A hierarchical sitting posture classification method and system

Abstract

The application discloses a layered sitting posture classification method and system, which comprises the following steps: obtaining pressure sensor array data on a seat cushion to obtain a pressure distribution matrix; preprocessing and partition calculation are performed on the pressure distribution matrix to obtain a regional pressure ratio; the overall sitting posture category of an operator is judged based on the regional pressure ratio feature to obtain a general posture category; six states of the sitting posture are obtained through coarse classification, and a preset threshold rule is used for pattern recognition on the regional pressure ratio; when the coarse classification result remains stable, a fine classification discrimination process is triggered; if the coarse classification result frequently fluctuates or is unstable, the fine classification discrimination is suspended; during the stable coarse classification, the gravity center trajectory and micro-motion features of the pressure distribution are extracted, the gravity center trajectory and micro-motion features are input into the fine classification discrimination, the fine classification discrimination is combined with the general posture category to identify a fine classification posture type; and the posture recognition result is reported and fed back through an output interface.

Description

Technical Field

[0001] This invention relates to the fields of posture monitoring and human-machine engineering technology, and in particular to a hierarchical sitting posture classification method and system. Background Technology

[0002] The main control room of a nuclear power plant is the core location for the safe operation of the nuclear power unit. Operators typically monitor and operate the system in a seated position for extended periods. Maintaining good posture during work is crucial to prevent operational errors due to fatigue or distraction. Abnormal posture or unintentional movement by the operator can impair the speed of response to equipment monitoring, posing potential safety hazards.

[0003] Current methods for monitoring sitting posture mainly fall into two categories: vision-based image detection and wearable device-based monitoring. Vision-based sitting posture recognition methods use cameras to acquire images of people and then determine their posture using algorithms such as deep learning. However, in the environment of a nuclear power plant's main control room, camera-based solutions have several limitations: First, the control room contains numerous devices, and operators may be partially obstructed by the control console, leading to blind spots due to camera line-of-sight obstruction; second, continuous video monitoring may cause privacy concerns and psychological stress for operators, which is not conducive to long-term use; therefore, wearable sensor-based methods are also unsuitable for this scenario.

[0004] Embedding pressure sensor arrays into seat cushions offers a non-invasive alternative for posture monitoring. By sensing the pressure distribution exerted by the operator's buttocks and legs on the seat cushion, the posture of the person in the seat can be inferred. Existing research and products on smart seats utilize multi-point pressure sensors to detect poor posture and provide alerts. In practical applications, due to individual differences and subtle unconscious movements, pressure distribution dynamically changes. Directly classifying posture based on pressure data is prone to inconsistencies and false alarms, especially in high-reliability environments such as nuclear power plant control rooms. Therefore, there is an urgent need for a more robust, hierarchical posture classification method: first, a robust coarse posture classification is performed, and then fine-grained classification and precise recognition are triggered when an abnormal posture trend is confirmed, thereby improving recognition accuracy and reducing unnecessary computational overhead. Summary of the Invention

[0005] The purpose of this invention is to provide a method for classifying sitting postures in a hierarchical manner.

[0006] To achieve the above objectives, the present invention is implemented according to the following technical solution: This invention includes the following steps: Acquire data from the pressure sensor array on the seat cushion to obtain the pressure distribution matrix of the operator's buttocks and legs on each area of ​​the seat cushion at the current moment; The pressure distribution matrix is ​​preprocessed and partitioned. The seat cushion sensor array is divided into multiple regions: front, back, left, right, and center. The pressure values ​​in each region are accumulated, and the proportion of the pressure value in each region to the total pressure is calculated to obtain the region pressure ratio. Based on the regional pressure ratio characteristics, the operator's overall sitting posture category is determined to obtain the general posture category: the sitting posture is divided into six states through coarse classification: normal sitting, leaning forward, leaning back, leaning to the left, leaning to the right, and sitting up. The pressure ratio of each region is used for pattern recognition by a pre-set threshold rule. The continuity of the coarse classification results is monitored. When the coarse classification results remain stable, the fine classification process is triggered. If the coarse classification results fluctuate frequently or are unstable, the fine classification process is temporarily suspended. During the coarse classification stabilization period, the center-of-gravity trajectory and micro-motion features of the pressure distribution are extracted: by calculating the trajectory of the center-of-gravity position of the pressure distribution matrix over time, the center-of-gravity trajectory curve and dynamic feature parameters reflecting the operator's subtle movements are obtained; the dynamic feature parameters include the amplitude and velocity of the small fluctuations in the pressure center position. The center of gravity trajectory and micro-motion features are input into the fine classification discrimination, and combined with the general posture category to identify the fine-classified posture type; the fine classification discrimination adopts a combination of rule judgment and pattern recognition, analyzes the direction of the center of gravity trajectory to determine the left and right deviation of the forward leaning posture, and distinguishes between a stable tilt and a continuous sliding state through micro-motion features, and outputs the sitting posture recognition result; the fine-classified postures include left forward leaning, right forward leaning, sliding posture, and twisting posture; The posture recognition results are reported and fed back through the output interface. Under normal sitting posture, the system only records the posture or provides slight prompts; when abnormal states such as poor sitting posture or operator leaving the seat are detected, alarm or prompt signals are triggered.

