Cardiopulmonary resuscitation pressing action posture feature ai recognition and deviation correction method

Abstract

The application provides a cardiopulmonary resuscitation pressing action posture feature AI identification and deviation correction method, relates to the technical field of medical first aid, and comprises the following steps: acquiring trunk movement trajectory data and chest cavity pressure dynamic data when a rescuer performs a pressing action; reconstructing the trunk movement trajectory data in space, extracting trunk centroid movement law and angular velocity change trend, calculating posture stability component, and generating a steady-state support area; constructing a pressure response state space for the chest cavity pressure dynamic data, establishing a pressure elastic deformation field, and generating an optimal pressure application path; mapping the steady-state support area and the optimal pressure application path to an elastic potential energy field, optimizing the support structure based on dynamic balance constraints, and generating a real-time posture adjustment instruction sequence; and adjusting the standing position distance, support angle and force application direction of the rescuer according to the instruction sequence to complete deviation correction. The application can accurately identify pressing action deviation in real time, intelligently generate individualized correction instructions, and improve the success rate of cardiopulmonary resuscitation.

Description

Technical Field

[0001] This invention relates to the field of medical emergency technology, and in particular to an AI-based method for recognizing and correcting deviations in the posture features of cardiopulmonary resuscitation (CPR) compression movements. Background Technology

[0002] The quality of chest compressions during cardiopulmonary resuscitation (CPR) directly impacts patient survival and neurological recovery. Because compressions involve complex biomechanical mechanisms, rescuers often struggle to accurately master the correct posture and force application. With advancements in artificial intelligence and sensor technologies, real-time monitoring and guidance of compression movements using intelligent recognition methods has become a crucial research direction for improving CPR quality. Traditional training methods rely primarily on manual demonstration and feedback, lacking quantitative analysis of movement details and hindering personalized, precise guidance. While some sensor-based monitoring devices have emerged in recent years, technological bottlenecks remain in the comprehensive analysis and real-time correction of movement postures.

[0003] Existing technologies for monitoring and guiding CPR compressions have significant limitations. They typically focus on monitoring single indicators such as compression depth and frequency, lacking a systematic analysis of the rescuer's overall trunk movement. They cannot identify inefficient force transmission caused by improper posture, leading to poor compression effectiveness due to improper support point selection or shift in the center of gravity, while also increasing the rescuer's fatigue accumulation and risk of injury. Current technologies only superficially analyze the characteristics of thoracic pressure response, failing to establish a dynamic deformation model during pressure application. They cannot accurately assess the force transmission path and energy dissipation in the thoracic tissues, making it difficult for rescuers to find the optimal direction and point of application, potentially resulting in uneven pressure distribution or insufficient effective compression depth. Existing correction and guidance methods lack real-time and systematic approaches, failing to comprehensively consider the dynamic balance between the rescuer's fatigue state, support stability, and pressure application efficiency. The adjustment suggestions provided are often too general, difficult to translate into specific action correction parameters, and cannot achieve precise personalized guidance, thus affecting the continuous improvement of compression quality. Summary of the Invention

[0004] This invention provides an AI-based method for recognizing and correcting deviations in the posture features of cardiopulmonary resuscitation (CPR) compressions, which can solve the problems in the prior art.

[0005] A first aspect of this invention provides an AI-based method for recognizing and correcting deviations in cardiopulmonary resuscitation (CPR) chest compression posture features, comprising:

[0006] Acquire data on trunk movement trajectory and dynamic chest pressure during chest compressions performed by the rescuer;

[0007] Spatial reconstruction is performed on the trunk motion trajectory data to extract the trunk center of mass motion law and angular velocity change trend, calculate the stability components of the trunk posture in the vertical and horizontal planes, determine the optimal force application range and force support point, and generate the rescuer's steady-state support area.

[0008] A pressure response state space is constructed from the dynamic data of intrathoracic pressure, the dynamic characteristics of the pressure action sequence are extracted, a pressure elastic deformation field is established, the energy dissipation distribution of the conduction path is analyzed, and the optimal pressure application path is generated.

[0009] The steady-state support region and the optimal pressure application path are mapped to the elastic potential energy field. The support structure morphology is optimized based on dynamic equilibrium constraints, and a real-time correction attitude adjustment command sequence is generated based on fatigue and stability indices.

[0010] Based on the posture adjustment command sequence, the action correction is performed to adjust the rescuer's standing distance, support angle, and force direction, thereby completing the correction of the compression action deviation.

[0011] In one optional embodiment, spatial reconstruction of the torso motion trajectory data to extract the motion law of the torso's center of mass and the trend of angular velocity changes includes:

[0012] A three-dimensional coordinate system centered on the rescuer's torso is established, and the torso motion trajectory data is mapped to the three-dimensional coordinate system to obtain three-dimensional spatial motion data. The motion boundary is corrected by combining the range of motion of the human torso joints to obtain the corrected torso motion characteristics.

[0013] The corrected trunk motion features are decomposed using a quaternion rotation matrix to obtain the main direction motion component and the auxiliary direction motion component. The main direction motion component and the auxiliary direction motion component are projected and combined to obtain the trunk center of mass motion trajectory. By performing time-domain analysis on the trunk center of mass motion trajectory, the trunk center of mass velocity change is obtained. Combined with displacement compensation, the trunk center of mass motion law is generated.

[0014] The motion law of the torso's center of mass is processed using a spherical linear interpolation algorithm. The rotational features of the torso are extracted to calculate the three-axis rotation angles and determine the torso posture change sequence. The angle differentiation of the torso posture change sequence generates the angular velocity change trend.

[0015] In one optional embodiment, the stability components of the torso posture in the vertical and horizontal planes are calculated to determine the optimal force application range and force support points, generating the rescuer's steady-state support region, including:

[0016] The motion law of the torso's center of mass is decomposed into vertical and horizontal planes to obtain the planar projection trajectory of the torso's center of mass; the trend of angular velocity change is decomposed into vertical and horizontal planes to obtain the planar projection characteristics of the torso's angular motion.

[0017] Amplitude analysis is performed on the plane projection trajectory of the torso's center of mass to extract peak and valley points, and the fluctuation amplitude and time interval between adjacent peak and valley points are calculated to obtain the center of mass motion fluctuation characteristics. The fluctuation period mapping relationship is determined, and the center of mass motion stability characteristics are extracted. The plane projection characteristics of the torso's angular motion are periodically segmented to extract angular motion stability characteristics. The center of mass motion stability characteristics and the angular motion stability characteristics are weighted and combined to generate vertical plane stability components and horizontal plane stability components.

[0018] Threshold segmentation is performed on the vertical plane stability component and the horizontal plane stability component to extract the time period with the optimal stability index and determine it as the optimal stability interval of the torso posture. The torso posture stabilization point is determined based on the motion characteristics within the optimal stability interval of the torso posture. A polygonal envelope is constructed with the torso posture stabilization point as the center to generate the rescuer's steady-state support region.

[0019] In one optional embodiment, a pressure response state space is constructed from the dynamic data of the intrathoracic pressure, dynamic features of the pressure application sequence are extracted, a pressure elastic deformation field is established, the energy dissipation distribution of the conduction path is analyzed, and the optimal pressure application path is generated, including:

[0020] Dynamic data of intrathoracic pressure are mapped to a multidimensional state space, the phase space trajectory of the pressure state vector is extracted, the stable interval is determined according to the topological structure of pressure fluctuations, and the pressure data is nonlinearly reconstructed to obtain a single-cycle pressure action sequence.

[0021] Pressure conduction analysis is performed on the single-cycle pressure action sequence to extract the pressure gradient and propagation velocity at each sampling point. The propagation direction of the pressure wave in the thoracic tissue is determined based on the pressure gradient. The displacement vector of the pressure action point is calculated along the propagation direction, and temporal correlation analysis is performed to obtain the spatial position change sequence of the pressure action point. A pressure elastic deformation field is constructed based on the spatial position change sequence. The propagation path of the pressure wave in the thoracic cavity is determined based on the pressure elastic deformation field to obtain the pressure conduction region.

[0022] The pressure attenuation coefficient between adjacent sampling points is calculated within the pressure transmission region to determine the energy loss characteristics of the pressure wave during transmission. The energy dissipation distribution along the transmission path is obtained through cumulative calculation. Based on the energy dissipation distribution, a transmission efficiency evaluation index is constructed. Multiple transmission paths are sorted and screened according to the transmission efficiency evaluation index to determine the transmission path with the minimum energy dissipation and generate the optimal pressure application path.

[0023] In one optional embodiment, dynamic data of intrathoracic pressure is mapped to a multidimensional state space, the phase space trajectory of the pressure state vector is extracted, a stable interval is determined based on the topological structure of the pressure fluctuation, and a single-cycle pressure action sequence is obtained by nonlinear reconstruction of the pressure data, including:

[0024] Dynamic data of intrathoracic pressure are mapped to a multidimensional state space to construct a phase matrix of pressure response. A time-delay embedding transformation is performed on the phase matrix to generate a pressure state vector.

[0025] Calculate the distance matrix between adjacent state points in the pressure state vector, transform the distance matrix into a state transition diagram, extract the connected components of the state transition diagram, obtain the invariant set in the state space, and use the invariant set to map the phase space trajectory of pressure changes to form the topology of pressure fluctuation.

