Self and mutual rescue examination intelligent scoring system and scoring method

Abstract

This invention discloses an intelligent scoring system and method for self-rescue and mutual-rescue assessments. The scoring system includes a wearable motion capture module for real-time acquisition of acceleration, angular velocity, and attitude angle data of various parts of the human body; a 3D somatosensory depth camera and infrared reflective markers for capturing three-dimensional spatial coordinate information; a data fusion and attitude estimation module for inertial data attitude estimation and calculation of joint angles of various joints of the human body; a feature extraction and action recognition module for identifying the current action type and the quality of action completion; and an intelligent scoring and feedback module for automatically scoring the identified action sequence based on preset scoring items and generating scoring results, while outputting scoring details and action optimization feedback. The scoring system and method described in this invention can achieve unified scoring standards, accurate feedback, adaptability to large-scale training and assessments, and adaptability to complex emergency rescue scenarios.

Description

Technical Field

[0001] This invention relates to the field of intelligent processing device technology, specifically to an intelligent scoring system and scoring method for self-rescue and mutual rescue assessment. Background Technology

[0002] In scenarios involving sudden disasters and accidents, the "golden ten minutes" window for injury and illness treatment is crucial, directly determining the survival rate and subsequent rehabilitation outcomes of the injured and sick. This necessitates that emergency rescue personnel, grassroots medical staff, and various related personnel be proficient in standardized self-rescue and mutual-rescue skills, and that their skill attainment be verified through scientific assessments. Self-rescue and mutual-rescue operations in complex scenarios such as the wilderness, accident scenes, and tunnels require both standardized procedures and precise movements, placing extremely high demands on the objectivity, timeliness, and standardization of assessments. Therefore, various emergency rescue systems, medical institutions, and grassroots units are actively exploring technical pathways that integrate training with intelligent assessment.

[0003] Currently, self-rescue and mutual rescue assessments are still primarily based on manual scoring, with professional evaluators assigning scores according to established standards. While intelligent technologies such as motion capture and data fusion are being applied in fields such as sports training and industrial testing, a feasible "motion collection-feature analysis-precise scoring" solution has not yet been developed for self-rescue and mutual rescue assessment scenarios. Existing technologies can mostly only record actions and lack professional scoring algorithms and systems adapted to the operational logic of self-rescue and mutual rescue.

[0004] The existing manual scoring model has several significant shortcomings, making it difficult to meet the needs of standardized and large-scale assessments: First, the scoring lacks objectivity. Assessment results are easily influenced by factors such as the professional level, subjective preferences, and psychological state of the assessors. Inconsistency in scoring between different assessors or even the same assessor at different times is poor, which can easily lead to disputes, especially in scenarios with extremely high requirements for fairness, such as skills competitions and qualification certifications. Second, the feedback efficiency is low. Manual scoring requires real-time on-site monitoring and recording, which cannot accurately capture subtle deviations in actions, make it difficult to quantify and analyze technical shortcomings, and lack data support for post-training debriefing. This hinders trainees from quickly improving their skills and is also unsuitable for large-scale batch assessments. Third, the scenario adaptability is weak. In complex scenarios, manual assessment has a limited monitoring range and insufficient recording accuracy. Existing simple intelligent devices lack multi-sensor collaborative calibration capabilities, making it difficult to guarantee data reliability and failing to meet the actual needs of emergency rescue assessments. Summary of the Invention

[0005] In view of this, the present invention provides an intelligent scoring system and method for self-rescue and mutual rescue assessment. Combining the development of modern computer image recognition technology, it utilizes motion capture algorithms to identify the basic actions and key medical treatment actions of medical personnel or rescuers in self-rescue and mutual rescue activities. It constructs a system-based and data-based self-rescue and mutual rescue assessment scoring system, achieving unified scoring standards and accurate feedback. It can be adapted to large-scale training and assessment and has adaptability to complex emergency rescue scenarios, thus overcoming the shortcomings of existing manual scoring models, such as low feedback quality and efficiency, poor objectivity, and limited application scenarios.