[0007] Further, the method for obtaining the regional pressure ratio includes: The raw pressure signal from the sensor array is read and preprocessed through analog-to-digital conversion and filtering. Let the first... The two-dimensional pressure matrix output at each sampling time point is: ; in For the first Line number The number of sensor points at time The collected pressure values, and These are the row and column indexes, respectively. Represents the field of nonnegative real numbers. Index for discrete sampling time; The matrix is ​​divided into predetermined regions: the first 8 rows of the sensing matrix are grouped into the front region, and the last 8 rows into the back region; the left 8 columns are grouped into the left region, and the right 8 columns into the right region; in addition, the center part of the matrix is ​​defined as the center region. The array is divided into front, back, left, right, and center regions, and the results are summed for each region. ;in For a moment The sum of the pressure values ​​in the front region, For a moment The sum of the pressure values ​​in the rear region, For a moment The sum of the pressure values ​​in the left region, For a moment The sum of the pressure values ​​in the right region, For a moment The sum of pressure values ​​in the central region; Calculate total pressure The regional pressure ratio characteristics are defined based on the total pressure. , , , , Constructing the regional pressure ratio eigenvector Among them are For a moment Total pressure, The pressure in the anterior region and the total pressure, For the pressure in the rear region and the total pressure, The pressure in the left region is equal to the total pressure. The pressure in the right-side region is equal to the total pressure. The pressure in the central area and the total pressure; Furthermore, the method for obtaining the general pose category includes: like Significantly higher than and If the absolute value exceeds a certain threshold, it is judged as forward tilting; if Much higher Then it is judged as leaning back; or If the proportion on one side is abnormally high and the corresponding proportion on the other side is low, it is determined that the body is leaning towards the corresponding direction; if If the total pressure value is below the body weight threshold, it is determined that the person is currently in an unseated state; if all areas are within the normal range and If the value is close to the center of symmetry, it is considered a normal sitting posture. The coarse classification result, attitude label, is output and passed to the hierarchical triggering logic module 106 and the output interface module 107. The difference threshold before and after classification is set as follows: The left-right difference threshold is The coarse classification rule is: forward tilt. Leaning back Left rightward ;in and The threshold parameter is obtained through pre-experiment calibration. Furthermore, the method for determining the stability of the coarse classification result includes: Let the length of the sliding time window be... The sampling time interval is Then the window ,make Indicates time The coarse classification output labels, the coarse classification label set is ; Take discrete time intervals within the window The majority class is defined as , Using the majority-percentage stability criterion, we first calculate the coarse category that appears most frequently within the window, and then set a stability threshold. The stability criterion is defined as follows: ; in This is an indicator function; it takes the value 1 if the condition within the parentheses is true, and 0 otherwise. The independent variable that maximizes the objective function. The most frequently appearing coarse category label within the window; when If the coarse classification is considered stable, the fine classification process will be triggered; otherwise, the triggering will be delayed and the process will continue to wait.

[0008] Furthermore, the method for extracting the center of gravity trajectory includes: Once the fine classification process is triggered, the time series of the pressure data is analyzed. Starting from the trigger time, a predetermined time is traced back, and the pressure centroid position of each frame of the pressure matrix is ​​calculated within the predetermined time, forming a centroid movement trajectory curve: ; Define the centroid vector ;in The physical coordinates of the sensing point in the seat cushion coordinate system. Let be the total pressure at time t. To prevent zero constant.

