calculated based on the asynchronous rebound intensity and the peripheral final state recovery progress. The calculation formula is: in This is the rebound integrity coefficient for the k-th press-release cycle, with a value ranging from 0 to 1. The asynchronous rebound intensity of the table during the k-th press-release cycle. Let be the peripheral final return progress during the k-th press-release cycle. The reason for using this formula to construct the rebound integrity coefficient is that it achieves adaptive dynamic adjustment of the rebound judgment weights; when there is no asynchronous rebound phenomenon, The value is 0, at which point the springback integrity coefficient is... The value is 1, and the system uses the apparent return state of the center as the criterion for judging the integrity of the rebound. This conforms to the normal rebound judgment logic when there is no relaxation margin. The more obvious the asynchronous rebound phenomenon between the table and the interior, the more likely it is to occur. The larger the value, the more automatically the formula will improve the periphery final state homing progress. The weighting of the rebound integrity assessment weakens the sole role of the apparent center return, avoiding misjudging false center return as complete rebound. At the same time, the formula is a monotonically continuous function, so there will be no sudden changes in weighting, ensuring that the rebound integrity coefficient smoothly transitions with the change in the degree of asynchronous rebound, avoiding jumps in the scoring results, and ensuring the stability and rationality of the evaluation results.

[0038] Based on the changes in the central apparent retracement sequence and the changes in the peripheral entrapment retracement sequence within the release interval, the smaller of the two values ​​is taken as the effective displacement amount supported by both. The calculation formula is: in This represents the value of the central apparent homing sequence corresponding to the end of the central apparent release. To compress the central apparent homing sequence values ​​corresponding to the lowest point, To release the total change in the apparent return sequence within the interval, corresponding to the total displacement of the press-and-rebound observed in the central region, The value of the peripheral entanglement retracement sequence corresponding to the end of the central apparent release. To compress the peripheral entanglement sequence values ​​corresponding to the lowest point, To release the total change in the peripheral traction and rebound sequence within the interval, the minimum value operator is used to determine the total displacement observed in the peripheral region during compression. The reason for using the minimum of both is that the effective depth of the compression action must be supported by observations from both the central and peripheral regions. When asynchronous rebound occurs, the rebound displacement in the peripheral region will be less than the apparent displacement in the central region. Minimizing this value filters out false apparent displacement in the central region caused by chest skin laxity, retaining only the effective component that reflects the true displacement of both the chest skin surface and the internal compression structure. This avoids a systematic overestimation of the compression depth and ensures that the effective displacement accurately reflects the actual effective stroke of the compression action.

[0039] Multiplying the effective displacement by the rebound integrity coefficient yields the effective pressure value for a single press-release cycle. The calculation formula is: in This represents the effective pressure value during the k-th press-release cycle. This refers to the effective displacement supported by all parties within this period. This is the rebound integrity coefficient for that cycle. The reason for using this method to calculate the effective compression value is that the effective compression value simultaneously integrates two core evaluation dimensions: the true effectiveness of the compression displacement and the integrity of the rebound action. The effective displacement ensures the authenticity and reliability of the compression depth, while the rebound integrity coefficient ensures the accuracy of the judgment of the sufficiency of the rebound. Multiplying the two together can simultaneously incorporate two types of unqualified operations, insufficient compression depth and insufficient rebound, into the judgment system of effective compression. Only when the compression displacement is truly effective and the rebound action is fully sufficient can the full effective compression value be obtained, which perfectly matches the evaluation requirements for effective compression in the CPR practical examination.