[0026] The topology is matrix decomposed to extract the structural parameters of the attractor, the convergence domain boundary of the structural parameters is determined, critical state points are marked in the convergence domain boundary, the stable interval of pressure fluctuation is divided according to the critical state points, and the pressure data in the stable interval is nonlinearly reconstructed to generate a single-cycle pressure action sequence.

[0027] In one optional embodiment, mapping the steady-state support region to the optimal pressure application path onto an elastic potential energy field, optimizing the support structure morphology based on dynamic equilibrium constraints, and generating a real-time correction attitude adjustment command sequence based on fatigue and stability indices include:

[0028] The coordinate sequence and pressure value sequence of the support points in the steady-state support area are collected. The potential energy value sequence is obtained by elastic potential energy calculation. The elastic potential energy field is constructed using the potential energy value sequence. The optimal pressure application path is mapped to the elastic potential energy field to obtain the potential energy gradient distribution.

[0029] The strain characteristics of the support point are calculated based on the potential energy gradient distribution, the pressure transmission efficiency is analyzed, a dynamic reconstruction sequence of the support point is generated based on the pressure transmission efficiency, the dynamic reconstruction sequence of the support point is combined with the preset human biomechanical parameters to construct dynamic equilibrium constraints, the support structure morphology is optimized based on the dynamic equilibrium constraints, and the support structure morphology is converted into position coordinate parameters and posture angle parameters.

[0030] Real-time motion data of the rescuer is collected, and muscle fatigue curves and posture stability indices are calculated and determined. The muscle fatigue curves and posture stability indices are input into dynamic balance constraints to update the support point reconstruction parameters. Based on the support point reconstruction parameters, station coordinate parameters, and posture angle parameters, posture correction commands are generated. The posture correction commands are organized in time sequence to obtain a real-time posture adjustment command sequence.

[0031] In one optional embodiment, the strain characteristics of the support point are calculated based on the potential energy gradient distribution, and the pressure transmission efficiency is analyzed, including:

[0032] Extract the principal direction vector of the potential energy gradient distribution, construct a pressure transmission path along the principal direction vector, and calculate the stress concentration factor of the support point on the pressure transmission path;

[0033] The local strain threshold of the support point is determined based on the stress concentration factor. The support points that exceed the strain threshold are classified by strain characteristics. The mapping relationship between the strain of each type of support point and the pressure transmission efficiency is calculated.

[0034] By combining the mapping relationship with the spatial distribution of support points, a pressure transmission network is constructed, and the overall pressure transmission efficiency is determined by calculating the pressure transmission network.

[0035] A second aspect of the present invention provides an AI-based system for recognizing and correcting deviations in cardiopulmonary resuscitation (CPR) chest compression posture features, comprising:

[0036] The data acquisition module is used to acquire data on the trunk movement trajectory and dynamic data on chest pressure when the rescuer performs chest compressions.

[0037] The support area optimization module is used to spatially reconstruct the torso motion trajectory data, extract the torso center of mass motion law and angular velocity change trend, calculate the stability components of the torso posture in the vertical and horizontal planes, determine the optimal force application range and force support point, and generate the rescuer's steady-state support area.

[0038] The pressure path analysis module is used to construct a pressure response state space from the dynamic data of the thoracic pressure, extract the dynamic characteristics of the pressure action sequence, establish a pressure elastic deformation field, analyze the energy dissipation distribution of the conduction path, and generate the optimal pressure application path.

[0039] The attitude correction module is used to map the steady-state support area and the optimal pressure application path to the elastic potential energy field, optimize the support structure morphology based on dynamic equilibrium constraints, and generate a real-time attitude adjustment command sequence based on fatigue and stability indicators.

[0040] The action execution module is used to perform action correction according to the posture adjustment command sequence, adjust the rescuer's standing distance, support angle and force direction, and complete the correction of the compression action deviation.

[0041] A third aspect of the present invention provides an electronic device, comprising:

[0042] processor;

[0043] Memory used to store processor-executable instructions;

[0044] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0045] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0046] In this embodiment of the invention, by acquiring trunk motion trajectory data and chest pressure dynamic data during chest compressions performed by the rescuer, spatial reconstruction of the trunk motion trajectory data and extraction of the trunk's center of mass motion law and angular velocity change trend can accurately calculate the stability components of the trunk posture in the vertical and horizontal planes, determine the optimal force application range and force support point, and generate a scientifically reasonable steady-state support area. This effectively improves the accuracy and comprehensiveness of CPR chest compression posture recognition and provides a reliable data foundation for subsequent deviation correction. By constructing a pressure response state space from the chest pressure dynamic data and extracting the dynamic characteristics of the pressure action sequence, establishing a pressure elastic deformation field, and analyzing the energy dissipation distribution of the conduction path, optimal pressure can be generated. By applying a path and then mapping the steady-state support area and the optimal pressure application path to the elastic potential energy field for dynamic balance constraint optimization, a deep analysis of the mechanical transmission mechanism in the compression action is achieved, ensuring the optimization of the compression effect and significantly improving the treatment efficiency and success rate of cardiopulmonary resuscitation. Based on fatigue and stability indicators, a real-time correction posture adjustment command sequence is generated. According to the command sequence, the rescuer's standing distance, support angle, and force direction are dynamically adjusted, realizing intelligent real-time correction of the compression action. This not only reduces the rescuer's physical exertion and operational fatigue, but also maintains the standardization and effectiveness of the compression action, effectively solving the technical problem that it is difficult to detect and correct movement deviations in a timely manner during traditional cardiopulmonary resuscitation training and implementation. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating the AI-based method for recognizing and correcting deviations in cardiopulmonary resuscitation (CPR) chest compression posture features according to an embodiment of the present invention.

[0048] Figure 2 A flowchart for analyzing the efficiency of thoracic pressure transmission and determining the optimal path. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0051] Figure 1 This is a flowchart illustrating the AI ​​recognition and deviation correction method for cardiopulmonary resuscitation (CPR) chest compression posture features according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0052] Acquire data on trunk movement trajectory and dynamic chest pressure during chest compressions performed by the rescuer;

[0053] Spatial reconstruction is performed on the trunk motion trajectory data to extract the trunk center of mass motion law and angular velocity change trend, calculate the stability components of the trunk posture in the vertical and horizontal planes, determine the optimal force application range and force support point, and generate the rescuer's steady-state support area.

[0054] A pressure response state space is constructed from the dynamic data of intrathoracic pressure, the dynamic characteristics of the pressure action sequence are extracted, a pressure elastic deformation field is established, the energy dissipation distribution of the conduction path is analyzed, and the optimal pressure application path is generated.

[0055] The steady-state support region and the optimal pressure application path are mapped to the elastic potential energy field. The support structure morphology is optimized based on dynamic equilibrium constraints, and a real-time correction attitude adjustment command sequence is generated based on fatigue and stability indices.

[0056] Based on the posture adjustment command sequence, the action correction is performed to adjust the rescuer's standing distance, support angle, and force direction, thereby completing the correction of the compression action deviation.

[0057] In one optional implementation, spatial reconstruction of the torso motion trajectory data to extract the motion law of the torso's center of mass and the trend of angular velocity changes includes:

[0058] A three-dimensional coordinate system centered on the rescuer's torso is established, and the torso motion trajectory data is mapped to the three-dimensional coordinate system to obtain three-dimensional spatial motion data. The motion boundary is corrected by combining the range of motion of the human torso joints to obtain the corrected torso motion characteristics.

[0059] The corrected trunk motion features are decomposed using a quaternion rotation matrix to obtain the main direction motion component and the auxiliary direction motion component. The main direction motion component and the auxiliary direction motion component are projected and combined to obtain the trunk center of mass motion trajectory. By performing time-domain analysis on the trunk center of mass motion trajectory, the trunk center of mass velocity change is obtained. Combined with displacement compensation, the trunk center of mass motion law is generated.

[0060] The motion law of the torso's center of mass is processed using a spherical linear interpolation algorithm. The rotational features of the torso are extracted to calculate the three-axis rotation angles and determine the torso posture change sequence. The angle differentiation of the torso posture change sequence generates the angular velocity change trend.

[0061] In one specific implementation, processing begins by establishing a three-dimensional coordinate system centered on the rescuer's torso, enabling accurate capture and analysis of the rescuer's torso posture characteristics during CPR. Posture sensors are placed at specific locations on the rescuer's torso to acquire raw motion trajectory data of the torso during CPR compressions. These sensors can be inertial measurement units (IMUs), including three-axis accelerometers and three-axis gyroscopes, capable of acquiring real-time acceleration and angular velocity data of the rescuer's torso. In practical applications, the sampling frequency is set to 100Hz to ensure the continuity and accuracy of data acquisition.

[0062] After acquiring the raw data, a three-dimensional coordinate system is established centered on the rescuer's torso, with the x-axis pointing directly in front of the rescuer, the y-axis pointing to the rescuer's right side, and the z-axis pointing vertically upwards. The raw sensor data is mapped to this three-dimensional coordinate system through coordinate transformation, enabling spatial localization of torso movement. This process involves calibrating the initial sensor pose, determining the sensor's initial orientation in space through static posture acquisition. Taking a cardiopulmonary resuscitation (CPR) procedure as an example, the sensor's initial position is at the rescuer's T7 thoracic vertebra, with its initial orientation aligned with the rescuer's torso.