[0006] The specific technical solution of the present invention is as follows:

[0007] 1. A self-rescue and mutual-rescue assessment intelligent scoring system, comprising:

[0008] The wearable motion capture module includes multiple inertial measurement units (IMUs) for real-time acquisition of acceleration, angular velocity, and attitude angle data of various parts of the human body.

[0009] The infrared vision assistance module includes a 3D motion-sensing depth camera and an infrared reflective marker. The infrared transmitting and receiving components of the 3D motion-sensing depth camera capture the three-dimensional spatial coordinate information of the infrared reflective marker.

[0010] The data fusion and attitude estimation module is used to receive high-frequency inertial data output by the wearable motion capture module and position data output by the infrared vision assistance module, use a filtering algorithm to fuse the two types of data, perform attitude estimation based on the fused inertial data, and calculate the joint angles of each joint of the human body.

[0011] The feature extraction and action recognition module is used to process the multi-channel time series data output by the data fusion and pose estimation module, extract motion features and construct a multi-dimensional motion feature vector, calculate the similarity between the multi-dimensional motion feature vector and the feature template in the standard action database, and identify the current action type and the quality of action completion.

[0012] The intelligent scoring and feedback module is used to automatically score the identified action sequences based on preset scoring items and generate scoring results, while outputting scoring details and action optimization feedback.

[0013] Furthermore, it also includes:

[0014] The data management and visualization module is configured with a PC-based software interface for data storage, display of scoring results, historical data query, statistical analysis, and training progress tracking.

[0015] Furthermore, the IMU is used to acquire limb activity data, which includes at least one of the following: triaxial acceleration measurement, triaxial angular velocity measurement, and triaxial magnetic field strength measurement.

[0016] Furthermore, the scoring items in the intelligent scoring and feedback module include the correctness of the operation method, the standardization of the action, the overall effect, and the time efficiency.

[0017] 2. A self-rescue and mutual-rescue assessment intelligent scoring method, comprising the following steps:

[0018] S1) Collect N standard sub-operations in self-rescue and mutual rescue scenarios, construct a standard action database and system scoring standards, and obtain the standard score corresponding to each standard sub-operation. (i=1,2,...,N);

[0019] S2) The IMU and infrared vision-assisted module are used to collect real-time motion data of the test subject, specifically including:

[0020] S21) System initialization: Perform time synchronization, spatial synchronization and zero-bias estimation to achieve spatiotemporal benchmark unification of multimodal data;

[0021] S22) Data fusion: A loosely coupled, low-computing-power fusion method is adopted to fuse the inertial data collected by the IMU with the position data output by the infrared vision auxiliary module in real time to calculate the limb position information;

[0022] S23) Micro-motion enhancement: For the minute motion amplitude in self-rescue and mutual rescue operations, a multi-scale wavelet transform and adaptive gain amplification mechanism are used to separate and enhance the micro-motion features.

[0023] S24) Data Output: Based on the solution results of step S22) and the micro-motion feature enhancement results of step S23), the quaternion Q calculated by the IMU is output. t Convert to Euler angles form and enhance the micro-motion characteristics Encapsulated into a data packet, serving as the basis for subsequent scoring;

[0024] S3) Sequential action recognition and logical scoring: Perform action segmentation, sub-operation recognition, timing verification and comprehensive scoring on the data packet output in step S2) to obtain the final assessment score.

[0025] Furthermore, in step S21),

[0026] Time synchronization uses the master control device as the sole clock source, and through periodic broadcasting and delayed response, the synchronization error between the IMUs deployed in various parts of the human body and the system timestamp of the master control device is ≤10ms.

[0027] Spatial synchronization requires the subject to maintain a "T-shaped" standing posture for 5 seconds. The system calculates the gravitational acceleration vector and the geomagnetic north pole vector output by the IMU to establish the initial rotation matrix from the sensor coordinate system to the world navigation coordinate system. At the same time, using the position of the reflective point captured by the infrared vision auxiliary module, the rigid body transformation relationship from the visual coordinate system to the world navigation coordinate system is established to complete the spatial registration of multimodal data. Zero bias estimation uses no less than 300 frames of gyroscope output data collected during the static period to calculate and store the static zero bias of each axis.