[0040] It should be noted that the rebound integrity coefficient refers to a value used to characterize whether the chest compression area has achieved sufficient return to its original position when recovering from the compressed state to the released state during a single compression and release process. It comprehensively reflects the consistency between the apparent recovery of the central chest surface and the traction recovery of the peripheral chest area. This coefficient is not determined solely by whether the central chest area has recovered visually, but rather by correcting for the rebound state based on the asynchronous rebound phenomenon. Therefore, it can indicate the degree of consistency between the visible return of the outer chest layer and the effective return of the inner chest layer during the current compression and release process. A larger coefficient indicates that at the end of the apparent central release, the peripheral area has also synchronously completed a high degree of recovery, resulting in a more complete chest rebound. A smaller coefficient indicates that although the central chest area may have shown apparent return, the peripheral area still lags behind, and the effective return of the inner chest layer is insufficient, thus indicating poor integrity of the current rebound process.

[0041] It should be noted that the effective compression value of a single compression-release cycle refers to a value obtained by comprehensively considering the effective displacement and rebound integrity of the chest compression area during a complete compression and release process. This value characterizes whether the compression truly constitutes an effective compression contribution worthy of evaluation. This value does not simply represent the magnitude of surface displacement during compression, but rather is based on the displacement changes supported by both the central and peripheral chest areas, and then corrected for by a rebound integrity coefficient. Therefore, it reflects both the actual compression effect produced by the compression action and whether sufficient rebound was achieved during the release phase. A larger value indicates that the compression has both good compression displacement and high rebound integrity, truly reflecting a high-quality effective compression; a smaller value indicates insufficient displacement formation or rebound recovery, and should not be considered a sufficiently effective compression action.

[0042] In embodiments of the present invention, a final scoring result is generated based on the rebound integrity coefficient and effective pressure value of each press-release cycle, combined with the press rhythm, including: The entire press-release cycle during the statistical operation was analyzed, and the average asynchronous rebound intensity, average rebound integrity coefficient, and average effective press value were calculated throughout the entire process. Calculate the compression rhythm frequency based on the total number of all compression-release cycles and their start and end times; The average asynchronous rebound intensity, average rebound integrity coefficient, average effective compression value, and compression rhythm frequency throughout the entire process are input into the evaluation rule engine, and the final score result including the total score is output. Specifically, the total number of effective compression-release cycles during the practical exam is recorded, denoted as N, where N is a positive integer, corresponding to the total number of effective compression actions completed by the examinee in this practical exam. The average asynchronous rebound intensity, average rebound integrity coefficient, and average effective compression value are calculated. The formula for calculating the average asynchronous rebound intensity is as follows: in The average asynchronous rebound intensity throughout the entire process, where N is the total number of effective press-release cycles. Let be the intensity of the asynchronous rebound sign in the k-th compression-release cycle, and the summation term is the sum of the asynchronous rebound sign intensities in all cycles. Using an arithmetic mean to calculate the overall average can equally reflect the asynchronous rebound of each compression action, comprehensively characterizing the overall level of risk of rebound misjudgment due to chest skin laxity margin throughout the entire practical operation, avoiding excessive influence of a single abnormal cycle on the overall assessment, and providing a unified global quantitative indicator for correcting the overall evaluation results.

[0043] The formula for calculating the average rebound integrity coefficient throughout the entire process is as follows: in The average rebound integrity coefficient over the entire process. This represents the rebound integrity coefficient for the k-th compression-release cycle. The remaining parameters are consistent with the parameter definitions in the asynchronous rebound strength calculation formula in the aforementioned overall average table. The reason for using the arithmetic mean is that this method comprehensively reflects the overall level of rebound adequacy of all compression actions performed by the examinee throughout the entire practical exercise. The closer this value is to 1, the better the overall rebound integrity, which better meets the assessment requirements of CPR practice. Simultaneously, it directly corresponds to the overall proportion of insufficient rebound operations, providing a direct quantitative basis for score deductions.

[0044] The formula for calculating the average effective pressure value over the entire process is as follows: in This represents the average effective pressure value throughout the entire compression process. This represents the effective compression value during the k-th compression-release cycle, with the remaining parameters remaining consistent with the definitions in the aforementioned formula. The average value comprehensively characterizes the overall effectiveness of the candidate's compression movements throughout the entire practical exercise. It integrates the two core assessment dimensions of the authenticity of compression depth and the completeness of the rebound action, perfectly aligning with the core assessment requirements for effective compression in the CPR practical exam, and directly reflecting the overall quality of the candidate's compression operation.