[0063] After completing the three-dimensional spatial mapping, motion boundary correction is performed based on the range of motion of the human torso joints. The human torso has physiological limitations in its range of motion; for example, the maximum angle of forward flexion is approximately 90 degrees, the maximum angle of backward extension is approximately 30 degrees, the maximum angle of lateral flexion is approximately 35 degrees, and the maximum angle of rotation is approximately 45 degrees. Motion boundaries are set based on these physiological parameters, and data exceeding the reasonable range is corrected. In actual cases, when the rescuer's torso flexion angle is detected to reach 92 degrees, it is corrected to 90 degrees to conform to human physiological limits, thereby avoiding unreasonable data caused by measurement errors.

[0064] To address the corrected trunk motion characteristics, a quaternion rotation matrix was used for decomposition. The three-dimensional motion of the trunk was decomposed into rotations around three orthogonal axes: the x-axis, y-axis, and z-axis. In practice, quaternions were used to represent the spatial posture of the rescuer's trunk, and rotational transformations between consecutive time points were calculated using quaternion multiplication. The current posture was represented by the quaternion q, and the principal and secondary motion components were extracted by analyzing the real and imaginary parts of the quaternion. During CPR compressions, the principal direction is typically vertical (z-axis), while the secondary directions are the forward / backward and left / right directions in the horizontal plane (x-axis and y-axis).

[0065] The trajectory of the torso's center of mass is obtained by projecting and combining the primary and secondary motion components. The positional change of the center of mass is calculated by analyzing the geometric relationship between the sensor location and the rescuer's center of mass. In a real-world case, assuming a vertical distance of 20 cm between the sensor location and the center of mass, the sensor trajectory is converted into the center of mass trajectory through spatial geometric transformation. In ideal cardiopulmonary resuscitation (CPR), the center of mass trajectory should exhibit a regular up-and-down reciprocating motion, with a vertical displacement of approximately 5-6 cm and a horizontal displacement kept to a minimum.

[0066] A time-domain analysis was performed on the trajectory of the trunk's center of mass to calculate the velocity changes in each direction. Velocity changes were obtained through positional differences between adjacent time points, and the instantaneous velocity was derived by combining these differences with the time intervals. During standard cardiopulmonary resuscitation (CPR), the maximum vertical velocity of the center of mass is approximately 0.5-0.6 m / s, while the horizontal velocity should be controlled within 0.1 m / s. By comparing the actual measured velocity with the ideal value, displacement compensation was performed to eliminate cumulative errors caused by sensor drift or measurement errors, thus generating an accurate law governing the trunk's center of mass motion.

[0067] To analyze the generated torso center-of-mass motion patterns, a spherical linear interpolation algorithm is applied to extract torso rotational features. This algorithm can interpolate discrete attitude data to generate continuous rotation sequences while maintaining rotational smoothness. In practical applications, attitude data is interpolated every 0.1 seconds to generate continuous attitude sequences with 10-millisecond intervals, ensuring smoothness of attitude changes. By analyzing the interpolated rotation sequences, the torso rotation angles around three axes are calculated: pitch angle (rotation around the y-axis), roll angle (rotation around the x-axis), and yaw angle (rotation around the z-axis).

[0068] After determining the sequence of trunk posture changes, angular differentiation is performed on the sequence to generate the trend of angular velocity changes. Angular velocity is calculated by dividing the angle difference between adjacent time points by the time interval, reflecting the rate of trunk rotation change of the rescuer. During standard cardiopulmonary resuscitation (CPR), the maximum angular velocity in the pitch direction is approximately 80-100 degrees / second, while the angular velocities in the roll and yaw directions should be controlled within 20 degrees / second. By analyzing the trend of angular velocity changes, abnormal changes in the rescuer's compression posture can be identified, such as unnecessary trunk rotation or tilting during compressions.

[0069] In a real-world CPR case, analysis of the rescuer's compressions over two minutes (approximately 240 compression cycles) revealed that the rescuer's torso gradually tilted to the right during compressions, with the roll angle increasing from an initial 3 degrees to 12 degrees, exceeding the ideal compression posture. Simultaneously, the compression depth decreased from an initial 5.5 cm to 4.2 cm, falling below the requirements for effective compression. By instructing the rescuer to adjust their compression posture, keeping their torso vertical, the quality of compressions was effectively improved, and the compression depth returned to approximately 5.5 cm, meeting the quality requirements for CPR.

[0070] In one optional implementation, the stability components of the torso posture in the vertical and horizontal planes are calculated to determine the optimal force application range and force support points, generating the rescuer's steady-state support region, including:

[0071] The motion law of the torso's center of mass is decomposed into vertical and horizontal planes to obtain the planar projection trajectory of the torso's center of mass; the trend of angular velocity change is decomposed into vertical and horizontal planes to obtain the planar projection characteristics of the torso's angular motion.

[0072] Amplitude analysis is performed on the plane projection trajectory of the torso's center of mass to extract peak and valley points, and the fluctuation amplitude and time interval between adjacent peak and valley points are calculated to obtain the center of mass motion fluctuation characteristics. The fluctuation period mapping relationship is determined, and the center of mass motion stability characteristics are extracted. The plane projection characteristics of the torso's angular motion are periodically segmented to extract angular motion stability characteristics. The center of mass motion stability characteristics and the angular motion stability characteristics are weighted and combined to generate vertical plane stability components and horizontal plane stability components.

[0073] Threshold segmentation is performed on the vertical plane stability component and the horizontal plane stability component to extract the time period with the optimal stability index and determine it as the optimal stability interval of the torso posture. The torso posture stabilization point is determined based on the motion characteristics within the optimal stability interval of the torso posture. A polygonal envelope is constructed with the torso posture stabilization point as the center to generate the rescuer's steady-state support region.

[0074] In one specific implementation, during emergency treatment, after acquiring the three-dimensional motion trajectory data of the torso's center of mass, this trajectory is projected onto two mutually perpendicular planes. The spatial coordinates of the center of mass are decomposed according to the direction of gravity and the direction parallel to the ground. The direction of gravity and its perpendicular direction constitute the vertical projection plane, and the two orthogonal directions parallel to the ground constitute the horizontal projection plane. The center of mass position coordinates are acquired in 100-millisecond time windows, and 600 data points are continuously acquired to form a complete projection trajectory curve. In the vertical plane, the center of mass trajectory exhibits an undulating curve shape, reflecting the torso's motion pattern in the vertical direction. In the horizontal plane, the center of mass trajectory presents a forward, backward, left, and right movement trajectory, reflecting the torso's swinging characteristics in the horizontal direction.

[0075] For the angular motion analysis of the torso, the angular velocity vector of the torso's principal axis relative to the reference coordinate system is extracted. This vector contains three components. The angular velocity vector is projected onto both the vertical and horizontal planes. The angular velocity component in the vertical plane reflects the forward and backward tilting velocities of the torso, while the angular velocity component in the horizontal plane reflects the left and right torsional twisting velocities. The angular velocity values ​​are recorded at a frequency of 10 sampling points per second for 60 seconds, resulting in 600 sets of angular velocity data. These data are then categorized and organized according to the projection planes, yielding two independent angular motion feature sequences.

[0076] Amplitude extraction is performed on the centroid trajectory curve within the vertical plane. A sliding window width of 20 data points is set. Local maxima are searched within the window as peak points, and local minima are searched as valley points. A peak point is marked when five consecutive data points have values ​​less than the center point value, and a valley point is marked when five consecutive data points have values ​​greater than the center point value. For example, in a certain data collection, the centroid height of the 12th data point is 152 cm, and the heights of the five data points before and after it are all less than 152 cm; therefore, this point is marked as a peak point. The extracted peak point sequence contains both a timestamp and a height value, and the valley point sequence also contains these two pieces of information.

[0077] Calculate the height difference between adjacent peak and trough points; this difference represents the fluctuation amplitude. If, within a certain time period, the peak height is 155 cm and the subsequent trough height is 148 cm, then the fluctuation amplitude for that period is 7 cm. Calculate the time difference between adjacent peak points as the fluctuation period. In one actual measurement, the first peak point occurred at 3.2 seconds, and the second peak point occurred at 5.8 seconds; therefore, the fluctuation period is 2.6 seconds. Statistically analyze all fluctuation periods. When a period's value occurs more than 30% of the total number of periods, that period is considered the dominant period. Divide all fluctuations into time periods according to the dominant period, with each dominant period serving as an analysis unit.

[0078] Within each analysis unit, the standard deviation of the fluctuation amplitude is calculated. A smaller standard deviation indicates a more regular centroid motion within that time period. One time period contains eight fluctuation cycles with amplitudes of 6.2 cm, 6.5 cm, 6.1 cm, 6.4 cm, 6.3 cm, 6.6 cm, 6.2 cm, and 6.4 cm, respectively, yielding a standard deviation of 0.17 cm. Another time period has a standard deviation of 0.52 cm for its eight fluctuation amplitudes, indicating that the stability of the centroid motion in the former time period is better than that in the latter. The standard deviation is normalized to the range of 0 to 1; a value closer to 1 indicates higher stability. For vertical plane centroid motion, a normalized stability value exceeding 0.75 is considered to meet the support requirements for that time period.

[0079] The same amplitude analysis procedure was applied to the centroid trajectory within the horizontal plane to extract the peak and trough points in the horizontal direction, and to calculate the fluctuation amplitude and period of the horizontal displacement. In one measurement, the dominant period of the horizontal displacement was 2.4 seconds, and the displacement amplitude within a single period ranged from 8 cm to 12 cm. The stability index of the centroid motion in the horizontal plane was calculated, and a normalized value was obtained. When this value exceeded 0.70, the centroid motion within the horizontal plane was considered to be in a stable state.

