A multi-region pressure sensing based posture recognition method and intelligent high-speed rail seat

Abstract

This invention relates to a posture recognition method based on multi-area pressure sensing and an intelligent high-speed rail seat. The method includes: collecting real-time pressure distribution data for various parts of the passenger's body using a flexible thin-film pressure sensor matrix laid on the headrest, upper backrest, lumbar region of the backrest, buttocks, thighs, and leg rest; identifying the passenger's current posture based on the pressure distribution data, including sitting, supine, and lateral positions; and generating corresponding control commands based on the identified posture to control the operation of airbags and / or massage motors distributed in various areas of the seat. This invention, through refined partitioning of the backrest and seat cushion, can accurately distinguish between supine and lateral positions and intelligently control the massage function accordingly: thigh massage is disabled when sitting to avoid discomfort; full-body synergistic massage is activated when supine; and regular massage is paused and micro-pressure migration is initiated to protect the shoulder and hip joints when lateral. This invention achieves intelligent matching of posture and massage function, significantly improving riding comfort and safety.

Description

Technical Field

[0001] This invention relates to the field of intelligent seating and ergonomics technology, specifically to a posture recognition method based on multi-region pressure sensing and an intelligent high-speed rail seat using this method. Background Technology

[0002] With the widespread adoption of high-speed rail and increased travel demands, passengers are spending significantly more time seated during their journeys. Maintaining a fixed sitting posture for extended periods, especially poor static posture, can easily lead to stiffness in the back muscles, poor blood circulation, and consequently, deep fatigue, aches and pains, and even an increased risk of lumbar strain. This not only seriously affects passengers' travel experience and health but also negatively impacts their subsequent work and daily life.

[0003] To enhance passenger comfort, various massage-equipped seats have been disclosed in existing technologies. These seats typically achieve massage modes such as tapping and kneading by integrating vibration motors or airbags into the seat cushion and backrest. However, such solutions have significant limitations: Firstly, their operating modes are mostly pre-set fixed programs, manually triggered by the passenger or activated at set times, failing to sense the passenger's actual physiological state and exhibiting low levels of intelligence. For example, when a passenger is in deep sleep, a strong massage may disturb their rest; and when a passenger is already fatigued but does not actively seek help, the system cannot provide timely intervention. This lack of intelligent intervention results in a poor passenger experience and low energy efficiency.

[0004] To further optimize the user experience, some improvement solutions attempt to introduce sensors for status awareness. For example, some solutions use pressure sensors to detect whether someone is seated, controlling the automatic start and stop of functions; others monitor sitting time to determine whether intervention is needed. In recent years, with the development of sensor technology and artificial intelligence, some high-end seats have begun to explore posture recognition and fatigue detection functions based on pressure perception. For example, Voyah Automobile's AI Cloud Comfort Seat has 66 embedded high-precision sensors that can sense sitting posture and weight distribution in real time and dynamically adjust massage points; Yanfeng's new generation of zero-pressure seats uses pressure pad sensors to identify passenger body shape and sitting posture, and can detect fatigue levels to automatically trigger zoned massage; Yaole Technology's intelligent seat solution also proposes to place flexible fabric pressure sensors in areas such as the seat, backrest, and side wings to achieve human posture recognition and vital sign detection.

[0005] However, the aforementioned existing technologies still have the following shortcomings:

[0006] First, the pose recognition dimension is limited.

[0007] While existing solutions can recognize basic sitting postures, they fail to differentiate precisely from lying down. In real-world long-distance travel scenarios, passengers may naturally roll over to their side while resting on their back. When lying on their side, the body's center of gravity shifts, with the shoulders and hips becoming the main pressure points, and the muscles in the lower back and back under asymmetrical stretching. Current technology cannot effectively distinguish between these two different lying positions, meaning that if the massage function is activated in the supine mode while lying on the side, it may cause additional strain on the pressure points, potentially leading to discomfort or safety hazards (such as the shoulder airbag rising and compressing the joints when lying on the side).