[0009] Furthermore, the method for identifying the fine-classified pose type includes: By combining coarse classification results and dynamic features, a refined judgment is made on the specific posture type, based on the stable coarse classification results. Limit the set of candidate subcategories; Candidate subclasses are determined based on the direction of center of gravity displacement, trajectory trend, and intensity of micro-motion: Let the coordinates of the seat cushion center be... And define the relative displacement at the end of the window: ; make The net displacement amplitude is defined as ,in It is a Euclidean second norm. for; Based on the broad category of the coarse classification, the scope of the fine classification is limited, and the current center of gravity trajectory is matched with multiple fine classification posture models. The matching results include left forward tilt posture, right forward tilt posture, glide posture, and torsional posture. In a coarse classification, if the center of gravity is leaning forward and has both forward and leftward displacement components relative to the center of the seat, and the pressure on the left side is higher than that on the right side, it is judged as a left forward leaning posture; similarly, if the center of gravity is shifted forward and biased to the right, and the pressure on the right side is larger, it is judged as a right forward leaning posture. The sliding type is determined by detecting the continuous movement of the center of gravity trajectory in the forward and backward directions, accompanied by a downward shift of the total pressure center: for the main direction sequence within the window. Perform linear regression, setting the main direction... If we take the coordinates of the center of gravity in the forward and backward directions, then Define the trend slope output ,in The window length used for fine-class trend estimation. For the fitting intercept, The fitted slope, At the end of the window The estimated trend slope is obtained at that point. for, for, for; When the ratio of trajectory path length to net displacement increases significantly and the fretting frequency is high, it is determined to be a torsional type: calculate the trajectory path length. Define the ratio index ,when When the frequency exceeds the threshold and the micro-motion frequency is high, it is determined to be torsional; among which To prevent zero constant, Based on predefined discrimination logic or machine learning classifiers, the input trajectory and micro-motion features are assigned to the most suitable fine-classification pose type, and the fine-classification result is output.

[0010] Secondly, a hierarchical sitting posture classification system includes: Pressure data acquisition module: used to acquire data from the pressure sensor array in the seat cushion of the nuclear power plant's main control room, and to collect the pressure distribution matrix in real time when the operator is seated; Regional pressure characteristic calculation module: used to summarize and calculate the pressure distribution matrix according to the predetermined front, back, left, right and center regions to obtain the pressure ratio characteristic parameters of each region; Coarse classification and discrimination module: used to perform coarse classification of the operator's sitting posture based on the regional pressure characteristics, and to determine whether the current posture belongs to one of the following: sitting upright, leaning forward, leaning back, leaning to the left, leaning to the right, or sitting up. Layered triggering logic module: Used to monitor changes in coarse classification results, and activates the fine classification module only when the coarse classification results are stable and meet the conditions; when the coarse classification results are unstable, fine classification is not triggered temporarily to reduce misjudgments; Center of gravity trajectory and micro-motion feature extraction module: used to calculate the center of gravity position change trajectory and micro-motion feature parameters of the pressure distribution over a recent period during fine classification activation, providing the dynamic feature input required for fine classification; The fine classification and discrimination module is used to further distinguish specific sitting postures based on the coarse classification results, including posture types such as left forward leaning, right forward leaning, sliding type, and twisting type, according to the center of gravity trajectory and micro-motion characteristics. Output interface and prompt module: Used to output the sitting posture recognition results and issue prompt signals or alarm information when abnormal sitting posture or leaving the seat is detected, and provide the posture classification results to the operator or safety monitoring system.

[0011] The beneficial effects of this invention are: This invention is a hierarchical sitting posture classification method and system. Compared with the prior art, this invention has the following technical advantages: This invention acquires posture information based on a pressure sensor built into the seat cushion, eliminating the need for video surveillance of operators and avoiding privacy and line-of-sight issues in control room monitoring. The monitoring process is imperceptible to the operator and does not interfere with normal work. It employs a two-tiered architecture of coarse and fine classification, initiating fine classification analysis only when coarse classification results indicate a clearly abnormal and persistent posture. This effectively filters out interference from transient posture changes, reducing false alarms and misclassifications, and improving the stability and accuracy of the identification results. It promptly detects poor operator posture or absence from duty, prompting corrections or reporting to management via a notification module. This helps prevent safety risks caused by operator fatigue or inattention, boasting high accuracy and low false alarm rates, ensuring reliable alarms when needed. This enhances the human factors engineering monitoring level of the nuclear power plant's control room and guarantees the safe operation of the unit. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating the steps of a hierarchical sitting posture classification method according to the present invention. Figure 2 This is a functional module structure diagram of a hierarchical sitting posture classification system in an embodiment of this specification. Detailed Implementation