[0080] To analyze the angular motion characteristics, the angular velocity sequence in the vertical plane is segmented according to the dominant period of the center-of-mass motion. Within each period segment, the average and peak angular velocities are calculated. The average reflects the overall angular motion trend of that period, while the peak reflects the maximum intensity of the angular motion. Within a given period segment, the average angular velocity is 12 degrees per second, the peak is 18 degrees per second, and the ratio of the peak to the average is 1.5. The ratios for all period segments are statistically analyzed; when the coefficient of variation of the ratio is less than 0.25, the angular motion is considered to have good stability. The stability of the angular motion is also normalized to the range of 0 to 1; a value greater than 0.72 is considered a stable state.

[0081] The stability characteristics of the center of mass motion and angular motion are combined, and a weighted summation method is used to generate a comprehensive stability index. For the vertical plane, the weight of the center of mass motion stability is set to 0.6, and the weight of the angular motion stability is set to 0.4. At a certain time period, the centroid stability of the vertical plane is 0.82, and the angular motion stability is 0.76, so the comprehensive stability of the vertical plane is 0.796. The same weighting is applied to the horizontal plane; at a certain time period, the centroid stability of the horizontal plane is 0.78, and the angular motion stability is 0.74, so the comprehensive stability of the horizontal plane is 0.764. The comprehensive stability values ​​for continuous time periods are plotted as time-series curves.

[0082] A vertical stability threshold of 0.75 and a horizontal stability threshold of 0.72 were set. Time periods that simultaneously met both thresholds were extracted from the time-series curve. During a rescue operation, from the 8th to the 15th second, both vertical stability and horizontal stability consistently exceeded 0.75 and 0.72, respectively; this time period was marked as the optimal stability interval. The average position coordinates of the torso's center of mass were extracted within this interval; these coordinates are the point of action for torso posture stability. In actual measurements, this point of action was located in the middle of the torso, 42 cm from the upper end and 38 cm from the lower end, offset 3 cm to the left of the body's central axis.

[0083] A rectangular envelope region is constructed in the horizontal plane with the stable point of action as the geometric center. The side length of the rectangle is determined based on the maximum offset of the center of mass within the optimal stable range. The maximum horizontal offset is 9 cm, so the rectangle length is set to 18 cm. The maximum forward / backward offset is 7 cm, so the rectangle width is set to 14 cm. This rectangular region represents the rescuer's steady-state support range in the horizontal direction. In the vertical direction, the support height range is determined based on the fluctuation range of the center of mass height. Within the optimal stable range, the center of mass height fluctuates between 146 cm and 154 cm, so the vertical support range is set within this range. The rescuer's hands should be positioned within the horizontal rectangular region, and the support height should be within the vertical range to ensure the most effective stable support for the injured person's torso.

[0084] In one optional implementation, a pressure response state space is constructed from the dynamic data of the intrathoracic pressure, the dynamic characteristics of the pressure application sequence are extracted, a pressure elastic deformation field is established, the energy dissipation distribution of the conduction path is analyzed, and the optimal pressure application path is generated, including:

[0085] Dynamic data of intrathoracic pressure are mapped to a multidimensional state space, the phase space trajectory of the pressure state vector is extracted, the stable interval is determined according to the topological structure of pressure fluctuations, and the pressure data is nonlinearly reconstructed to obtain a single-cycle pressure action sequence.

[0086] Pressure conduction analysis is performed on the single-cycle pressure action sequence to extract the pressure gradient and propagation velocity at each sampling point. The propagation direction of the pressure wave in the thoracic tissue is determined based on the pressure gradient. The displacement vector of the pressure action point is calculated along the propagation direction, and temporal correlation analysis is performed to obtain the spatial position change sequence of the pressure action point. A pressure elastic deformation field is constructed based on the spatial position change sequence. The propagation path of the pressure wave in the thoracic cavity is determined based on the pressure elastic deformation field to obtain the pressure conduction region.

[0087] The pressure attenuation coefficient between adjacent sampling points is calculated within the pressure transmission region to determine the energy loss characteristics of the pressure wave during transmission. The energy dissipation distribution along the transmission path is obtained through cumulative calculation. Based on the energy dissipation distribution, a transmission efficiency evaluation index is constructed. Multiple transmission paths are sorted and screened according to the transmission efficiency evaluation index to determine the transmission path with the minimum energy dissipation and generate the optimal pressure application path.

[0088] In one specific implementation, the collected dynamic data of intrathoracic pressure are arranged in a time series, with each time point containing pressure value, sampling time, and corresponding spatial coordinates. Assuming 128 data points are collected within a complete cardiac cycle, each data point is recorded as a pressure value in three-dimensional coordinates. These pressure data are mapped to a six-dimensional state space, where three dimensions represent spatial coordinates, and the other three dimensions represent pressure value, pressure rate of change, and pressure acceleration, respectively. The time series data is converted into a state space trajectory using a delayed embedding method, with a delay time step of 8 sampling points and an embedding dimension of 6. The motion trajectory of the pressure state vector is plotted in the state space, and the geometric morphology of the trajectory is observed.

[0089] Topological analysis is performed on the state-space trajectory to identify attractor structures and periodic features. The regression characteristics of the trajectory are identified by calculating the Euclidean distance between trajectory points; points with a distance less than 0.15 times the average distance are considered regression points. The distribution density of regression points is statistically analyzed to determine the stable pressure fluctuation range. For example, in the region with pressure values ​​between 4 kPa and 7 kPa, the regression point density reaches 12 points per cubic centimeter, and this region is marked as a stable pressure range. Unstable pressure fluctuation data is filtered using a noise reduction method based on local linear fitting, retaining pressure data within the stable range as valid samples.

[0090] The filtered pressure data underwent nonlinear reconstruction, dividing the continuous pressure time series into single-cycle segments. The start and end points of the cycle were identified by detecting pressure peaks and troughs. Within a typical pressure cycle, the pressure rises from a baseline of 3.2 kPa to a peak of 8.5 kPa, then falls back to the baseline, with the entire cycle lasting 0.8 seconds. Pressure data from all sampling points within this cycle were extracted, forming a single-cycle pressure action sequence containing 128 data points. This sequence was normalized, mapping the pressure values ​​to the zero-to-one range for easier subsequent quantitative analysis.

[0091] For each sampling point in a single-cycle pressure action sequence, the pressure difference between it and its adjacent sampling points is calculated. Assuming the pressure value at sampling point 45 is 6.8 kPa, the pressure value at the preceding sampling point is 6.5 kPa, and the pressure value at the following sampling point is 7.1 kPa, then the forward pressure gradient at this point is 0.3 kPa / cm, and the backward pressure gradient is also 0.3 kPa / cm. A pressure gradient distribution map is generated by calculating the pressure gradients at all sampling points. The direction information of the pressure gradient is extracted, and the direction angle of the gradient vector represents the pressure conduction direction. For sampling points in three-dimensional space, the pressure gradient vector contains three components. By calculating the magnitude and direction angle of the vector, the main conduction direction of the pressure wave is determined.

[0092] The propagation speed of pressure waves is estimated based on the magnitude of the pressure gradient. In areas with a larger pressure gradient, the pressure wave propagates faster, while in areas with a smaller pressure gradient, it propagates slower. The local propagation speed is calculated by statistically analyzing the time difference of pressure changes between adjacent sampling points and combining this with the spatial distance between the sampling points. For example, on a certain propagation path, if the distance between two adjacent points is 2 mm and the time interval between pressure changes is 0.006 seconds, then the propagation speed of this path segment is 0.33 meters per second. The propagation speed data from all sampling points are then integrated to generate a propagation speed field distribution map.

[0093] The spatial position changes of each sampling point are tracked along the direction of pressure transmission. Under pressure, the thoracic tissue undergoes elastic deformation, and the spatial coordinates of the sampling points shift over time. The displacement vector is calculated by comparing the coordinate values ​​of the same sampling point at different times. For example, assuming the initial coordinates of a sampling point are 12 cm horizontally, 8 cm vertically, and 5 cm deep, after 0.2 seconds of pressure application, the coordinates change to 12.15 cm horizontally, 8.08 cm vertically, and 5.03 cm deep. The displacement vector of this point is then 0.15 cm horizontally, 0.08 cm vertically, and 0.03 cm deep. A temporal correlation analysis is performed on the displacement vectors of all sampling points to identify the continuity and correlation of displacement changes.

[0094] A pressure-elastic deformation field model was constructed based on the spatial position change sequence of sampling points. The thoracic cavity was divided into several voxel units, each with a size of 2 mm cubic. The displacement data of the sampling points within each voxel were interpolated to obtain the average deformation of the voxel. An interpolation method based on radial basis functions was used to infer the deformation of unsampled locations based on the displacement data of known sampling points. A complete three-dimensional deformation field was generated, which describes the spatial deformation distribution of thoracic tissue under pressure. The deformation field was visualized as a cloud map using visualization technology, with different colors representing different deformation intensities. Areas with deformation greater than 0.2 cm were marked in red, and areas with deformation less than 0.05 cm were marked in blue.

[0095] The actual propagation path of the pressure wave within the thoracic cavity is determined based on the pressure-elastic deformation field. The pressure wave propagates along the direction of the maximum deformation gradient, forming the main conduction channel. By tracing the gradient direction in the deformation field, a spatial curve of the conduction path is generated, extending gradually from the point of pressure application. This curve connects a series of voxel units with decreasing deformation intensity, forming a continuous conduction region. The influence range of the pressure wave is marked within this conduction region, and the area where the pressure value decays to less than 10% of the initial value is defined as the conduction boundary.