[0008] Second, the massage function is not well matched with the posture.

[0009] Current seat massage triggers are mostly based on time settings or simple fatigue assessments, lacking "functional threshold" control based on refined posture. For example, when a passenger is sitting upright, the thigh muscles are under pressure and tense; activating a thigh massage at this time can produce a strong foreign body sensation and discomfort. However, when lying down, the thigh muscles are relaxed, and massage at this time can effectively promote blood circulation and relieve long-distance fatigue. Current technology has failed to establish a strict "function-posture" mapping relationship, leading to frequent instances of massage being activated on uncomfortable areas at inappropriate times.

[0010] Third, the sensor layout is not precise enough.

[0011] Existing solutions mostly use pressure sensors in single areas or at limited points, failing to design zones based on the functional characteristics of different physiological parts. For example, the shoulder is a key pressure-bearing area when lying on one's side, the lower back is the core area of ​​fatigue from sitting, and the thigh is the key discrimination area for distinguishing between sitting and lying flat. The lack of targeted zone layout limits the accuracy and anti-interference ability of posture recognition.

[0012] Fourth, it lacks smooth handling of posture transitions.

[0013] Existing technologies often switch or stop the massage function directly when a change in posture is detected. The switching process is abrupt and affects the user experience.

[0014] Therefore, there is an urgent need for a seating system that can accurately and automatically recognize various passenger postures (including sitting, lying down, and side-lying), and can intelligently select the optimal massage strategy (including enabling / disabling specific areas and adjusting massage modes) based on different postures, so as to proactively and efficiently prevent and alleviate travel fatigue while ensuring safety. Summary of the Invention

[0015] This invention designs a posture recognition method and an intelligent high-speed rail seat based on multi-region pressure sensing. The technical problem it solves is that existing intelligent seats cannot accurately identify passengers' supine and side-lying postures, which leads to the massage function being falsely triggered when lying on the side, causing discomfort. The invention achieves accurate posture identification and intelligent massage control based on multi-region pressure sensing, improving riding comfort and safety.

[0016] To solve the aforementioned technical problems, the present invention adopts the following solution:

[0017] A posture recognition method based on multi-region pressure sensing, characterized by the following steps:

[0018] A flexible thin-film pressure sensor matrix is ​​installed on the headrest, backrest, seat cushion, and leg rest of the seat to collect pressure distribution data of various parts of the passenger in real time; wherein, the backrest is divided into the upper backrest area and the lumbar backrest area, and the seat cushion is divided into the buttock area and the thigh area.

[0019] Based on the pressure distribution data, extract the pressure statistical characteristics of each region;

[0020] Using a preset posture recognition model, the passenger's current posture is determined based on the pressure statistical characteristics. The posture includes at least sitting posture, supine posture, and side-lying posture.

[0021] Based on the identified posture, corresponding control commands are generated to control the working status of the massage execution units distributed in various areas of the seat.

[0022] Furthermore, the step of determining the passenger's current posture further includes:

[0023] First, a coarse classification is performed between sitting and lying down. When the first discrimination condition is met, it is determined to be a sitting posture, and when the second discrimination condition is met, it is determined to be a lying posture.

[0024] Once the position is determined to be supine, further fine classification of supine and lateral positions is performed, and supine and lateral positions are distinguished according to the third discrimination condition.

[0025] The third criterion includes: calculating the lateral deviation of the pressure center in the upper backrest area and the pressure ratio of the left and right halves, and combining the pressure ratio of the left and right halves of the buttock area to comprehensively determine whether it is a side-lying posture.

[0026] Furthermore, the second discrimination condition includes:

[0027] The pressure in the upper backrest area, lumbar backrest area, hip area, thigh area, leg support area, and headrest area all exceeded their respective preset thresholds.

[0028] The distribution range of the pressure center along the length of the seat exceeds the preset length;

[0029] The variance of the pressure distribution in the thigh region is less than the variance threshold, exhibiting a planar support characteristic.