[0045] The compression rhythm frequency is calculated based on the total number of compression-release cycles and their start and end times. The formula for calculating the compression rhythm frequency is as follows: in The compression rhythm frequency is expressed in compressions per minute, where N is the total number of effective compression-release cycles. In this embodiment, the frame rate for capturing the practical video is set to 30 frames per second. This is the frame number corresponding to the center apparent release end time of the Nth press-release cycle. This is the frame number corresponding to the lowest compression point of the first press-release cycle. reduce This represents the total number of frames corresponding to the entire effective compression sequence. This calculation method starts at the lowest point of the initial compression and ends at the release point of the final compression, fully covering the entire process of the candidate's continuous compressions. It accurately calculates the number of compressions per minute that meets the CPR assessment standards, avoiding rhythm calculation errors caused by invalid initial or final compressions or pauses. Furthermore, it aligns with industry-standard compression rhythm assessment calculation methods, ensuring the compliance and universality of the assessment results.

[0046] The evaluation rules engine inputs the average intensity of the asynchronous rebound sign, the average rebound integrity coefficient, the average effective compression value, and the compression rhythm frequency throughout the entire compression process. The engine outputs a final score including the total score. Based on the CPR and cardiovascular emergency care guidelines and the unified evaluation standards for domestic emergency practice examinations, the evaluation rules engine sets weights and passing thresholds for four evaluation dimensions. The average effective compression value throughout the entire compression process accounts for 40% of the weight, with a passing threshold of a compression depth of 5 to 6 centimeters. The average rebound integrity coefficient throughout the entire compression process accounts for 30% of the weight, with a passing threshold of ≥0.9. The compression rhythm frequency accounts for 20% of the weight, with a passing range of 100 to 120 compressions per minute. The average intensity of the asynchronous rebound sign throughout the entire compression process accounts for 10% of the weight, used for additional deductions to correct abnormal rebound conditions.

[0047] The specific deduction rules are set as follows: A perfect score is achieved if the actual depth corresponding to the average effective compression value throughout the entire compression is between 5 cm and 6 cm. For compressions below 5 cm, 20% of the total score for that dimension is deducted for every 0.5 cm below 5 cm; for compressions above 6 cm, 10% of the total score for that dimension is deducted for every 0.5 cm above 6 cm. A perfect score is achieved if the average rebound integrity coefficient throughout the entire compression is greater than or equal to 0.9. For compressions below 0.9, 15% of the total score for that dimension is deducted for every 0.05 below 0.9. A perfect score is achieved if the compression rhythm frequency is between 100 and 120 compressions per minute. For compressions below 100 or above 120, 20% of the total score for that dimension is deducted for every 10 compressions exceeding this range. If the average asynchronous rebound intensity throughout the entire compression exceeds a preset abnormal threshold, an additional 5% of the total score is deducted for every 0.5 units exceeding the threshold, up to a maximum deduction of 10%. The rules engine outputs a total score of 100 points, and simultaneously outputs the deduction items, deduction scores, and corresponding abnormal cycle details for each dimension, as well as the comprehensive judgment result of rebound abnormality, completing the full-process intelligent assessment of this cardiopulmonary resuscitation practical examination.

[0048] like Figure 2 The diagram shown is a flowchart of an intelligent assessment method for emergency practical examinations based on lightweight visual recognition, provided by an embodiment of the present invention.