[0028] Furthermore, in step S22), the system performs data processing cyclically at a frequency of 120Hz, and performs the following operations sequentially for each time step t:

[0029] 1) Inertial attitude acquisition: Directly read the quaternion output by the IMU. This represents the current inertial posture of the limbs.

[0030] 2) Inertial position calculation: Obtain the triaxial linear acceleration output by the IMU. The acceleration is obtained by transforming the current attitude to the navigation coordinate system and subtracting the gravitational component g. ,right Double integration yields displacement calculated using pure inertia. The calculation formula is: ,in, The position at time t-1 Let be the velocity at time t-1, and Δt be the time step interval;

[0031] 3) Infrared position correction: via marker position Determine whether the infrared vision assistance module has captured a valid marker point:

[0032] when At the same time, high-precision position measured by infrared vision Covering the inertial estimation location, i.e. Simultaneously calculate and cache the integral drift compensation amount. ;

[0033] when At that time, use the most recent effective drift compensation amount Correcting the inertial estimation position, the output position is .

[0034] Furthermore, step S23) includes the following specific operations:

[0035] (1) Multi-scale detail separation: The raw inertial data stream (angular velocity, linear acceleration) acquired by the IMU is subjected to discrete wavelet transform, and the signal is decomposed into approximation coefficients and detail coefficients using the Daubechies wavelet basis, satisfying:

[0036] ,

[0037] in, This is the approximate component of the Jth layer, corresponding to large-amplitude limb movements; For the j-th layer detail component, it corresponds to micro-movements such as finger tremors and end-effector fine adjustments;

[0038] (2) Adaptive micro-motion feature amplification: targeting the extracted micro-motion signals Construct a nonlinear gain amplifier and define a dynamic threshold. ;when When the feature is amplified by an exponential gain function, the formula is:

[0039] ,

[0040] in These are gain control parameters, and All > 0;

[0041] (3) Extraction of fingertip / end micro-motion features: The IMU worn on the wrist is used to capture high-frequency micro-vibrations generated by tendon traction. The energy features of the 15-30Hz frequency band are extracted by spectrum analysis as a proxy indicator of finger movement activity.

[0042] Furthermore, step S3) includes the following specific operations:

[0043] S31) Adaptive Action Segmentation: A sliding time window is used to slice the continuous data stream. The window length W and step size S are set, and the cosine similarity of the feature vectors of adjacent windows is calculated. When the similarity is lower than a threshold, it is determined as an action switching point, and the continuous data stream is divided into a set of discrete action segments. ;

[0044] S32) Sub-operation recognition and individual scoring: This involves identifying and scoring action segments. The feature vectors are projected onto the standard sub-operation feature subspace constructed in step S1), and the Euclidean or Mahalanobis distance is calculated; the sub-operation type is identified based on the minimum distance principle. And generate sub-operation quality scores based on Gaussian mapping functions. D i S is the feature distance. max The full score is given by σ, where σ is the distance coefficient.

[0045] S33) Sequential logic verification: Introduce a state transition matrix model to define an ordered state sequence for the standard operation procedure. and the allowed state transition probability matrix Maintain a real-time state pointer; when a new sub-operation type is identified... At that time, the verification starts from arrive Is the transfer in If the statement is valid, a penalty will be deducted based on the logic. and to Multiply by the penalty coefficient β (0 < β < 1);

[0046] S34) Weighted Comprehensive Score: Final Assessment Score The weighted sum of the scores for each sub-operation is calculated using the following formula:

[0047] ,

[0048] Among them, w k As the importance weight of sub-operations, As a macroscopic action quality score, Add points to the micro-motions calculated in step S23). The sequential logic coefficients calculated in step S33) Additional points will be deducted for redundant / ineffective actions.

[0049] The beneficial effects of this invention are as follows:

[0050] (1) Significantly improved objectivity of scoring: Automatic scoring based on quantitative data from wearable device inertial sensors eliminates the influence of subjective factors and psychological biases. The scoring standards are unified and the results are reproducible, ensuring consistency and fairness of scoring for different assessees and at different times. It is suitable for scenarios with strict requirements for fairness, such as skills competitions and qualification certifications.