[0030] Furthermore, the third discrimination condition also includes:

[0031] The principal axis angle of the pressure image in the upper region of the backrest is calculated. If the principal axis angle is greater than a second angle threshold, it is used to help determine a side-lying posture. The second angle threshold (e.g., 30°) is determined based on ergonomic statistics and a large number of experimental samples, and is used to quantify the degree of body torsion. This threshold can be adaptively adjusted according to seat size, sensor accuracy, and target population (e.g., different percentiles of human bodies), and its range is typically between 25° and 35°.

[0032] A smart high-speed rail seat, employing the posture recognition method based on multi-region pressure sensing as described above, is characterized by comprising:

[0033] The multi-zone pressure sensing module includes a flexible thin-film pressure sensor matrix laid on the headrest, upper backrest, lumbar backrest, buttocks, thighs, and leg rest.

[0034] The posture recognition module is electrically connected to the multi-area pressure sensing module and identifies the passenger's current posture based on the pressure distribution data.

[0035] The intelligent massage execution module includes multiple independently controllable execution units distributed in various areas, each execution unit including an airbag and / or a massage motor;

[0036] The adaptive control module is electrically connected to the posture recognition module and the intelligent massage execution module, and generates corresponding control commands based on the recognized posture.

[0037] Furthermore, the adaptive control module is configured as follows:

[0038] When the user is identified as sitting, the massage execution unit in the thigh area is disabled, while the massage execution unit in the backrest and lumbar area is allowed to operate.

[0039] When the patient is identified as lying supine, the massage execution units in all areas are allowed to work, and a multi-point coordinated massage mode for the whole body is activated.

[0040] When the patient is identified as lying on their side, the conventional massage actuators in all areas are paused, and a micro-pressure migration mode is activated. The actuators corresponding to the high-pressure areas are controlled to perform micro-lifting movements at a level below a first amplitude threshold and a first frequency threshold. The first amplitude threshold (e.g., 0.5 cm) is the minimum controllable lifting step of the airbag, ensuring gentle and imperceptible movements. The first frequency threshold (e.g., once every 2 minutes, or 0.5 times / minute) is a preset minimum movement frequency, designed to slowly transfer pressure points at an extremely low frequency that does not disturb the passenger's sleep. These two thresholds can be preset in the adaptive control module based on actual hardware performance and control precision.

[0041] Furthermore, the full-body multi-point coordinated massage mode includes:

[0042] Gentle vibration in the headrest area;

[0043] Wave-like pressure transfer is performed in coordination between the upper backrest area and the lumbar backrest area;

[0044] Alternating lifting of the hip and thigh areas;

[0045] Periodic lifting of the leg support area;

[0046] The coordinated movement sequence in each area creates a gentle, flowing motion throughout the body from head to toe.

[0047] Furthermore, the trace pressure migration pattern includes:

[0048] Identify high-pressure areas in a side-lying position, including the upper backrest area corresponding to the shoulder and / or the buttock area corresponding to the hip.

[0049] The actuator corresponding to the high-pressure area is controlled to slowly inflate and deflate the gas in a manner with a lifting amplitude of less than 0.5 cm and an action frequency of not less than once every 2 minutes, so that the pressure point is slowly transferred.

[0050] Furthermore, the adaptive control module is also configured to perform smooth transition processing when a change in attitude is detected:

[0051] The currently performed massage movements gradually decrease until they stop within the first time period; the current massage progress is saved.

[0052] The massage movements in the new posture gradually increase to the target intensity during the second time period.

[0053] Furthermore, it also includes a reinforcement learning module, which uses the current posture, pressure distribution in each area, historical massage patterns, and passenger manual adjustment records as states, the selection and adjustment of massage strategies as actions, and passenger comfort feedback as rewards to optimize massage strategies online.