[0096] Within a defined pressure transmission area, adjacent sampling point pairs along the transmission path are selected, and their pressure attenuation coefficients are calculated. Assuming a starting pressure of 7.2 kPa and an ending pressure of 6.9 kPa in a sampling point pair, with a distance of 3 mm between the two points, the pressure attenuation coefficient for this segment of the path is 0.1 kPa per centimeter. A pressure attenuation coefficient distribution matrix is ​​established by traversing all adjacent sampling point pairs within the transmission area. The attenuation coefficients are then cumulatively calculated, starting from the pressure application point and adding the pressure loss segment by segment along the transmission path. After transmission through 20 sampling points, the cumulative pressure loss reaches 1.8 kPa.

[0097] Pressure loss is converted into energy dissipation. The energy carried by the pressure wave is gradually dissipated during conduction due to the viscoelastic properties of the tissue. The energy value is calculated based on the pressure value and the volume change. The product of pressure and volume change represents the work done, which is the dissipated energy. In a certain conduction path segment, a pressure drop of 0.3 kPa corresponds to a volume change of 0.008 cubic centimeters in a voxel, so the energy dissipation in that segment is 0.0024 millijoules. By summing the energy dissipation values ​​of each segment along the entire conduction path, the total energy dissipation distribution curve of the path is obtained.

[0098] A conduction efficiency evaluation index system is constructed, which comprehensively considers three factors: conduction path length, total energy dissipation, and conduction time. Conduction efficiency is defined as the ratio of effective conduction distance to conduction time per unit energy dissipation. For multiple possible conduction paths, their conduction efficiency index values ​​are calculated separately. For example, path A has an effective conduction distance of 8 cm, energy dissipation of 0.045 mJ, and a conduction time of 0.15 seconds, with an efficiency index value of 1185. Path B has an effective conduction distance of 7.5 cm, energy dissipation of 0.038 mJ, and a conduction time of 0.12 seconds, with an efficiency index value of 1644.

[0099] All candidate conduction paths are sorted in descending order of efficiency index value, and the path with the highest index value is selected as the optimal conduction path. During the selection process, the stability of the path also needs to be verified by repeatedly calculating the conduction efficiency across different periods to ensure that the selected path maintains high efficiency over multiple periods. The final determined minimum energy dissipation conduction path has the shortest conduction distance, the lowest energy loss, and the fastest propagation speed. The spatial coordinate sequence of this path constitutes the geometric description of the optimal pressure application path and is used to guide actual pressure application operations.

[0100] like Figure 2 As shown, a flowchart illustrating the analysis of intrathoracic pressure transmission efficiency and the determination of the optimal path is presented.

[0101] In one optional implementation, dynamic data of intrathoracic pressure is mapped to a multidimensional state space, the phase space trajectory of the pressure state vector is extracted, a stable interval is determined based on the topological structure of the pressure fluctuation, and a single-cycle pressure action sequence is obtained by nonlinear reconstruction of the pressure data, including:

[0102] Dynamic data of intrathoracic pressure are mapped to a multidimensional state space to construct a phase matrix of pressure response. A time-delay embedding transformation is performed on the phase matrix to generate a pressure state vector.

[0103] Calculate the distance matrix between adjacent state points in the pressure state vector, transform the distance matrix into a state transition diagram, extract the connected components of the state transition diagram, obtain the invariant set in the state space, and use the invariant set to map the phase space trajectory of pressure changes to form the topology of pressure fluctuation.

[0104] The topology is matrix decomposed to extract the structural parameters of the attractor, the convergence domain boundary of the structural parameters is determined, critical state points are marked in the convergence domain boundary, the stable interval of pressure fluctuation is divided according to the critical state points, and the pressure data in the stable interval is nonlinearly reconstructed to generate a single-cycle pressure action sequence.

[0105] In one specific implementation, during the mapping of dynamic intrathoracic pressure data to a multidimensional state space, a continuous pressure signal sequence with a sampling frequency of 100 Hz is received. This sequence contains intrathoracic pressure measurements over a duration of 60 seconds. Assuming the acquired raw pressure data is a one-dimensional time series with values ​​ranging from -8 cmH2O to +15 cmH2O, this one-dimensional sequence is extended according to a time delay parameter. Specifically, a delay time step of 5 sampling points is selected, and the embedding dimension is set to 3 dimensions. The nth, n+5th, and n+10th data points in the original sequence are combined into a three-dimensional vector. All possible three-dimensional vectors are constructed sequentially using a sliding window method, ultimately resulting in 5990 three-dimensional state vectors. These vectors constitute the phase matrix of the pressure response, with a matrix size of 5990 rows and 3 columns.

[0106] After the phase matrix is ​​constructed, a time-delay embedding transformation is performed to further process the matrix. The Euclidean spatial distance between each 3D vector and all other vectors is calculated, forming a 5990 x 5990 distance matrix. The distance calculation uses the square root of the sum of the squares of the differences between the three components. For vector A, which contains components a1, a2, and a3, and vector B, which contains components b1, b2, and b3, the distance value is equal to the square root of the sum of the squares of the differences between the three corresponding components. Each element in the distance matrix represents the geometric distance between two corresponding state points in 3D space, with distance values ​​ranging from 0 to 32.7 centimeters (water column units).

[0107] When converting the distance matrix into a state transition graph, a threshold parameter of 1.2 cm of water column was set. When the distance between two state points is less than this threshold, a connection edge was established in the state transition graph, indicating that the two states can transition to each other. After thresholding, the state transition graph presents a sparse network structure containing 5990 nodes and approximately 47000 edges. Connectivity analysis was performed on this network graph, using a depth-first search algorithm to traverse all nodes and identify the sets of nodes that can reach each other. The analysis results show that the state transition graph contains 7 main connected components, with the largest connected component containing 4823 nodes, accounting for 80.5% of the total number of nodes. The remaining 6 connected components are smaller, with the number of nodes ranging from 50 to 200.

[0108] Each connected component corresponds to an invariant set region in the state space, and the invariant set corresponding to the largest connected component represents the main range of pressure state activity within a normal respiratory cycle. The system connects the pressure state vectors at each moment in chronological order in three-dimensional space, forming a continuous phase space trajectory curve. This trajectory exhibits a periodic, circling characteristic in three-dimensional space, moving cyclically around a central region. The projection of the trajectory onto various coordinate planes displays an elliptical or spiral shape. By tracking the transitions between different connected components of the trajectory, the topological structure of pressure fluctuations is identified, which contains seven stable attraction domains and corresponding connection paths.

[0109] The matrix decomposition of the topology employs singular value decomposition (SVD) to process the phase matrix. The 5990-row, 3-column phase matrix is ​​decomposed into the product of three matrices, yielding a left singular matrix, a singular value diagonal matrix, and a right singular matrix. The singular value diagonal matrix contains three non-zero singular values: 186.4, 52.3, and 18.7. These values ​​reflect the magnitude of the pressure state vector's variation along the three principal directions. The singular vector corresponding to the largest singular value is extracted as the directional feature of the principal attractor. The three components of this vector are 0.68, 0.59, and 0.43, indicating the relative contribution of the pressure state across the three embedding dimensions.

[0110] The structural parameters of the attractor include the center position coordinates and the diffusion radius. The center position of the main attractor is determined by calculating the arithmetic mean of all state vectors in the largest connected component, with coordinates of -2.1, -1.8, and -1.6 cm water columns for the three components. The distance from each state point to the center position is calculated, and the cumulative probability of the distance distribution is statistically analyzed. When the cumulative probability reaches 95%, the corresponding distance is 8.4 cm water columns, which is defined as the effective diffusion radius of the attractor. The convergence domain boundary of the structural parameters is determined by analyzing the trajectory point density. The three-dimensional state space is divided into a cubic grid with a side length of 0.5 cm water columns. The number of trajectory points in each grid is counted, decreasing from high-density regions outwards. When the number of trajectory points in a grid drops below 10% of the average density, that grid is marked as the convergence domain boundary region.

[0111] Critical state points in the boundary region meet specific criteria: the point is located at the boundary between high-density and low-density grids, and the corresponding pressure change rate exceeds a threshold of 3 cm / s water column. The system identified 236 critical state points, which correspond to moments of pressure abrupt changes or cycle transitions in the time series. Based on the temporal location of the critical state points, the entire 60-second pressure data was divided into 23 stable intervals, each ranging from 1.8 to 3.5 seconds in duration, with an average duration of 2.6 seconds, which is essentially consistent with a normal respiratory cycle.

[0112] When performing nonlinear reconstruction of pressure data within a stable interval, a locally linear embedding method is employed. Taking the first stable interval as an example, this interval contains 180 sampling points, corresponding to 180 three-dimensional state vectors. For each state vector, its 12 nearest neighbor vectors in the state space are found, and reconstruction weight coefficients are calculated so that the vector can be approximated by a linear combination of its neighbor vectors. The reconstruction weight coefficients are obtained by minimizing the reconstruction error, which is defined as the squared distance between the original vector and the reconstructed vector. After obtaining the reconstruction weights of all state vectors, the high-dimensional state space is mapped to a one-dimensional time axis, preserving the topological relationship of the neighborhood structure, to generate a single-cycle pressure action sequence. This sequence contains 180 reconstructed pressure values, ranging from -6.2 to +8.9 cmH2O. The sequence curve exhibits a smooth three-stage characteristic of inhalation-exhalation-interval. During the inhalation stage, the pressure decreases from zero to -6.2 cmH2O for 0.9 seconds; during the exhalation stage, the pressure rises from negative to +8.9 cmH2O for 1.2 seconds; and during the interval stage, the pressure returns to near zero for 0.7 seconds.