[0051] (2) The quality and efficiency of feedback are greatly improved: high-frequency sampling fully records the motion state data of the entire operation, accurately locates the moment of erroneous action, quantitatively analyzes parameters such as action deviation and speed abnormality, intuitively presents the differences in operation through visual comparison charts, provides targeted improvement suggestions, shortens the skill improvement cycle, and is suitable for large-scale training.

[0052] (3) Significantly improved overall training efficiency: Wireless wearable devices can monitor multiple examinees at the same time, infrared cameras can be flexibly deployed to achieve multi-point monitoring, the scoring process is fully automated, and the entire process from motion capture to score generation can be completed in just a few seconds, greatly reducing labor costs.

[0053] (4) Data accumulation supports precision training: Automatically establish a skill profile database, fully store training data, movement characteristics, scoring details and other information, identify common problems and individual differences through historical data mining, provide a scientific basis for formulating targeted training plans, and form a virtuous cycle of "data-driven training improvement".

[0054] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0055] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0056] Figure 1 This is a schematic diagram of the modules of the intelligent scoring system for self-rescue and mutual rescue assessment shown in the embodiment.

[0057] Figure 2 This is a flowchart illustrating the intelligent scoring system for self-rescue and mutual rescue assessment, as shown in the embodiment. Detailed Implementation

[0058] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should also be noted that technical means not described in detail in this invention can be implemented using conventional technical means.

[0059] It should also be noted that the flowcharts shown in the attached figures are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily need to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0060] Reference Figure 1 One embodiment of the present invention discloses an intelligent scoring system for self-rescue and mutual rescue assessment, comprising:

[0061] The wearable motion capture module includes multiple inertial measurement units (IMUs) for real-time acquisition of acceleration, angular velocity, and attitude angle data from various parts of the human body. The IMUs can be worn on key body parts such as the head, torso, upper arm, forearm, wrist, thigh, and calf. Each sensor node incorporates a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer to acquire real-time acceleration, angular velocity, and attitude angle data for each part. The sensor nodes transmit data to a data processing center via Bluetooth. The gyroscope can be a fiber optic gyroscope (FOG) or a microelectromechanical system (MEMS). Flexible angle sensors (such as strain gauges or potentiometer-type angle sensors) can also be used at key joints to directly measure joint bending angles, in conjunction with the IMU sensors.

[0062] The infrared vision assistance module includes a 3D motion-sensing depth camera and an infrared reflective marker. The infrared transmitter and receiver components of the 3D motion-sensing depth camera capture the three-dimensional spatial coordinate information of the infrared reflective marker. For example, it consists of a Kinect infrared depth camera and an infrared reflective marker. The infrared reflective marker is pasted at key locations in the training area, and the Kinect camera is placed at an appropriate location in the training area. The infrared transmitter and receiver capture the three-dimensional spatial coordinate information of the infrared marker.

[0063] The data fusion and attitude estimation module is used to receive high-frequency inertial data output by the wearable motion capture module and position data output by the infrared vision assistance module, use a filtering algorithm to fuse the two types of data, perform attitude estimation based on the fused inertial data, and calculate the joint angles of each joint of the human body.

[0064] The feature extraction and action recognition module is used to process the multi-channel time series data output by the data fusion and pose estimation module, extract motion features and construct a multi-dimensional motion feature vector, calculate the similarity between the multi-dimensional motion feature vector and the feature template in the standard action database, and identify the current action type and the quality of action completion.

[0065] The intelligent scoring and feedback module is used to automatically score the identified action sequences based on preset scoring items and generate scoring results, while outputting scoring details and action optimization feedback; scoring items may include the correctness of operation method, action standardization, overall effect and time efficiency.

[0066] The intelligent scoring system for self-rescue and mutual-rescue assessments may also include:

[0067] The data management and visualization module is configured with a PC-based software interface for data storage, display of scoring results, historical data query, statistical analysis, and training progress tracking.

[0068] Another embodiment of the present invention provides an intelligent scoring method for self-rescue and mutual rescue assessment. First, a scoring standard is established using a sufficient sample size of standard actions. Then, a scoring mechanism based on statistical learning and manifold learning is adopted to measure the statistical distance between the assessment subject's actions and the distribution of standard actions, and finally, the assessment score is obtained.