[0054] The attitude recognition method based on multi-region pressure sensing and the intelligent high-speed rail seat have the following beneficial effects:

[0055] (1) This invention divides the backrest into an upper backrest area and a lumbar backrest area, and the seat cushion into a hip area and a thigh area. Combined with a multi-area pressure sensor matrix, it can accurately identify supine and lateral lying positions. By analyzing the lateral deviation of the pressure center of the upper backrest, the pressure ratio of the left and right halves, and the unilateral high pressure characteristics of the hip area, it effectively solves the problem that the existing technology cannot distinguish between supine and lateral lying positions, laying a precise posture recognition foundation for subsequent intelligent massage control.

[0056] (2) Based on the identified different postures, this invention automatically matches differentiated massage strategies: when sitting, thigh massage is disabled to avoid discomfort caused by massage in a tense muscle state; when lying on the back, multi-point coordinated massage of the whole body is activated to achieve comprehensive relaxation from head to toe; when lying on the side, regular massage is paused and a micro-pressure migration mode is activated to maintain blood circulation while avoiding compression of the shoulder and hip joints. This posture-based "functional threshold" control completely solves the problem of existing seats activating massage of uncomfortable areas at inappropriate times, significantly improving riding comfort and safety.

[0057] (3) When the present invention detects a side-lying posture, it does not simply stop all functions, but intelligently identifies high-pressure areas such as the shoulder and hip, and performs a slight lifting motion with an extremely low amplitude (<0.5cm) and an extremely low frequency (≥2 minutes / time) to slowly shift the pressure point. This mode effectively prevents the risk of pressure sores caused by prolonged side-lying without disturbing the passenger's sleep, reflecting proactive care for human health.

[0058] (4) When the present invention detects a change in posture, it executes a smooth transition mechanism that gradually weakens to stop, saves progress, and gradually strengthens, avoiding the abruptness caused by the sudden switching of functions in traditional seats, so that passengers can have a seamless and comfortable experience when turning over or adjusting their posture.

[0059] (5) The whole body multi-point coordinated massage mode activated in the supine position of the present invention forms a whole body soothing flow from head to toe through the time sequence coordination of gentle vibration of the head pillow, wave-like pressure migration of the backrest, alternating lifting of the buttocks and thighs, and periodic relaxation of the leg support. It simulates professional physiotherapy techniques, effectively promotes blood circulation, relieves muscle fatigue, and significantly improves the comfort of long-distance travel.

[0060] (6) This invention introduces a reinforcement learning module, which can optimize the massage strategy online based on the passenger's historical usage data and manual adjustment preferences. As the number of uses increases, the seat will continuously adapt to the passenger's personalized needs, achieving intelligent evolution that understands you better with use, and providing a truly customized comfort experience. Attached Figure Description

[0061] Figure 1 : Overall system architecture diagram of the present invention;

[0062] Figure 2 : Hardware connection diagram of the present invention;

[0063] Figure 3 : Flowchart I of the attitude recognition and control of this invention;

[0064] Figure 4 : Flowchart II of the posture recognition and control of this invention. Detailed Implementation

[0065] The following is combined with Figures 1 to 4 The present invention will be further described as follows:

[0066] Example 1: As Figure 1 As shown, the control module of the present invention is as follows: a multi-area pressure sensing module, including a flexible thin-film pressure sensor matrix laid on the headrest, upper backrest, lumbar region of the backrest, buttocks, thighs, and leg rest; a posture recognition module, electrically connected to the multi-area pressure sensing module, which identifies the passenger's current posture based on pressure distribution data; an intelligent massage execution module, including multiple independently controllable execution units distributed in each area, the execution units including airbags and / or massage motors; and an adaptive control module, electrically connected to the posture recognition module and the intelligent massage execution module, which generates corresponding control commands according to the identified posture.

[0067] The adaptive control module is configured to: disable the massage execution unit in the thigh area and allow the massage execution unit in the backrest and lumbar area when the posture is identified as sitting; allow the massage execution units in all areas to work when the posture is identified as supine and start the full-body multi-point collaborative massage mode; and pause the regular massage execution units in all areas when the posture is identified as side-lying and start the micro-pressure migration mode, controlling the execution unit corresponding to the high-pressure area to perform a micro-lifting action in a mode lower than the first amplitude threshold and the first frequency threshold.