[0113] In one optional implementation, the steady-state support region is mapped to the optimal pressure application path onto an elastic potential energy field, the support structure morphology is optimized based on dynamic equilibrium constraints, and a real-time correction attitude adjustment command sequence is generated based on fatigue and stability indices, including:

[0114] The coordinate sequence and pressure value sequence of the support points in the steady-state support area are collected. The potential energy value sequence is obtained by elastic potential energy calculation. The elastic potential energy field is constructed using the potential energy value sequence. The optimal pressure application path is mapped to the elastic potential energy field to obtain the potential energy gradient distribution.

[0115] The strain characteristics of the support point are calculated based on the potential energy gradient distribution, the pressure transmission efficiency is analyzed, a dynamic reconstruction sequence of the support point is generated based on the pressure transmission efficiency, the dynamic reconstruction sequence of the support point is combined with the preset human biomechanical parameters to construct dynamic equilibrium constraints, the support structure morphology is optimized based on the dynamic equilibrium constraints, and the support structure morphology is converted into position coordinate parameters and posture angle parameters.

[0116] Real-time motion data of the rescuer is collected, and muscle fatigue curves and posture stability indices are calculated and determined. The muscle fatigue curves and posture stability indices are input into dynamic balance constraints to update the support point reconstruction parameters. Based on the support point reconstruction parameters, station coordinate parameters, and posture angle parameters, posture correction commands are generated. The posture correction commands are organized in time sequence to obtain a real-time posture adjustment command sequence.

[0117] In one specific implementation, during the mapping of the steady-state support region and the optimal pressure application path to the elastic potential energy field, the support point coordinate sequence is acquired through a pressure sensor array. This sensor array is arranged at 10 mm intervals on the contact surface of the rescuer's foot, recording the three-dimensional spatial coordinate values ​​of each support point in real time. The pressure value sequence acquisition uses a piezoresistive sensor with a range of 0 to 500 Newtons, and the sampling frequency is set to 100 Hz to ensure the capture of the complete dynamic process of pressure changes. When the rescuer is in a standard standing position, the coordinates of the left heel support point are 200 mm on the X-axis, 150 mm on the Y-axis, and 0 mm on the Z-axis, corresponding to a pressure value of 320 Newtons; the coordinates of the forefoot support point are 220 mm on the X-axis, 280 mm on the Y-axis, and 0 mm on the Z-axis, corresponding to a pressure value of 280 Newtons. The support points of the right foot exhibit a mirror-symmetric distribution.

[0118] The elastic potential energy calculation process involves multiplying the pressure value at the support point by the vertical displacement at that point, and then multiplying by a coefficient of 0.5 to obtain the potential energy value at a single point. When the left heel support point is subjected to a pressure of 320 Newtons, the vertical downward displacement at that point is 8 millimeters, and the calculated potential energy value is 1280 millijoules. The forefoot support point produces a displacement of 6 millimeters under a pressure of 280 Newtons, corresponding to a potential energy value of 840 millijoules. The potential energy values ​​of all support points are correlated according to their spatial positions to construct a three-dimensional potential energy distribution matrix covering the entire sole area. This matrix is ​​divided into 50 units along the X-axis and 80 units along the Y-axis, with each unit storing the potential energy value at the corresponding location, forming a complete elastic potential energy field.

[0119] During optimal pressure application path mapping, the pre-determined chest compression trajectory is transformed into the movement trajectory of the rescuer's plantar pressure center. This trajectory starts at the midpoint of the line connecting the two feet, shifted forward 30 mm, and ends at the center of the forefoot region. Each sampling point along this path is matched with its spatial location in the elastic potential energy field, and the corresponding potential energy value is extracted to generate a potential energy variation curve distributed along the path. The potential energy value is 950 mJ at the starting point, reaches a maximum of 1450 mJ in the middle section, and drops back to 1100 mJ at the end point. The potential energy gradient is obtained by calculating the ratio of the potential energy difference between adjacent sampling points to the spatial distance. The average potential energy gradient from the starting point to the middle section increases by 5 mJ per millimeter, while the gradient from the middle section to the end point decreases by 3.5 mJ per millimeter.

[0120] The strain characteristics of the support points were calculated based on the potential energy gradient distribution data. The strain value was determined by the ratio of the vertical displacement of the support point to its initial height. The initial height of the heel support point was 60 mm, and the strain value was 0.133 with an 8 mm displacement. The initial height of the forefoot support point was 45 mm, and the strain value was 0.133 with a 6 mm displacement. Due to its unique structure, the mid-arch of the foot had an initial height of 25 mm, resulting in only a 2 mm displacement and a strain value of 0.08. Pressure transmission efficiency analysis was achieved by tracking the attenuation law of pressure diffusion from the initial contact point to the surrounding area. When the pressure at the center of the heel was 320 N, it decreased to 270 N after diffusion 10 mm, with a transmission efficiency of 84.4%. At 20 mm, the pressure dropped to 210 N, and the transmission efficiency decreased to 65.6%. Due to the dense bone structure in the forefoot area, the pressure transmission efficiency remained above 90% within a 10 mm range.

[0121] The dynamic reconstruction sequence of support points identifies the locations of support points requiring adjustment based on the spatial distribution characteristics of pressure transmission efficiency. When the average transmission efficiency within a 10 mm radius of a support point is below 75%, the point is determined to require position reconstruction. The reconstruction operation is achieved by adjusting the point's Y-axis coordinate; the adjustment amount is proportional to the transmission efficiency deviation, with a 1% increase in deviation corresponding to an increase of 0.5 mm in adjustment. For example, the transmission efficiency detected at the lateral arch support point was 68%, 7 percentage points below the threshold. The Y-axis coordinate of this point was adjusted inward by 3.5 mm, resulting in a reconstruction that increased the transmission efficiency to 79%.

[0122] The biomechanical parameters for the rescuer include basic data such as height, weight, range of motion of joints, and muscle strength level. For a rescuer with a height of 1750 mm, the maximum knee flexion angle is set at 130 degrees, the upper limit of the hip flexion angle is 90 degrees, and the ankle dorsiflexion angle ranges from -15 degrees to +30 degrees. With a weight of 700 Newtons, the maximum output torque of the lower limb muscles is 180 Newton-meters at the knee joint and 90 Newton-meters at the ankle joint. During the construction of dynamic balance constraints, the coordinates of each point in the dynamic reconstruction sequence of support points are correlated with the projected position of the human body's center of mass. The deviation between the projected center of mass and the centroid of the supporting polygon is required to be no more than 40 mm and the deviation angle no more than 5 degrees. The supporting polygon is formed by connecting all effective support points, and its centroid coordinates are determined by the arithmetic mean of the coordinates of all support points.

[0123] The optimization of the support structure morphology employed an iterative adjustment method. After each iteration, the deviation between the centroid projection position and the support centroid was recalculated after adjusting the support point coordinates. Initially, the deviation was 52 mm, exceeding the constraint range. After adjusting the outer support point to move inward by 8 mm, the deviation was reduced to 35 mm, satisfying the balance constraint. The position coordinate parameter conversion mapped the key support point coordinates in the support structure morphology to the coordinates of the two foot centers and the distance between the feet. The left foot center coordinates were 210 mm X-axis and 200 mm Y-axis, and the right foot center coordinates were 210 mm X-axis and -200 mm Y-axis, with a foot distance of 400 mm. Posture angle parameters consisted of the trunk forward tilt angle, knee flexion angle, and ankle dorsiflexion angle. The optimized trunk forward tilt angle was 15 degrees, the knee flexion angle was 25 degrees, and the ankle dorsiflexion angle was 8 degrees.

[0124] Real-time motion data acquisition was achieved through an inertial measurement unit (IMU), fixed at the rescuer's third lumbar vertebra, recording triaxial acceleration and angular velocity information. During chest compressions, the peak Z-axis acceleration reached 18 meters per second squared, and the peak Y-axis angular velocity was 35 degrees per second. Muscle fatigue curve calculation was based on the temporal variation characteristics of electromyographic (EMG) signal amplitude. The initial EMG amplitude of the quadriceps femoris was 450 microvolts, increasing to 620 microvolts after 180 seconds of continuous work, a growth rate of 37.8%, corresponding to a moderate fatigue level. The EMG amplitude of the gastrocnemius muscle increased from an initial 380 microvolts to 540 microvolts, also reaching a moderate fatigue level. Postural stability was quantified by the offset range of the center of mass trajectory. The radius of the center of mass's projection on the horizontal plane was less than 30 millimeters in a stable state; when the radius increased to 45 millimeters, stability was considered to have decreased, requiring posture correction.

[0125] The dynamic balance constraint update process uses muscle fatigue curves and postural stability indices as input variables to correct the original balance constraint parameters. When fatigue reaches a moderate level, the allowable deviation distance between the centroid projection and the support centroid is increased from 40 mm to 55 mm, and the deviation angle limit is adjusted from 5 degrees to 7 degrees. When the stability index shows a centroid offset radius of 45 mm, it is determined that the support area needs to be expanded, and the foot spacing in the support point reconstruction parameters is increased by 50 mm, from the original 400 mm to 450 mm. At the same time, the Y-axis coordinates of some support points are adjusted: the outer heel support point is moved outward by 6 mm, and the inner forefoot support point is moved inward by 4 mm, increasing the area of ​​the reconstructed support polygon by 12%.