[0013] The present invention will be further described below through specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0014] The present invention provides a method and system for classifying seated postures, comprising the following steps: like Figure 1 As shown, this embodiment includes the following steps: Acquire data from the pressure sensor array on the seat cushion to obtain the pressure distribution matrix of the operator's buttocks and legs on each area of ​​the seat cushion at the current moment; The pressure distribution matrix is ​​preprocessed and partitioned. The seat cushion sensor array is divided into multiple regions: front, back, left, right, and center. The pressure values ​​in each region are accumulated, and the proportion of the pressure value in each region to the total pressure is calculated to obtain the region pressure ratio. Based on the regional pressure ratio characteristics, the operator's overall sitting posture category is determined to obtain the general posture category: the sitting posture is divided into six states through coarse classification: normal sitting, leaning forward, leaning back, leaning to the left, leaning to the right, and sitting up. The pressure ratio of each region is used for pattern recognition by a pre-set threshold rule. The continuity of the coarse classification results is monitored. When the coarse classification results remain stable, the fine classification process is triggered. If the coarse classification results fluctuate frequently or are unstable, the fine classification process is temporarily suspended. During the coarse classification stabilization period, the center-of-gravity trajectory and micro-motion features of the pressure distribution are extracted: by calculating the trajectory of the center-of-gravity position of the pressure distribution matrix over time, the center-of-gravity trajectory curve and dynamic feature parameters reflecting the operator's subtle movements are obtained; the dynamic feature parameters include the amplitude and velocity of the small fluctuations in the pressure center position. The center of gravity trajectory and micro-motion features are input into the fine classification discrimination, and combined with the general posture category to identify the fine-classified posture type; the fine classification discrimination adopts a combination of rule judgment and pattern recognition, analyzes the direction of the center of gravity trajectory to determine the left and right deviation of the forward leaning posture, and distinguishes between a stable tilt and a continuous sliding state through micro-motion features, and outputs the sitting posture recognition result; the fine-classified postures include left forward leaning, right forward leaning, sliding posture, and twisting posture; The posture recognition results are reported and fed back through the output interface. Under normal sitting posture, the system only records the posture or provides slight prompts; when abnormal states such as poor sitting posture or operator leaving the seat are detected, alarm or prompt signals are triggered.

[0015] In this embodiment, the method for obtaining the regional pressure ratio includes: The raw pressure signal from the sensor array is read and preprocessed through analog-to-digital conversion and filtering. Let the first... The two-dimensional pressure matrix output at each sampling time point is: ; in For the first Line number The number of sensor points at time The collected pressure values, and These are the row and column indexes, respectively. Represents the field of nonnegative real numbers. Index for discrete sampling time; The matrix is ​​divided into predetermined regions: the first 8 rows of the sensing matrix are grouped into the front region, and the last 8 rows into the back region; the left 8 columns are grouped into the left region, and the right 8 columns into the right region; the center of the matrix is ​​randomly defined as the central region, and the array is divided into the front, back, left, right, and central regions, and the results are summed for each region. ;in For a moment The sum of the pressure values ​​in the front region, For a moment The sum of the pressure values ​​in the rear region, For a moment The sum of the pressure values ​​in the left region, For a moment The sum of the pressure values ​​in the right region, For a moment The sum of pressure values ​​in the central region; Calculate total pressure The regional pressure ratio characteristics are defined based on the total pressure. , , , , Constructing the regional pressure ratio eigenvector Among them are For a moment Total pressure, The pressure in the anterior region and the total pressure, For the pressure in the rear region and the total pressure, The pressure in the left region is equal to the total pressure. The pressure in the right-side region is equal to the total pressure. The pressure in the central area and the total pressure; In actual assessment, the seat cushion of the main control room of the nuclear power plant is embedded with a 16×16 pressure sensor array 101, which is used to collect the pressure distribution data of the operator's buttocks and thighs on the surface of the seat cushion in real time. The pressure data acquisition module 102 outputs a two-dimensional pressure matrix; the regional pressure characteristic calculation module 103 divides the matrix according to a predetermined region.

[0016] In this embodiment, the method for obtaining the general pose category includes: like Significantly higher than and If the absolute value exceeds a certain threshold, it is judged as forward tilting; if Much higher Then it is judged as leaning back; or If the proportion on one side is abnormally high and the corresponding proportion on the other side is low, it is determined that the body is leaning towards the corresponding direction; if If the total pressure value is below the body weight threshold, it is determined that the person is currently in an unseated state; if all areas are within the normal range and If the value is close to the center of symmetry, it is considered a normal sitting posture. The coarse classification result, attitude label, is output and passed to the hierarchical triggering logic module 106 and the output interface module 107. The difference threshold before and after classification is set as follows: The left-right difference threshold is The coarse classification rule is: forward tilt. Leaning back Left rightward ;in and The threshold parameter is obtained through pre-experiment calibration. In actual evaluation, the coarse classification module 104 determines the general posture category based on the regional pressure feature vector. Module 104 pre-stores the discrimination rules for each posture category and monitors the stability of the coarse classification results through the hierarchical triggering logic module 106. When module 106 detects that the coarse classification output remains unchanged within a preset time window, it determines that the current posture has entered a stable state and activates the center of gravity trajectory and micro-motion feature extraction module 105 and the fine classification module 108 to begin in-depth fine classification analysis. Conversely, if the coarse classification results change repeatedly in a short period of time, module 106 delays the fine classification trigger signal, temporarily does not start the fine classification module, and continues to wait for the coarse classification results to stabilize.