[0049] In this embodiment, the steps of the intelligent assessment method for emergency practical examinations based on lightweight visual recognition are as follows: S1. Acquire the practical video captured by a single-channel ordinary camera device, extract the positioning points and contour boundaries from the image frames of the practical video using a lightweight visual recognition model, and divide the pressing area into a central area and a peripheral area. S2. Calculate the optical flow field between adjacent image frames, and obtain the normal displacement of the pixel by combining the normal projection direction. Construct the center apparent repositioning sequence and the peripheral entanglement repositioning sequence based on the normal displacement in the center area and the peripheral area, respectively. S3. Determine the release interval of each compression release cycle based on the central apparent return sequence. Perform time series analysis on the central apparent return sequence and the peripheral entanglement return sequence within the release interval to extract the intensity of the asynchronous rebound sign and the peripheral final state return progress. S4. Calculate the rebound integrity coefficient based on the asynchronous rebound intensity and the peripheral final state return progress, and generate the effective pressure value of a single press release cycle by combining the effective displacement of the center apparent return sequence and the peripheral entangled return sequence. S5. Based on the rebound integrity coefficient and effective pressure value of each press-release cycle, and combined with the press rhythm, generate the final score result.

[0050] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An intelligent assessment system for emergency practical examinations based on lightweight visual recognition, characterized in that: include: The region segmentation module is used to acquire the practical video captured by a single-channel ordinary camera device, extract positioning points and contour boundaries from the image frames of the practical video through a lightweight visual recognition model, and divide the pressing area into a central area and a peripheral area. The displacement construction module is used to calculate the optical flow field between adjacent image frames, and obtain the normal displacement of the pixel by combining the normal projection direction. Based on the normal displacement in the central area and the peripheral area, the center apparent retrieval sequence and the peripheral entanglement retrieval sequence are constructed respectively. The intensity extraction module is used to determine the release interval of each compression release cycle based on the central apparent return sequence. Within the release interval, the module performs time series analysis on the central apparent return sequence and the peripheral entanglement return sequence to extract the intensity of the asynchronous rebound sign and the peripheral terminal return progress. The press correction module is used to calculate the rebound integrity coefficient based on the asynchronous rebound intensity and the peripheral final state return progress, and to generate the effective press value of a single press release cycle by combining the effective displacement of the center apparent return sequence and the peripheral entangled return sequence. The intelligent evaluation module is used to generate the final score result based on the rebound integrity coefficient and effective pressure value of each press-release cycle, combined with the press rhythm.

2. The intelligent assessment system for emergency practical examinations based on lightweight visual recognition as described in claim 1, characterized in that, A lightweight visual recognition model is used to extract positioning points and contour boundaries from image frames of the practical video, and the pressing area is divided into a central area and a peripheral area, including: Extract the left and right shoulder positioning points, the center point of the pressed palm, and the outer contour boundary of the chest from the image frames of the practical video; The horizontal reference vector in front of the chest is determined based on the positioning points of the left and right shoulders, and the normal projection direction in front of the chest is constructed based on the horizontal reference vector in front of the chest. Calculate the minimum distance from the center point of the pressing palm to the outer contour boundary of the chest, and use it as the current pressing action scale; Using the center point of the pressing palm as the center, divide the chest pressing area into a central area and a peripheral area according to the current pressing force.

3. The intelligent assessment system for emergency practical examinations based on lightweight visual recognition as described in claim 2, characterized in that, Construct the central apparent reticular sequence and the peripheral entanglement reticular sequence, including: Calculate the dense optical flow field between adjacent practical video image frames; The dense optical flow field is projected and accumulated along the normal projection direction in front of the chest to obtain the normal displacement of the pixel. The median of the normal displacement of pixels in the central and peripheral regions is taken to obtain the apparent repositioning sequence of the center and the entanglement repositioning sequence of the periphery.