[0126] The posture correction command generates comprehensive support point reconstruction parameters, stance coordinate parameters, and posture angle parameters, and outputs specific action adjustment commands. The foot spacing adjustment command instructs the rescuer to move their right foot 25 mm outwards and their left foot 25 mm outwards as well; the trunk forward tilt angle adjustment command instructs the rescuer to reduce the forward tilt angle from 15 degrees to 12 degrees, reducing upper body load; the knee flexion angle adjustment command instructs the rescuer to increase the flexion angle from 25 degrees to 30 degrees, reducing hip joint pressure. The command sequence is organized chronologically, with the foot spacing adjustment command executed first, with an execution duration set to 2 seconds; the trunk angle adjustment command is initiated 0.5 seconds after the foot spacing adjustment is completed and lasts for 1.5 seconds; the knee angle adjustment command is performed synchronously with the trunk adjustment. The entire command sequence has a total duration of 4 seconds. Each command includes execution progress feedback parameters, and the execution effect is verified based on inertial measurement unit data. The adjustment is confirmed complete when the actual foot spacing reaches 445 mm.

[0127] In one optional implementation, the strain characteristics of the support point are calculated based on the potential energy gradient distribution, and the pressure transmission efficiency is analyzed, including:

[0128] Extract the principal direction vector of the potential energy gradient distribution, construct a pressure transmission path along the principal direction vector, and calculate the stress concentration factor of the support point on the pressure transmission path;

[0129] The local strain threshold of the support point is determined based on the stress concentration factor. The support points that exceed the strain threshold are classified by strain characteristics. The mapping relationship between the strain of each type of support point and the pressure transmission efficiency is calculated.

[0130] By combining the mapping relationship with the spatial distribution of support points, a pressure transmission network is constructed, and the overall pressure transmission efficiency is determined by calculating the pressure transmission network.

[0131] In one specific implementation, after acquiring the potential energy gradient distribution data of the support points, the extraction of the principal direction vector begins. A local coordinate system is established for each support point, with the support point as the origin. The potential energy gradient distribution data is projected onto three orthogonal directions of this coordinate system. The gradient magnitudes of the projected data are statistically analyzed, and the average gradient magnitude in the three orthogonal directions is calculated. The direction with the largest average gradient magnitude is selected as the principal direction vector of that support point. For example, if the average gradient magnitude of a support point is 0.8 MPa per meter in the X direction, 0.3 MPa per meter in the Y direction, and 0.5 MPa per meter in the Z direction, then the X direction is determined as the principal direction vector of that support point.

[0132] When constructing the pressure transmission path along the extracted principal direction vector, the path extends in three-dimensional space starting from the current support point, following the direction of the principal direction vector. A sampling node is set at a preset sampling interval of 5 mm. During the extension process, it is determined whether an adjacent support point is encountered. If an adjacent support point is detected within the search radius of the extension path, it is included in the pressure transmission path; the search radius is set to 10 mm. By traversing the principal direction vectors of all support points, multiple interconnected pressure transmission paths are established, forming a preliminary path network structure.

[0133] When calculating the stress concentration factor on a constructed pressure transmission path, stress tensor data is acquired for each sampling node along the path. The stress tensor contains six independent components, corresponding to three normal stress components and three shear stress components. The first invariant of the stress tensor, the sum of the three normal stress components, is calculated; this sum reflects the hydrostatic pressure state of the node. The second deviator invariant of the stress tensor is calculated; this invariant reflects the degree of shear deformation at the node. The stress concentration factor of the node is calculated by comparing the maximum principal stress value of the node with the average stress value of all nodes along the path. For example, if the maximum principal stress of a sampling node is 120 MPa and the average stress of the path is 80 MPa, then the stress concentration factor of that node is 1.5.

[0134] When determining the local strain threshold of support points based on the calculated stress concentration coefficient, a statistical analysis is performed on the stress concentration coefficients of all support points. The median and standard deviation of the stress concentration coefficient are calculated. The median value plus 1.5 times the standard deviation is used as the first strain threshold, and the median value plus 2.5 times the standard deviation is used as the second strain threshold. In a certain implementation scenario, the median value is 1.2 and the standard deviation is 0.4, so the first strain threshold is 1.8 and the second strain threshold is 2.2. Based on the comparison between the stress concentration coefficient of the support point and the two thresholds, support points with a stress concentration coefficient less than the first strain threshold are classified as low-strain characteristic support points, those with a stress concentration coefficient between the first and second strain thresholds are classified as medium-strain characteristic support points, and those with a stress concentration coefficient exceeding the second strain threshold are classified as high-strain characteristic support points.

[0135] When calculating the strain of various strain-characteristic support points after classification, displacement data of each support point before and after loading is obtained. The strain tensor of the support point is calculated using the displacement data. The strain tensor also contains six independent components. The equivalent strain value of the strain tensor is calculated, which comprehensively reflects the degree of deformation of the support point in three directions. For support points with low strain characteristics, the equivalent strain value is typically in the range of 0.001 to 0.005; for support points with medium strain characteristics, the equivalent strain value is typically in the range of 0.005 to 0.01; and for support points with high strain characteristics, the equivalent strain value typically exceeds 0.01.

[0136] When establishing the mapping relationship between strain and pressure conduction efficiency, pressure conduction efficiency is defined as the ratio of effective stress transmitted to input stress at the support point. For support points with low strain characteristics, the deformation is small, the internal lattice structure remains stable, and the pressure conduction efficiency is relatively high, typically in the range of 0.85 to 0.95. For support points with medium strain characteristics, the deformation reaches a certain level, and some energy is consumed by internal deformation, reducing the pressure conduction efficiency to the range of 0.7 to 0.85. For support points with high strain characteristics, the deformation is significant, and a large amount of energy is converted into plastic deformation energy, resulting in a significant decrease in pressure conduction efficiency to the range of 0.5 to 0.7. By fitting the pressure conduction efficiency data points corresponding to different strain ranges, a piecewise linear mapping relationship model is established.

[0137] When constructing a pressure transmission network by combining the mapping relationship with the spatial distribution of support points, each support point is treated as a network node, and the pressure transmission paths between adjacent support points are treated as network edges. Each edge is assigned a weight value, which is determined by the pressure transmission efficiency of the support points at both ends of the path and the path length. Specifically, the average pressure transmission efficiency of the support points at both ends of the path is taken, and then multiplied by the reciprocal of the path length to obtain the weight value of the edge. For example, if a path connects two support points with pressure transmission efficiencies of 0.8 and 0.9 respectively, and the path length is 20 mm, then the weight value of this edge is 0.0425. By traversing all support points and pressure transmission paths, a complete weighted pressure transmission network is established.

[0138] When calculating the overall pressure transmission efficiency using a pressure transmission network, the point where the external load is applied is selected as the source node, and the key support point on the final bearing surface is selected as the target node. A network analysis algorithm is used to search for the transmission path with the highest weight between the source and target nodes; this path represents the optimal pressure transmission path. The weighted product of all edges on the optimal path is calculated, and this product reflects the cumulative pressure transmission efficiency along that path. For multiple target nodes, their corresponding optimal paths and cumulative pressure transmission efficiencies are calculated separately. The cumulative pressure transmission efficiencies of all target nodes are then weighted and averaged according to their bearing areas to obtain the overall pressure transmission efficiency. In a certain implementation case, the cumulative pressure transmission efficiencies calculated for the three target nodes are 0.68, 0.72, and 0.65, respectively, corresponding to bearing areas of 100 square millimeters, 150 square millimeters, and 80 square millimeters, respectively. Therefore, the overall pressure transmission efficiency is 0.69.

[0139] The AI-based cardiopulmonary resuscitation (CPR) chest compression posture feature recognition and deviation correction system of this invention includes:

[0140] The data acquisition module is used to acquire data on the trunk movement trajectory and dynamic data on chest pressure when the rescuer performs chest compressions.

[0141] The support area optimization module is used to spatially reconstruct the torso motion trajectory data, extract the torso center of mass motion law and angular velocity change trend, calculate the stability components of the torso posture in the vertical and horizontal planes, determine the optimal force application range and force support point, and generate the rescuer's steady-state support area.

[0142] The pressure path analysis module is used to construct a pressure response state space from the dynamic data of the thoracic pressure, extract the dynamic characteristics of the pressure action sequence, establish a pressure elastic deformation field, analyze the energy dissipation distribution of the conduction path, and generate the optimal pressure application path.

[0143] The attitude correction module is used to map the steady-state support area and the optimal pressure application path to the elastic potential energy field, optimize the support structure morphology based on dynamic equilibrium constraints, and generate a real-time attitude adjustment command sequence based on fatigue and stability indicators.

[0144] The action execution module is used to perform action correction according to the posture adjustment command sequence, adjust the rescuer's standing distance, support angle and force direction, and complete the correction of the compression action deviation.