[0017] In this embodiment, the method for determining the stability of the coarse classification result includes: Let the length of the sliding time window be... The sampling time interval is Then the window ,make Indicates time The coarse classification output labels, the coarse classification label set is ; Take discrete time intervals within the window The majority class is defined as , Using the majority-percentage stability criterion, we first calculate the coarse category that appears most frequently within the window, and then set a stability threshold. The stability criterion is defined as follows: ; in This is an indicator function; it takes the value 1 if the condition within the parentheses is true, and 0 otherwise. The independent variable that maximizes the objective function. The most frequently appearing coarse category label within the window; when If the coarse classification is considered stable, the fine classification process will be triggered; otherwise, the triggering will be delayed and the process will continue to wait.

[0018] In this embodiment, the method for extracting the center of gravity trajectory includes: Once the fine classification process is triggered, the time series of the pressure data is analyzed. Starting from the trigger time, a predetermined time is traced back, and the pressure centroid position of each frame of the pressure matrix is ​​calculated within the predetermined time, forming a centroid movement trajectory curve: ; Define the centroid vector ;in The physical coordinates of the sensing point in the seat cushion coordinate system. Let be the total pressure at time t. To prevent zero constant; In actual assessment, when the fine classification process is triggered, the center of gravity trajectory and micro-motion feature extraction module 105 begins to analyze the time series of pressure data. The module 105 traces back a predetermined time from the triggering time. The center of gravity trajectory reflects the path of the operator's center of gravity on the seat cushion during the current period of time; corresponding to the forward-leaning posture, the center of gravity trajectory shows a smooth forward movement and a new equilibrium point towards the front; if it is a gliding posture, the trajectory will show a continuous trend of slow movement in one direction. The micro-motion feature extraction module 105 calculates the micro-motion feature parameters of the center of gravity position. The micro-motion feature parameters include the swing amplitude of the trajectory curve and the frequency of back-and-forth shaking. The micro-motion feature parameters measure the stability of the operator's body when maintaining the posture. For a static posture, the micro-motion amplitude of the center of gravity is very small. However, for unstable twisting or adjustment, the frequency and amplitude of the micro-motion of the center of gravity will be larger. The module 105 sends the generated trajectory features and micro-motion parameters to the fine classification and discrimination module 108.