4. The intelligent assessment system for emergency practical examinations based on lightweight visual recognition as described in claim 3, characterized in that, The release interval for each compression cycle is determined based on the central apparent return sequence, including: Extreme value search is performed on the central apparent return sequence to determine the time of the lowest compression point and the time of the end of the central apparent release in each compression-release cycle; The time interval corresponding to the moment from the lowest point of compression to the end of the apparent release in each compression-release cycle is defined as the release interval. The central apparent reticular sequence and the peripheral entanglement reticular sequence are normalized based on the maximum and minimum values ​​of the central apparent reticular sequence and the peripheral entanglement reticular sequence within the release interval, respectively.

5. The intelligent assessment system for emergency practical examinations based on lightweight visual recognition according to claim 4, characterized in that, Extract the asynchronous rebound intensity and peripheral terminal state recovery progress from the table, including: Calculate the centroid time of the positive change of the normalized central apparent homing sequence and the peripheral entanglement homing sequence, respectively; The intensity of the asynchronous rebound pattern is calculated based on the difference between the positive change center time of the peripheral entanglement repositioning sequence and the positive change center time of the central apparent repositioning sequence, combined with the cumulative positive difference between the central apparent repositioning sequence and the peripheral entanglement repositioning sequence within the release interval. The normalized peripheral entanglement retrieval sequence value at the end of the central apparent release is determined as the peripheral final state retrieval progress.

6. The intelligent assessment system for emergency practical examinations based on lightweight visual recognition as described in claim 5, characterized in that, The formula for calculating the springback integrity coefficient is as follows: In the formula, For the first The rebound integrity coefficient per press-release cycle. For the first The intensity of asynchronous rebound characteristics within each press-release cycle. For the first The peripheral final state return progress of each press-release cycle.

7. The intelligent assessment system for emergency practical examinations based on lightweight visual recognition as described in claim 6, characterized in that, Generate the effective pressure value for a single press-release cycle, including: Based on the change in the central apparent retracement sequence and the change in the peripheral entrapment retracement sequence within the release interval, the smaller of the two values ​​is taken as the effective displacement supported by both. Multiply the effective displacement by the rebound integrity coefficient to obtain the effective pressure value for a single press-release cycle.

8. The intelligent assessment system for emergency practical examinations based on lightweight visual recognition according to claim 7, characterized in that, The final score is generated based on the rebound integrity coefficient and effective compression value of each compression-release cycle, combined with the compression rhythm, including: The entire press-release cycle during the statistical operation was analyzed, and the average asynchronous rebound intensity, average rebound integrity coefficient, and average effective press value were calculated throughout the entire process. Calculate the compression rhythm frequency based on the total number of all compression-release cycles and their start and end times; The evaluation rules engine is input with the average asynchronous rebound intensity, average rebound integrity coefficient, average effective compression value, and compression rhythm frequency throughout the entire process. The final score result, including the total score, is output.

9. An intelligent assessment method for emergency practical examinations based on lightweight visual recognition, applied to the intelligent assessment system for emergency practical examinations based on lightweight visual recognition as described in any one of claims 1-8, characterized in that, The method includes: S1. Acquire the practical video captured by a single-channel ordinary camera device, extract the positioning points and contour boundaries from the image frames of the practical video using a lightweight visual recognition model, and divide the pressing area into a central area and a peripheral area. S2. Calculate the optical flow field between adjacent image frames, and obtain the normal displacement of the pixel by combining the normal projection direction. Construct the center apparent repositioning sequence and the peripheral entanglement repositioning sequence based on the normal displacement in the center area and the peripheral area, respectively. S3. Determine the release interval of each compression release cycle based on the central apparent return sequence. Perform time series analysis on the central apparent return sequence and the peripheral entanglement return sequence within the release interval to extract the intensity of the asynchronous rebound sign and the peripheral final state return progress. S4. Calculate the rebound integrity coefficient based on the asynchronous rebound intensity and the peripheral final state return progress, and generate the effective pressure value of a single press release cycle by combining the effective displacement of the center apparent return sequence and the peripheral entangled return sequence. S5. Based on the rebound integrity coefficient and effective pressure value of each press-release cycle, and combined with the press rhythm, generate the final score result.

Images