[0069] The specific process is as follows:

[0070] S1) Collect standard sub-operations, construct a standard action database, and establish scoring criteria. For example, assuming there are a total of n assessment items, which are n sub-operations, these n sub-operations form a sub-operation set. Each sub-operation establishes its own scoring criteria and obtains n standard scores. .

[0071] S2) Collect real-time action data of the person being assessed, such as Figure 2 The flowchart illustrates how high-precision attitude information of the subject is acquired through multi-sensor fusion technology. Specifically, a loosely coupled fusion strategy is adopted, where "IMU (Inertial Measurement Unit) dominates attitude and infrared vision assists in position correction." The system directly utilizes the AHRS (Attitude and Heading Reference System) algorithm integrated within the IMU to obtain high-frequency attitude quaternions, including the following steps:

[0072] Step S21) Initialize and multi-dimensionally align the sensor network.

[0073] First, high-precision time synchronization is performed. The wearable IMU nodes are connected to the main control computer of the infrared vision auxiliary module via a wireless local area network. The system adopts the IEEE 1588V2 Precision Time Protocol (PTP), with the main control computer as the sole clock source. Through periodic broadcasting and delayed response mechanisms, the IMU nodes distributed on various parts of the body are synchronized with the system timestamps of the vision host. ;

[0074] Next, spatial coordinate alignment is performed, requiring the examinee to maintain a static "T-shaped" standing posture. During this period, the system calculates the gravitational acceleration vector output by the IMU. Using the geomagnetic north pole vector, establish the initial rotation matrix from the sensor coordinate system to the world navigation coordinate system. Meanwhile, by using the positions of reflective points captured by the infrared vision system, a rigid body transformation relationship from the visual coordinate system to the navigation coordinate system is established to complete the spatial registration of multimodal data.

[0075] Finally, zero-bias estimation is performed. Using at least 300 frames of gyroscope output data collected during the stationary period, the static zero-bias of each axis is calculated and stored. This is used for initial compensation in subsequent integration operations.

[0076] Step S22) Perform loosely coupled complementary correction based on location anchor points.

[0077] During real-time assessment, the system performs data processing cyclically at a frequency of 120Hz, executing the following logic for each time step t:

[0078] First, the inertial attitude is directly acquired by the system, which directly reads the quaternion output from the wearable IMU. As the body posture at the current moment, that is ;

[0079] Then, the inertial position is integrated and calculated, and the system obtains the triaxial linear acceleration output by the IMU. Using the current attitude Q t Transform it to the navigation coordinate system and subtract the gravitational component g to obtain the acceleration. ,right By performing double integration, the displacement calculated using pure inertia is obtained. Its calculation formula satisfies:

[0080] ;

[0081] Finally, infrared position correction (position reset) is performed, and the system determines whether the infrared vision system has captured a valid marker point (i.e., a flag bit). ),when When the marker point is visible, high-precision positioning is directly measured using infrared vision. Covering the inertial estimation location, i.e. Simultaneously, calculate the "integral drift compensation" at the current moment. (defined as) ), and update the cache; when When the marker is in an obstructed or blind spot, the most recent effective drift compensation amount is used. The inertial estimation position is corrected, and the output position is... Through the above logic, the system achieves millimeter-level absolute positioning accuracy when infrared is visible, and maintains smooth trajectory by using inertial short-time calculation when infrared is blocked, thus realizing highly robust tracking with low computing power cost.

[0082] Step S23) Multi-scale detail enhancement and feature magnification for minute motion amplitudes

[0083] First aid procedures involve numerous minute movements such as "checking for a carotid pulse," "checking for breathing," or "meticulous bandaging and knotting." These movements have a low signal-to-noise ratio and are easily drowned out by sensor background noise or masked by macroscopic movements. To accurately capture these movements, this step introduces a multi-scale wavelet transform and adaptive gain amplification mechanism. The specific steps are as follows:

[0084] (1) Multi-scale detail separation based on wavelet transform: The system performs Discrete Wavelet Transform (DWT) on the original inertial data stream (especially angular velocity and linear acceleration), and uses the Daubechies wavelet basis to decompose the signal into approximation coefficients and detail coefficients.