[0068] The multi-point coordinated massage mode for the whole body includes: gentle vibration in the headrest area; wave-like pressure transfer executed in coordination between the upper backrest area and the lumbar backrest area; alternating lifting in the hip and thigh areas; and periodic lifting in the leg support area. The coordinated timing of the movements in each area creates a soothing flow from head to toe throughout the body.

[0069] The system identifies high-pressure areas in the side-lying position, including the upper backrest area corresponding to the shoulder and / or the hip area corresponding to the buttocks. The actuators corresponding to these high-pressure areas are controlled to slowly inflate and deflate with a lift of less than 0.5 cm and an action frequency of at least once every 2 minutes, allowing the pressure point to gradually shift. These parameters are determined based on ergonomic experiments and subjective comfort evaluations, effectively alleviating localized pressure without waking the passenger.

[0070] The adaptive control module is also configured to perform a smooth transition when a change in posture is detected: the currently performing massage action gradually weakens to stop during the first time period; the current massage progress is saved; and the massage action under the new posture gradually increases to the target intensity during the second time period.

[0071] It also includes a reinforcement learning module, which uses the current posture, pressure distribution in each area, historical massage patterns, and passenger manual adjustment records as status, the selection and adjustment of massage strategies as actions, and passenger comfort feedback as rewards to optimize massage strategies online.

[0072] Example 2: Seat Structure and Sensor Layout

[0073] like Figure 2 As shown, the intelligent high-speed rail seat of this invention includes a headrest, backrest, seat cushion, and leg rest. Specifically: the headrest area is equipped with a 4x8 flexible thin-film pressure sensor matrix for detecting head pressure; the backrest area is equipped with a 16x20 sensor matrix for detecting back pressure distribution; the backrest area is divided into an upper backrest sensor area (corresponding to the scapula region) and a lumbar backrest sensor area (corresponding to the lumbar spine region), both equipped with a 16x20 flexible thin-film pressure sensor matrix. The value of this partitioned design lies in: the upper backrest is primarily used for lateral posture recognition—when a passenger lies on their side, the shoulder becomes the main pressure point, and the pressure value in this area increases significantly with a noticeable shift in the pressure center; the lumbar backrest is primarily used for sitting fatigue analysis—the lumbar muscles are most prone to fatigue when sitting, and the high-frequency pressure micro-motion signals in this area can be used to calculate the continuous muscle activation, providing a basis for active recovery strategies.

[0074] The seating area is divided into two parts: the buttocks area and the thigh area. A 16x16 sensor matrix is ​​installed in the buttocks area, and an 8x16 sensor matrix is ​​installed in the thigh area to distinguish between buttock and thigh weight-bearing states. An 8x16 sensor matrix is ​​installed in the leg support area to detect calf support. The value of this zoning design lies in the following: the buttocks area is the core area for determining sitting posture—the ischial tuberosity pressure is stable, which is the basic basis for determining "someone is sitting," and it is also the monitoring area for high hip pressure when lying on one's side; the thigh area is the key area for determining posture transition—when sitting, the thigh muscles are tense, and the pressure distribution is locally concentrated; when lying flat, the muscles are relaxed, and the pressure distribution is planar. By analyzing the morphological characteristics of the pressure distribution in this area, sitting and lying down can be accurately distinguished, avoiding misjudgment.

[0075] All sensors utilize piezoresistive flexible thin-film sensors based on carbon nanotube composite materials, seamlessly embedded under the seat upholstery without compromising ride comfort. The sensor matrix outputs pressure values ​​in real time at a sampling frequency of no less than 10Hz.

[0076] The seat also features multiple independently controllable airbags and vibration motors, corresponding to the headrest, backrest (zones), buttocks, thighs, and leg rest areas, for performing massage and pressure adjustment functions.