[0145] A third aspect of the present invention provides an electronic device, comprising:

[0146] processor;

[0147] Memory used to store processor-executable instructions;

[0148] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0149] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0150] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for AI recognition and deviation correction of cardiopulmonary resuscitation (CPR) chest compression posture features, characterized in that, include: Acquire data on trunk movement trajectory and dynamic chest pressure during chest compressions performed by the rescuer; Spatial reconstruction is performed on the trunk motion trajectory data to extract the trunk center of mass motion law and angular velocity change trend, calculate the stability components of the trunk posture in the vertical and horizontal planes, determine the optimal force application range and force support point, and generate the rescuer's steady-state support area. A pressure response state space is constructed from the dynamic data of intrathoracic pressure, the dynamic characteristics of the pressure action sequence are extracted, a pressure elastic deformation field is established, the energy dissipation distribution of the conduction path is analyzed, and the optimal pressure application path is generated. The steady-state support region and the optimal pressure application path are mapped to the elastic potential energy field. The support structure morphology is optimized based on dynamic equilibrium constraints, and a real-time correction attitude adjustment command sequence is generated based on fatigue and stability indices. Based on the posture adjustment command sequence, the action correction is performed to adjust the rescuer's standing distance, support angle, and force direction, thereby completing the correction of the compression action deviation.

2. The method according to claim 1, characterized in that, Spatial reconstruction of the trunk motion trajectory data, and extraction of the trunk center of mass motion patterns and angular velocity variation trends, including: A three-dimensional coordinate system centered on the rescuer's torso is established, and the torso motion trajectory data is mapped to the three-dimensional coordinate system to obtain three-dimensional spatial motion data. The motion boundary is corrected by combining the range of motion of the human torso joints to obtain the corrected torso motion characteristics. The corrected trunk motion features are decomposed using a quaternion rotation matrix to obtain the main direction motion component and the auxiliary direction motion component. The main direction motion component and the auxiliary direction motion component are projected and combined to obtain the trunk center of mass motion trajectory. By performing time-domain analysis on the trunk center of mass motion trajectory, the trunk center of mass velocity change is obtained. Combined with displacement compensation, the trunk center of mass motion law is generated. The motion law of the torso's center of mass is processed using a spherical linear interpolation algorithm. The rotational features of the torso are extracted to calculate the three-axis rotation angles and determine the torso posture change sequence. The angle differentiation of the torso posture change sequence generates the angular velocity change trend.

3. The method according to claim 1, characterized in that, Calculate the stability components of the torso posture in the vertical and horizontal planes, determine the optimal force application range and force support points, and generate the rescuer's steady-state support region, including: The motion law of the torso's center of mass is decomposed into vertical and horizontal planes to obtain the planar projection trajectory of the torso's center of mass; the trend of angular velocity change is decomposed into vertical and horizontal planes to obtain the planar projection characteristics of the torso's angular motion. Amplitude analysis is performed on the plane projection trajectory of the torso's center of mass to extract peak and valley points, and the fluctuation amplitude and time interval between adjacent peak and valley points are calculated to obtain the center of mass motion fluctuation characteristics. The fluctuation period mapping relationship is determined, and the center of mass motion stability characteristics are extracted. The plane projection characteristics of the torso's angular motion are periodically segmented to extract angular motion stability characteristics. The center of mass motion stability characteristics and the angular motion stability characteristics are weighted and combined to generate vertical plane stability components and horizontal plane stability components. Threshold segmentation is performed on the vertical plane stability component and the horizontal plane stability component to extract the time period with the optimal stability index and determine it as the optimal stability interval of the torso posture. The torso posture stabilization point is determined based on the motion characteristics within the optimal stability interval of the torso posture. A polygonal envelope is constructed with the torso posture stabilization point as the center to generate the rescuer's steady-state support region.

4. The method according to claim 1, characterized in that, A pressure response state space is constructed from the dynamic data of intrathoracic pressure, the dynamic characteristics of the pressure application sequence are extracted, a pressure elastic deformation field is established, the energy dissipation distribution of the conduction path is analyzed, and the optimal pressure application path is generated, including: Dynamic data of intrathoracic pressure are mapped to a multidimensional state space, the phase space trajectory of the pressure state vector is extracted, the stable interval is determined according to the topological structure of pressure fluctuations, and the pressure data is nonlinearly reconstructed to obtain a single-cycle pressure action sequence. Pressure conduction analysis is performed on the single-cycle pressure action sequence to extract the pressure gradient and propagation velocity at each sampling point. The propagation direction of the pressure wave in the thoracic tissue is determined based on the pressure gradient. The displacement vector of the pressure action point is calculated along the propagation direction, and temporal correlation analysis is performed to obtain the spatial position change sequence of the pressure action point. A pressure elastic deformation field is constructed based on the spatial position change sequence. The propagation path of the pressure wave in the thoracic cavity is determined based on the pressure elastic deformation field to obtain the pressure conduction region. The pressure attenuation coefficient between adjacent sampling points is calculated within the pressure transmission region to determine the energy loss characteristics of the pressure wave during transmission. The energy dissipation distribution along the transmission path is obtained through cumulative calculation. Based on the energy dissipation distribution, a transmission efficiency evaluation index is constructed. Multiple transmission paths are sorted and screened according to the transmission efficiency evaluation index to determine the transmission path with the minimum energy dissipation and generate the optimal pressure application path.

5. The method according to claim 4, characterized in that, Dynamic data of intrathoracic pressure are mapped to a multidimensional state space, the phase space trajectory of the pressure state vector is extracted, the stable interval is determined based on the topological structure of pressure fluctuations, and a single-cycle pressure action sequence is obtained by nonlinear reconstruction of the pressure data, including: Dynamic data of intrathoracic pressure are mapped to a multidimensional state space to construct a phase matrix of pressure response. A time-delay embedding transformation is performed on the phase matrix to generate a pressure state vector. Calculate the distance matrix between adjacent state points in the pressure state vector, transform the distance matrix into a state transition diagram, extract the connected components of the state transition diagram, obtain the invariant set in the state space, and use the invariant set to map the phase space trajectory of pressure changes to form the topology of pressure fluctuation. The topology is matrix decomposed to extract the structural parameters of the attractor, the convergence domain boundary of the structural parameters is determined, critical state points are marked in the convergence domain boundary, the stable interval of pressure fluctuation is divided according to the critical state points, and the pressure data in the stable interval is nonlinearly reconstructed to generate a single-cycle pressure action sequence.

6. The method according to claim 1, characterized in that, Mapping the steady-state support region and the optimal pressure application path to an elastic potential energy field, optimizing the support structure morphology based on dynamic equilibrium constraints, and generating a real-time correction attitude adjustment command sequence based on fatigue and stability indices include: The coordinate sequence and pressure value sequence of the support points in the steady-state support area are collected. The potential energy value sequence is obtained by elastic potential energy calculation. The elastic potential energy field is constructed using the potential energy value sequence. The optimal pressure application path is mapped to the elastic potential energy field to obtain the potential energy gradient distribution. The strain characteristics of the support point are calculated based on the potential energy gradient distribution, the pressure transmission efficiency is analyzed, a dynamic reconstruction sequence of the support point is generated based on the pressure transmission efficiency, the dynamic reconstruction sequence of the support point is combined with the preset human biomechanical parameters to construct dynamic equilibrium constraints, the support structure morphology is optimized based on the dynamic equilibrium constraints, and the support structure morphology is converted into position coordinate parameters and posture angle parameters. Real-time motion data of the rescuer is collected, and muscle fatigue curves and posture stability indices are calculated and determined. The muscle fatigue curves and posture stability indices are input into dynamic balance constraints to update the support point reconstruction parameters. Based on the support point reconstruction parameters, station coordinate parameters, and posture angle parameters, posture correction commands are generated. The posture correction commands are organized in time sequence to obtain a real-time posture adjustment command sequence.

7. The method according to claim 6, characterized in that, Based on the potential energy gradient distribution, the strain characteristics of the support point were calculated, and the pressure transmission efficiency was analyzed, including: Extract the principal direction vector of the potential energy gradient distribution, construct a pressure transmission path along the principal direction vector, and calculate the stress concentration factor of the support point on the pressure transmission path; The local strain threshold of the support point is determined based on the stress concentration factor. The support points that exceed the strain threshold are classified by strain characteristics. The mapping relationship between the strain of each type of support point and the pressure transmission efficiency is calculated. By combining the mapping relationship with the spatial distribution of support points, a pressure transmission network is constructed, and the overall pressure transmission efficiency is determined by calculating the pressure transmission network.

8. A cardiopulmonary resuscitation (CPR) chest compression posture feature AI recognition and deviation correction system, used to implement the method of any one of claims 1-7, characterized in that, include: The data acquisition module is used to acquire data on the trunk movement trajectory and dynamic data on chest pressure when the rescuer performs chest compressions. The support area optimization module is used to spatially reconstruct the torso motion trajectory data, extract the torso center of mass motion law and angular velocity change trend, calculate the stability components of the torso posture in the vertical and horizontal planes, determine the optimal force application range and force support point, and generate the rescuer's steady-state support area. The pressure path analysis module is used to construct a pressure response state space from the dynamic data of the thoracic pressure, extract the dynamic characteristics of the pressure action sequence, establish a pressure elastic deformation field, analyze the energy dissipation distribution of the conduction path, and generate the optimal pressure application path. The attitude correction module is used to map the steady-state support area and the optimal pressure application path to the elastic potential energy field, optimize the support structure morphology based on dynamic equilibrium constraints, and generate a real-time attitude adjustment command sequence based on fatigue and stability indicators. The action execution module is used to perform action correction according to the posture adjustment command sequence, adjust the rescuer's standing distance, support angle and force direction, and complete the correction of the compression action deviation.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images