[0085]

[0086] in, This is the approximate component of the Jth layer, corresponding to a large displacement of the limb; The j-th layer detail component corresponds to micro-movements such as finger tremors and end-effector fine-tuning.

[0087] (2) Adaptive micro-motion feature amplification: targeting the extracted micro-motion signals Construct a nonlinear gain amplifier and define a dynamic threshold. When the signal amplitude At that time, the exponential gain function is applied for feature amplification:

[0088] ,

[0089] in The gain control parameter is used; this mechanism significantly enhances the saliency of effective micro-motions without amplifying background noise.

[0090] (3) Extraction of fingertip / end-point micro-motion features: For finger parts where sensors cannot be directly worn, an IMU worn on the wrist is used to capture high-frequency micro-vibrations caused by tendon traction. Energy features in the 15-30Hz frequency band are extracted through spectrum analysis as a proxy indicator of "finger movement activity". For example, in the "check pulse" action, the wrist remains still, but pressure on the fingertip will cause micro-vibrations at a specific frequency. The system uses this to determine whether the action has been performed correctly.

[0091] Step S24) Output a continuous whole-body pose data stream and enhanced features.

[0092] Based on the solution result of step S22) and the micro-motion feature enhancement of step S23), the system will calculate the quaternion Q. t Convert to Euler angles form and enhance the micro-motion characteristics It is encapsulated in a data packet and used as the basis for subsequent scoring.

[0093] S3) Perform temporal sub-action recognition and logical scoring on the collected action data.

[0094] Given that emergency rescue actions consist of multiple non-independent sub-operations and have strict timing requirements, this step adopts a comprehensive evaluation strategy of "segmentation identification - logical verification - weighted scoring". The specific implementation steps are as follows:

[0095] Step S31) Construction of multidimensional motion feature matrix and adaptive segmentation

[0096] The system employs a sliding time window mechanism to slice the continuous data stream. It sets the window length W and step size S, calculates the cosine similarity of feature vectors from adjacent time windows, and identifies action switching points when the similarity falls below a preset abrupt change threshold. Based on this, the continuous data stream is divided into a series of discrete action segment sequences. Each segment represents a potential sub-operation;

[0097] Step S32) Sub-operation recognition and individual scoring based on subspace projection

[0098] The system predefines standard sub-operation sets (i.e., steps S1) for each emergency rescue task in the standard database. For each real-time action segment... The system projects its feature vectors onto the feature subspace of each standard suboperation and calculates the Euclidean or Mahalanobis distance:

[0099] First, the sub-operation type to which the segment belongs is identified based on the minimum distance principle. Secondly, calculate the feature distance D between the segment and the corresponding standard template. i And generate a single quality score for this sub-operation based on the Gaussian mapping function. This step enables a quantitative evaluation of individual sub-operations.

[0100] Step S33) Operation timing logic verification based on state transition matrix

[0101] To evaluate the correctness of the order of operations of the n sub-operations in step S1), the system introduces a State Transition Matrix model, defining the standard operation flow as an ordered sequence of states. And construct the allowed state transition probability matrix. The system maintains a real-time status pointer; when a new sub-operation type is identified in step S32). At that time, the system checks from the previous state To the current state Is the transfer in Medium to legal.

[0102] -Legitimate transition: If the transition from a "rotate" action to a "fixed" action is logical, the quality score of that sub-operation is retained. .

[0103] - Illegal transfer: Performing the "fix" action directly without first performing the "rotate" action (missing step), or performing "fix" before "rotate" (reversed order). The system will trigger a logical penalty mechanism and record the penalty points. And may affect the quality score of the current sub-operation. Perform weight reduction (e.g., multiply by a penalty coefficient) ).

[0104] Step S34) Construct a global comprehensive scoring model based on time-series weights.

[0105] Final assessment results Instead of a simple average, the system uses a weighted sum that combines the importance of sub-operations, timing logic, and operation quality. The system assigns a weight w to each standard sub-operation. k (Key steps have high weight), the calculation formula is as follows:

[0106] ,in, Assign a score to the quality of macroscopic actions; Add points to the micro-actions calculated in step S23); The sequential logic coefficients calculated in step S33); Additional points are deducted for redundant or ineffective actions. Through this model, the system can achieve intelligent, logical, and precise scoring of the entire emergency rescue operation process.