[0077] The present invention also includes a control module, a drive circuit, and an execution module.

[0078] The core of the control module is a microcontroller (MCU) or embedded processor, which runs posture recognition algorithms (based on threshold logic or AI models) and massage strategy control logic. It receives data from various sensor areas, processes it, and generates corresponding control commands, such as activating an airbag in a specific area or adjusting the vibration motor frequency. These control commands are sent to the drive circuit via PWM signals, I²C, or serial communication.

[0079] The drive circuit is divided into two parts: the airbag drive and the vibration motor drive. The airbag drive typically consists of an array of solenoid valves or a miniature air pump in conjunction with a multi-way valve, and the inflation and deflation rate is controlled by PWM. The vibration motor drive is composed of MOSFETs or H-bridge circuits, which can independently control the start / stop and vibration intensity of each motor. The drive circuit receives the low-voltage signal from the control module and converts it into a high-power signal sufficient to drive the actuator.

[0080] The actuators consist of an airbag array and a vibration motor array. The airbag array is distributed in areas such as the backrest, seat cushion, and leg rest, and can be independently inflated and deflated to achieve lifting action. The vibration motor array is also distributed in multiple points, providing vibration massage at different frequencies and intensities. The actuators perform physical actions based on the power signals output from the drive circuit, achieving dynamic pressure adjustment.

[0081] The data generated by the pressure sensor is sent to the control module for attitude recognition and fatigue assessment. The control module sends mode commands to the drive circuit, which then drives the actuator's airbag / motor.

[0082] Example 3: Pose Recognition Method

[0083] The pose recognition module of this invention employs multi-level discrimination logic, the specific process of which is as follows: Figure 3 and Figure 4 As shown:

[0084] Step 1: Data Acquisition and Preprocessing

[0085] The control module reads the pressure values ​​from all sensors at a frequency of 10Hz to construct a real-time pressure distribution map. The raw data undergoes preprocessing such as normalization and noise reduction, and the following features are extracted:

[0086] Mean, peak, and variance of pressure in each region;

[0087] The coordinates of the center of pressure (Xcog, Ycog) are calculated using a weighted average formula:

[0088] Xcog =Σ(P ij × X i ) / ΣP ij ,Ycog=Σ(P ij ×Y j ) / ΣP ij ;

[0089] Xcog is the horizontal coordinate of the pressure center (along the width of the seat, i.e., left-right); Ycog is the vertical coordinate of the pressure center (along the length of the seat, i.e., head-to-feet); P ij X represents the pressure value detected by the sensor unit located in the i-th row and j-th column; i Y represents the physical x-coordinate position of the sensor in the i-th row (with a reference point on the seat as the origin); j The physical vertical coordinate position of the sensor in column j (with a certain reference point on the seat as the origin).

[0090] Pressure asymmetry ratio between the left and right halves (calculated separately for backrest and seat cushion);

[0091] R= P left / P right ;

[0092] P left This is the sum of all sensor pressure values ​​in the left-side area (such as the left half of the backrest or the left half of the seat cushion); P right This is the sum of all sensor pressure values ​​in the right-side area (such as the right half of the backrest or the right half of the seat cushion);

[0093] The principal axis direction angle of the pressure image is calculated using the second moment.

[0094] Step 2: Seated / Lying Down Discrimination

[0095] First, a preliminary classification is performed to determine whether the passenger is sitting or lying down. The judgment rules are as follows:

[0096] The sitting and standing postures must simultaneously meet the following requirements:

[0097] Pressure in the hip area > first threshold (e.g., 5 kPa);

[0098] If the pressure in the thigh region is greater than the second threshold (e.g., 3 kPa) and the pressure distribution is locally concentrated, the pressure gradient is calculated. If the gradient value is higher than the gradient threshold, it is determined to be concentrated.

[0099] Backrest lumbar pressure > third threshold (e.g., 2 kPa);

[0100] Pressure in the leg support area is less than the fourth threshold (e.g., 0.5 kPa).