[0019] In this embodiment, the method for identifying the fine-classified pose type includes: By combining coarse classification results and dynamic features, a refined judgment is made on the specific posture type, based on the stable coarse classification results. Limit the set of candidate subcategories; Candidate subclasses are determined based on the direction of center of gravity displacement, trajectory trend, and intensity of micro-motion: Let the coordinates of the seat cushion center be... And define the relative displacement at the end of the window: ; make The net displacement amplitude is defined as ,in It is a Euclidean second norm. for; Based on the broad category of the coarse classification, the scope of the fine classification is limited, and the current center of gravity trajectory is matched with multiple fine classification posture models. The matching results include left forward tilt posture, right forward tilt posture, glide posture, and torsional posture. In a coarse classification, if the center of gravity is leaning forward and has both forward and leftward displacement components relative to the center of the seat, and the pressure on the left side is higher than that on the right side, it is judged as a left forward leaning posture; similarly, if the center of gravity is shifted forward and biased to the right, and the pressure on the right side is larger, it is judged as a right forward leaning posture. The sliding type is determined by detecting the continuous movement of the center of gravity trajectory in the forward and backward directions, accompanied by a downward shift of the total pressure center: for the main direction sequence within the window. Perform linear regression, setting the main direction... If we take the coordinates of the center of gravity in the forward and backward directions, then Define the trend slope output ,in The window length used for fine-class trend estimation. For the fitting intercept, The fitted slope, At the end of the window The estimated trend slope is obtained at that point. for, for, for; When the ratio of trajectory path length to net displacement increases significantly and the fretting frequency is high, it is determined to be a torsional type: calculate the trajectory path length. Define the ratio index ,when When the frequency exceeds the threshold and the micro-motion frequency is high, it is determined to be torsional; among which To prevent zero constant, Based on predefined discrimination logic or machine learning classifiers, the input trajectory and micro-motion features are assigned to the most suitable fine-classification pose type, and the fine-classification result is output. In actual assessment, the characteristic of large path length but small net displacement is used to characterize the repeated left and right swaying; the torsional posture is characterized by the center of gravity swaying back and forth slightly from side to side, the pressure center is unstable, and the micro-motion characteristics are obvious. When the coarse classification is "sitting upright" but the center of gravity trajectory shows repeated small left and right movements, it corresponds to the fine classification of "unstable sitting posture"; when the coarse classification is "leaving the seat" and the pressure gradually decreases, the fine classification can be judged as "getting up and leaving the seat". Module 108 assigns the input trajectory and micro-motion features to the most suitable fine-classification pose type based on predefined discrimination logic or machine learning classifier, and outputs the fine-classification result. After obtaining the detailed classification results, the system provides the final posture category to the user interface or alarm device through the output interface and the prompt module 107. The output module 107 includes a signal processing unit and a prompt device: when it is detected that the operator is in an abnormal posture for a long time, such as continuously leaning forward for more than a preset threshold time, the module 107 triggers vibration or sound prompts on the seat to remind the operator to correct the sitting posture; if it detects that the operator is out of seat for more than the allowed time, the system will issue an alarm to notify other on-duty personnel; the output module can also display the posture status in real time on the monitoring screen of the control room interface for the shift leader or safety supervisor to view; once the operator returns to the correct sitting posture or the alarm is cleared, the prompts will stop; at the same time, the system saves all identification records (timestamp, posture category, duration) to the log for future analysis of operator behavior and reactions; As described in the above embodiments, the current hierarchical sitting posture classification method and system can reliably monitor the operator's sitting posture in the nuclear power plant main control room environment. Experiments show that the current hierarchical sitting posture classification system has improved accuracy in identifying static correct sitting posture, dynamic poor sitting posture, and personnel leaving their posts. Brief changes in body posture will not cause false alarms. When the operator shows an abnormal posture trend (gradually leaning forward and dozing off), the system can promptly detect and issue a reminder to prevent potential safety hazards. Due to the adoption of pressure array sensing technology and hierarchical processing algorithm, the system operates efficiently and is unaffected by light or electromagnetic interference, making it very suitable for deployment in high-safety human-machine interface environments such as nuclear power plants.

[0020] Secondly, a hierarchical sitting posture classification system includes: Pressure data acquisition module: used to acquire data from the pressure sensor array in the seat cushion of the nuclear power plant's main control room, and to collect the pressure distribution matrix in real time when the operator is seated; Regional pressure characteristic calculation module: used to summarize and calculate the pressure distribution matrix according to the predetermined front, back, left, right and center regions to obtain the pressure ratio characteristic parameters of each region; Coarse classification and discrimination module: used to perform coarse classification of the operator's sitting posture based on the regional pressure characteristics, and to determine whether the current posture belongs to one of the following: sitting upright, leaning forward, leaning back, leaning to the left, leaning to the right, or sitting up. Layered triggering logic module: Used to monitor changes in coarse classification results, and activates the fine classification module only when the coarse classification results are stable and meet the conditions; when the coarse classification results are unstable, fine classification is not triggered temporarily to reduce misjudgments; Center of gravity trajectory and micro-motion feature extraction module: used to calculate the center of gravity position change trajectory and micro-motion feature parameters of the pressure distribution over a recent period during fine classification activation, providing the dynamic feature input required for fine classification; The fine classification and discrimination module is used to further distinguish specific sitting postures based on the coarse classification results, including posture types such as left forward leaning, right forward leaning, sliding type, and twisting type, according to the center of gravity trajectory and micro-motion characteristics. Output interface and prompt module: Used to output the sitting posture recognition results and issue prompt signals or alarm information when abnormal sitting posture or leaving the seat is detected, and provide the posture classification results to the operator or safety monitoring system.