[0107] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A self-rescue and mutual-rescue assessment intelligent scoring system, characterized in that, include: The wearable motion capture module includes multiple inertial measurement units (IMUs) for real-time acquisition of acceleration, angular velocity, and attitude angle data of various parts of the human body. The infrared vision assistance module includes a 3D motion-sensing depth camera and an infrared reflective marker. The infrared transmitting and receiving components of the 3D motion-sensing depth camera capture the three-dimensional spatial coordinate information of the infrared reflective marker. The data fusion and attitude estimation module is used to receive high-frequency inertial data output by the wearable motion capture module and position data output by the infrared vision assistance module, use a filtering algorithm to fuse the two types of data, perform attitude estimation based on the fused inertial data, and calculate the joint angles of each joint of the human body. The feature extraction and action recognition module is used to process the multi-channel time series data output by the data fusion and pose estimation module, extract motion features and construct a multi-dimensional motion feature vector, calculate the similarity between the multi-dimensional motion feature vector and the feature template in the standard action database, and identify the current action type and the quality of action completion. The intelligent scoring and feedback module is used to automatically score the identified action sequences based on preset scoring items and generate scoring results, while outputting scoring details and action optimization feedback.

2. The intelligent scoring system for self-rescue and mutual rescue assessment as described in claim 1, characterized in that, Also includes: The data management and visualization module is configured with a PC-based software interface for data storage, display of scoring results, historical data query, statistical analysis, and training progress tracking.

3. The intelligent scoring system for self-rescue and mutual rescue assessment as described in claim 1, characterized in that, The IMU is used to acquire limb activity data, which includes at least one of the following: triaxial acceleration measurement, triaxial angular velocity measurement, and triaxial magnetic field strength measurement.

4. The intelligent scoring system for self-rescue and mutual rescue assessment as described in claim 1, characterized in that, The scoring items in the intelligent scoring and feedback module include the correctness of the operation method, the standardization of the action, the overall effect, and the time efficiency.

5. A self-rescue and mutual-rescue assessment intelligent scoring method, characterized in that, Includes the following steps: S1) Collect N standard sub-operations in self-rescue and mutual rescue scenarios, construct a standard action database and system scoring standards, and obtain the standard score corresponding to each standard sub-operation. (i=1,2,...,N); S2) The IMU and infrared vision-assisted module are used to collect real-time motion data of the test subject, specifically including: S21) System initialization: Perform time synchronization, spatial synchronization and zero-bias estimation to achieve spatiotemporal benchmark unification of multimodal data; S22) Data fusion: A loosely coupled, low-computing-power fusion method is adopted to fuse the inertial data collected by the IMU with the position data output by the infrared vision auxiliary module in real time to calculate the limb position information; S23) Micro-motion enhancement: For the minute motion amplitude in self-rescue and mutual rescue operations, a multi-scale wavelet transform and adaptive gain amplification mechanism are used to separate and enhance the micro-motion features. S24) Data Output: Based on the solution results of step S22) and the micro-motion feature enhancement results of step S23), the quaternion Q calculated by the IMU is output. t Convert to Euler angles form and enhance the micro-motion characteristics Encapsulated into a data packet, serving as the basis for subsequent scoring; S3) Sequential action recognition and logical scoring: Perform action segmentation, sub-operation recognition, timing verification and comprehensive scoring on the data packet output in step S2) to obtain the final assessment score.

6. The intelligent scoring method for self-rescue and mutual rescue assessment as described in claim 5, characterized in that, In step S21), Time synchronization uses the master control device as the sole clock source, and through periodic broadcasting and delayed response, the synchronization error between the IMUs deployed in various parts of the human body and the system timestamp of the master control device is ≤10ms. Spatial synchronization requires the subject to maintain a "T-shaped" standing posture for 5 seconds. The system calculates the gravitational acceleration vector and the geomagnetic north pole vector output by the IMU to establish the initial rotation matrix from the sensor coordinate system to the world navigation coordinate system. At the same time, using the position of the reflective point captured by the infrared vision auxiliary module, the rigid body transformation relationship from the visual coordinate system to the world navigation coordinate system is established to complete the spatial registration of multimodal data. Zero bias estimation uses no less than 300 frames of gyroscope output data collected during the static period to calculate and store the static zero bias of each axis.