[0101] Pressure in the headrest area is less than the fifth threshold (e.g., 0.2 kPa, indicating that the head is not bearing weight).

[0102] For a supine posture to be valid, the pressure on the backrest, buttocks, thighs, leg rest, and headrest must all simultaneously exceed their respective thresholds. In this embodiment, the thresholds for determining the supine posture are set as follows: upper backrest > 1.5 kPa, lumbar backrest > 2 kPa, buttocks > 5 kPa, thighs > 3 kPa, leg rest > 1 kPa, and headrest > 0.5 kPa. It should be noted that the above values ​​are merely exemplary and can be adjusted in actual applications based on factors such as seat structure, sensor model, and target population. Those skilled in the art can determine the appropriate thresholds for specific scenarios through routine experiments.

[0103] The pressure center is distributed over a range exceeding a preset length along the length of the seat (Y-axis), such as 120cm.

[0104] The pressure distribution in the thigh area is flat (small variance), exhibiting planar support characteristics.

[0105] If the above rules cannot clearly distinguish the condition, such as a semi-reclining state, a pre-trained machine learning model (such as a random forest) is invoked for auxiliary judgment. The model input consists of the stress statistics of each region, and the output is the probability of sitting / lying down.

[0106] Step 3: Fine-grained distinction between supine and lateral positions (only when lying flat)

[0107] Once the patient is determined to be lying flat, further distinguish between supine and lateral decubitus positions (which can be further subdivided into left lateral decubitus and right lateral decubitus). Use one or a combination of the following two methods:

[0108] Method A: Rule-based discrimination based on pressure image morphology

[0109] Supine Characteristics: Lateral deviation of the pressure center refers to the distance between the pressure center and the seat's central axis in the width direction of the seat. Lateral deviation of the pressure center in the upper backrest |Xcog-Xmid| < 5cm (Xmid is the seat's central axis); the pressure ratio between the left and right halves of the upper backrest is between 0.8 and 1.2; the pressure ratio between the left and right halves of the buttocks area is between 0.8 and 1.2; the principal axis angle of the pressure image in the upper backrest is close to 0° (angle with the longitudinal direction < 15°). The principal axis angle of the pressure image refers to the angle between the principal direction calculated from the second moment of the pressure distribution image and the longitudinal axis of the seat.

[0110] Lateral lying characteristics (taking the right lateral lying position as an example): Lateral deviation of the pressure center at the upper backrest > 5cm (leaning to the right); pressure on the right side of the upper backrest is significantly higher than on the left, with a pressure ratio > 1.5; a high-pressure point appears on the right side of the buttocks (hip area), while the pressure on the left side is extremely low; the principal axis angle of the pressure image at the upper backrest > 30°. It should be noted that the angle thresholds here (e.g., <15° for supine, >30° for lateral lying) are not fixed but determined based on ergonomic statistics and a large number of experimental samples from volunteers of different body types, used to quantify the degree of body torsion. This threshold can be adaptively adjusted according to the actual size of the chair, sensor accuracy, and the target population; its reasonable range is usually between 25° and 35°, used to assist in determining whether it is a lateral lying posture.

[0111] Method B: Discrimination based on Convolutional Neural Network (CNN)

[0112] A lightweight CNN model was constructed, taking normalized pressure data from the backrest and seat cushion as input and stitching them into a 40×40 grayscale image. The output consists of three probability classes (supine, left lateral, right lateral). The model structure includes two convolutional layers (each containing 32 3×3 convolutional kernels with ReLU activation), two max pooling layers, one flattening layer, and two fully connected layers (64-dimensional and 3-dimensional), finally connected to a Softmax output. Training data was collected from pressure images of 100 volunteers of different body types in three postures, with 2000 images per class, divided into training, validation, and test sets in a 7:2:1 ratio. The test accuracy reached 98.5%. After quantization, the model was deployed to the control module, with a single inference time of less than 50ms, meeting real-time requirements.