[0021] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A hierarchical sitting posture classification method, characterized in that, Includes the following steps: Acquire data from the pressure sensor array on the seat cushion to obtain the pressure distribution matrix of the operator's buttocks and legs on each area of ​​the seat cushion at the current moment; The pressure distribution matrix is ​​preprocessed and partitioned. The seat cushion sensor array is divided into multiple regions: front, back, left, right, and center. The pressure values ​​in each region are accumulated, and the proportion of the pressure value in each region to the total pressure is calculated to obtain the region pressure ratio. Based on the regional pressure ratio characteristics, the operator's overall sitting posture category is determined to obtain the general posture category: the sitting posture is divided into six states through coarse classification: normal sitting, leaning forward, leaning back, leaning to the left, leaning to the right, and sitting up. The pressure ratio of each region is used for pattern recognition by a pre-set threshold rule. Monitor the continuity of the coarse classification results. When the coarse classification results remain stable, trigger the entry into the fine classification process. If the coarse classification results fluctuate frequently or are unstable, then the fine classification judgment should be postponed. During the coarse classification stabilization period, the center-of-gravity trajectory and micro-motion features of the pressure distribution are extracted: by calculating the trajectory of the center-of-gravity position of the pressure distribution matrix over time, the center-of-gravity trajectory curve and dynamic feature parameters reflecting the operator's subtle movements are obtained; the dynamic feature parameters include the amplitude and velocity of the small fluctuations in the pressure center position. The center of gravity trajectory and micro-motion features are input into the fine classification discrimination, and combined with the general posture category to identify the fine-classified posture type; the fine classification discrimination adopts a combination of rule judgment and pattern recognition, analyzes the direction of the center of gravity trajectory to determine the left and right deviation of the forward leaning posture, and distinguishes between a stable tilt and a continuous sliding state through micro-motion features, and outputs the sitting posture recognition result; the fine-classified postures include left forward leaning, right forward leaning, sliding posture, and twisting posture; The posture recognition results are reported and fed back through the output interface. Under normal sitting posture, the system only records the posture or provides slight prompts; when abnormal states such as poor sitting posture or operator leaving the seat are detected, alarm or prompt signals are triggered.

2. The layered sitting posture classification method according to claim 1, characterized in that, A method for obtaining the pressure ratio of the region includes: The raw pressure signal from the sensor array is read and preprocessed through analog-to-digital conversion and filtering. Let the first... The two-dimensional pressure matrix output at each sampling time point is: ； in For the first Line number The number of sensor points at time The collected pressure values, and These are the row and column indexes, respectively. Represents the field of nonnegative real numbers. Index for discrete sampling time; The matrix is ​​divided into predetermined regions: the first 8 rows of the sensing matrix are grouped into the front region, and the last 8 rows into the back region; the left 8 columns are grouped into the left region, and the right 8 columns into the right region; in addition, the center part of the matrix is ​​defined as the center region. The array is divided into front, back, left, right, and center regions, and the results are summed for each region. ;in For a moment The sum of the pressure values ​​in the front region, For a moment The sum of the pressure values ​​in the rear region, For a moment The sum of the pressure values ​​in the left region, For a moment The sum of the pressure values ​​in the right region, For a moment The sum of pressure values ​​in the central region; Calculate total pressure The regional pressure ratio characteristics are defined based on the total pressure. , , , , Constructing the regional pressure ratio eigenvector Among them are For a moment Total pressure, The pressure in the anterior region and the total pressure, For the pressure in the rear region and the total pressure, The pressure in the left region is equal to the total pressure. The pressure in the right-side region is equal to the total pressure. The pressure in the central region and the total pressure.

3. The layered sitting posture classification method according to claim 1, characterized in that, The method for obtaining the general pose category includes: like Significantly higher than and If the absolute value exceeds a certain threshold, it is judged as forward tilting; if Much higher Then it is judged as leaning back; or If the proportion on one side is abnormally high and the corresponding proportion on the other side is low, it is determined that the body is leaning towards the corresponding direction; if If the total pressure value is below the body weight threshold, it is determined that the person is currently in an unseated state; if all areas are within the normal range and If the value is close to the center of symmetry, it is considered a normal sitting posture. The coarse classification result, attitude label, is output and passed to the hierarchical triggering logic module 106 and the output interface module 107. The difference threshold before and after classification is set as follows: The left-right difference threshold is The coarse classification rule is: forward tilt. Leaning back Left rightward ;in and These are the threshold parameters obtained through pre-experiment calibration.

4. The layered sitting posture classification method according to claim 1, characterized in that, The methods for determining the stability of the coarse classification results include: Let the length of the sliding time window be... The sampling time interval is The current time is Then the window Discrete times within the window are denoted as ,in ; make Indicates time The coarse classification output labels, the coarse classification label set is ; majority class definition Using the majority probability stability criterion, the coarse classification category that appears most frequently within the calculation window is defined as follows: ； in The coarse category label that appears most frequently within the window; The independent variable that maximizes the objective function; This is a stability threshold used to constrain the proportion of the majority class within the window; This is an indicator function; it takes the value 1 if the condition inside the parentheses is true, and 0 otherwise. If the coarse classification is considered stable, the fine classification process will be triggered; otherwise, the triggering will be delayed and the process will continue to wait.