7. The intelligent scoring method for self-rescue and mutual rescue assessment as described in claim 5, characterized in that, In step S22), the system performs data processing cyclically at a frequency of 120Hz, and performs the following operations sequentially for each time step t: 1) Inertial attitude acquisition: Directly read the quaternion output by the IMU. This represents the current inertial posture of the limbs. 2) Inertial position calculation: Obtain the triaxial linear acceleration output by the IMU. The acceleration is obtained by transforming the current attitude to the navigation coordinate system and subtracting the gravitational component g. ,right Double integration yields displacement calculated using pure inertia. The calculation formula is: ,in, The position at time t-1 Let be the velocity at time t-1, and Δt be the time step interval; 3) Infrared position correction: via marker position Determine whether the infrared vision assistance module has captured a valid marker point: when At the same time, high-precision position measured by infrared vision Covering the inertial estimation location, i.e. Simultaneously calculate and cache the integral drift compensation amount. ; when At that time, use the most recent effective drift compensation amount Correcting the inertial estimation position, the output position is .

8. The intelligent scoring method for self-rescue and mutual rescue assessment as described in claim 5, characterized in that, Step S23 includes the following specific operations: (1) Multi-scale detail separation: The raw inertial data stream (angular velocity, linear acceleration) acquired by the IMU is subjected to discrete wavelet transform, and the signal is decomposed into approximation coefficients and detail coefficients using the Daubechies wavelet basis, satisfying: ， in, This is the approximate component of the Jth layer, corresponding to large-amplitude limb movements; For the j-th layer detail component, it corresponds to micro-movements such as finger tremors and end-effector fine adjustments; (2) Adaptive micro-motion feature amplification: targeting the extracted micro-motion signals Construct a nonlinear gain amplifier and define a dynamic threshold. ;when When the feature is amplified by an exponential gain function, the formula is: ， in These are gain control parameters, and All > 0; (3) Extraction of fingertip / end micro-motion features: The IMU worn on the wrist is used to capture high-frequency micro-vibrations generated by tendon traction. The energy features of the 15-30Hz frequency band are extracted by spectrum analysis as a proxy indicator of finger movement activity.

9. The intelligent scoring method for self-rescue and mutual rescue assessment as described in claim 5, characterized in that, Step S3 includes the following specific operations: S31) Adaptive Action Segmentation: A sliding time window is used to slice the continuous data stream. The window length W and step size S are set, and the cosine similarity of the feature vectors of adjacent windows is calculated. When the similarity is lower than a threshold, it is determined as an action switching point, and the continuous data stream is divided into a set of discrete action segments. ; S32) Sub-operation recognition and individual scoring: This involves identifying and scoring action segments. The feature vectors are projected onto the standard sub-operation feature subspace constructed in step S1), and the Euclidean or Mahalanobis distance is calculated; the sub-operation type is identified based on the minimum distance principle. And generate sub-operation quality scores based on Gaussian mapping functions. D i S is the feature distance. max The full score is given by σ, where σ is the distance coefficient. S33) Sequential logic verification: Introduce a state transition matrix model to define an ordered state sequence for the standard operation procedure. and the allowed state transition probability matrix ; Maintain a real-time state pointer when a new sub-operation type is identified. At that time, the verification starts from arrive Is the transfer in If the statement is valid, a penalty will be deducted based on the logic. and to Multiply by the penalty coefficient β (0 < β < 1); S34) Weighted Comprehensive Score: Final Assessment Score The weighted sum of the scores for each sub-operation is calculated using the following formula: ， Among them, w k As the importance weight of sub-operations, As a macroscopic action quality score, Add points to the micro-motions calculated in step S23). The sequential logic coefficients calculated in step S33) Additional points will be deducted for redundant / ineffective actions.

Images