5. The layered sitting posture classification method according to claim 1, characterized in that, The method for extracting the centroid trajectory includes: Once the fine classification process is triggered, the time series of the pressure data is analyzed. Starting from the trigger time, a predetermined time is traced back, and the pressure centroid position of each frame of the pressure matrix is ​​calculated within the predetermined time, forming a centroid movement trajectory curve: ; Define the centroid vector ;in and These are the horizontal and vertical coordinates of the center of gravity in the seat coordinate system, respectively. The physical coordinates of the sensing point in the seat cushion coordinate system. Let be the total pressure at time t. To prevent zero constant, Indicates time The pressure center of gravity coordinate vector.

6. The layered sitting posture classification method according to claim 1, characterized in that, The method for identifying the fine-classified pose type includes: By combining coarse classification results and dynamic features, a refined judgment is made on the specific posture type, based on the stable coarse classification results. Limit the set of candidate subcategories; Candidate subclasses are determined based on the direction of center of gravity displacement, trajectory trend, and intensity of micro-motion: Let the coordinates of the seat cushion center be... And define the relative displacement at the end of the window: ,in The reference coordinates representing the center position of the seat cushion are preferably the geometric center of the seat cushion or the average center of gravity coordinates under the operator's positive coordinate system. and Indicates the lateral and longitudinal displacements of the center of gravity relative to the reference center; make The net displacement amplitude is defined as ,in It is a relative displacement vector. This is the net displacement magnitude, used to measure the degree to which the current center of gravity deviates from the center. It is the Euclidean second norm, which is the square root of the sum of squares of the vector components; Based on the broad category of the coarse classification, the scope of the fine classification is limited, and the current center of gravity trajectory is matched with multiple fine classification posture models. The matching results include left forward tilt posture, right forward tilt posture, glide posture, and torsional posture. In a coarse classification, if the center of gravity is leaning forward and has both forward and leftward displacement components relative to the center of the seat, and the pressure on the left side is higher than that on the right side, it is judged as a left forward leaning posture; similarly, if the center of gravity is shifted forward and biased to the right, and the pressure on the right side is larger, it is judged as a right forward leaning posture. The sliding type is determined by detecting the continuous movement of the center of gravity trajectory in the forward and backward directions, accompanied by a downward shift of the total pressure center: for the main direction sequence within the window. Perform linear regression, setting the main direction... If we take the coordinates of the center of gravity in the forward and backward directions, then Define the trend slope output ,in The window length used for fine-class trend estimation. For the fitting intercept, The fitted slope, At the end of the window The estimated trend slope is obtained at that point. for, for, These are discrete time points within the window; When the ratio of trajectory path length to net displacement increases significantly and the fretting frequency is high, it is determined to be a torsional type: calculate the trajectory path length. Define the ratio index ,when When the frequency exceeds the threshold and the micro-motion frequency is high, it is determined to be torsional; among which To prevent zero constant, Based on predefined discrimination logic or machine learning classifiers, the input trajectory and micro-motion features are assigned to the most suitable fine-classification pose type, and the fine-classification result is output.

7. A hierarchical sitting posture classification system for performing the method according to any one of claims 1-6, characterized in that, include: Pressure data acquisition module: used to acquire data from the pressure sensor array in the seat cushion of the nuclear power plant's main control room, and to collect the pressure distribution matrix in real time when the operator is seated; Regional pressure characteristic calculation module: used to summarize and calculate the pressure distribution matrix according to the predetermined front, back, left, right and center regions to obtain the pressure ratio characteristic parameters of each region; Coarse classification and discrimination module: used to perform coarse classification of the operator's sitting posture based on the regional pressure characteristics, and to determine whether the current posture belongs to one of the following: sitting upright, leaning forward, leaning back, leaning to the left, leaning to the right, or sitting up. Layered triggering logic module: Used to monitor changes in coarse classification results, and only activates the fine classification module when the coarse classification results stably meet the conditions; When the coarse classification results are unstable, fine classification is temporarily suspended to reduce misjudgments; Center of gravity trajectory and micro-motion feature extraction module: used to calculate the center of gravity position change trajectory and micro-motion feature parameters of the pressure distribution over a recent period during fine classification activation, providing the dynamic feature input required for fine classification; The fine classification and discrimination module is used to further distinguish specific sitting postures based on the coarse classification results, including posture types such as left forward leaning, right forward leaning, sliding type, and twisting type, according to the center of gravity trajectory and micro-motion characteristics. Output interface and prompt module: Used to output the sitting posture recognition results and issue prompt signals or alarm information when abnormal sitting posture or leaving the seat is detected, and provide the posture classification results to the operator or safety monitoring system.

Images