[0113] Step 4: Maintaining State and Stabilizing Image

[0114] To prevent misjudgments due to transient interference, the system introduces a state-maintaining mechanism: the current pose is only updated when the recognition results are consistent for 5 consecutive frames (0.5 seconds). When the pose changes, the moment of change is recorded, and the original control strategy is maintained during the transition period until the new pose stabilizes before switching.

[0115] Example 4: Posture-based intelligent massage control

[0116] Based on the identified posture, the adaptive control module executes the following strategy:

[0117] (1) Sitting posture

[0118] The thigh area massage function is disabled: Regardless of whether the start command is issued manually by pressing a button or automatically by the program, the threshold control module blocks the drive signal of the actuator in this area to ensure that the thigh massage is not accidentally triggered when sitting.

[0119] The backrest and lumbar massage functions are fully functional and can be adjusted based on the user's state of consciousness (awake / lightly awake / sleeping): when awake and fatigued, an active recovery strategy is activated, such as fixed-point cyclical pressure reduction and wave-like pressure migration; when lightly awake, a guided stretching strategy is activated, using a gentle wave-like motion; when asleep, a non-invasive protection strategy is implemented, with very low-amplitude random lifting. The upper part of the backrest can cooperate with the lumbar region for wave-like pressure migration, but the amplitude is low to avoid interference.

[0120] (2) Supine position

[0121] All massage functions are available. The system automatically activates a multi-point coordinated full-body massage mode: the headrest area vibrates gently at a frequency of 20Hz and an amplitude of 0.5mm; the upper backrest and lumbar support work together to perform wave-like pressure transfer from bottom to top for 30 seconds; the lumbar support provides targeted, cyclical pressure relief, lifting the area by 1cm, holding for 5 seconds, and then lowering it for 15 seconds; the buttocks and thigh areas alternately lift, alternating left and right, for 20 seconds; the leg rest area periodically relaxes, lifting by 0.8cm, holding for 10 seconds, and then lowering for 20 seconds. The coordinated timing of each area creates a soothing full-body massage from head to toe. The massage intensity can be automatically adjusted based on passenger preference or physiological feedback (such as muscle activation).

[0122] (3) Side-lying position

[0123] Pause all regular massage actuators. Stop inflating and deflating the airbags and stop the vibration motors to avoid putting additional pressure on high-pressure areas such as the shoulders and hips.

[0124] Simultaneously, a micro-pressure migration mode is activated: it identifies high-pressure areas when lying on one's side, such as the sensor locations corresponding to the shoulder and hip; it controls the airbag under this area to slowly inflate and deflate at an extremely low amplitude, rising <0.5cm, and at an extremely low frequency (once every 2 minutes), causing the pressure point to slowly shift; the amplitude and frequency of the movement are far below the human perception threshold, so they will not disturb sleep at all, but can effectively promote local blood circulation and prevent pressure sores. Here, the first amplitude threshold (e.g., 0.5cm) is set based on the minimum controllable rise step of the airbag hardware to ensure gentle movements; the first frequency threshold (e.g., once every 2 minutes, i.e., 0.5 times / minute) is the minimum preset movement frequency to ensure that the pressure point is slowly shifted without disturbing sleep. These two thresholds are preset in the adaptive control module and can be calibrated according to the actual hardware performance and control accuracy.

[0125] When the posture returns to supine, the system automatically restores the previously saved massage mode or starts a new massage cycle.

[0126] (4) Smooth attitude switching

[0127] When a change in posture is detected, such as rolling from supine to side-lying, the system executes the following transition logic:

[0128] The massage action being performed gradually weakens and stops within 1 second, the airbag slowly deflates, and the motor frequency decreases linearly.

[0129] Save the current massage progress (mode, area, timestamp);

[0130] Once a new posture is entered, if a new strategy needs to be initiated, the actuator will gradually increase to the target intensity within 2 seconds to avoid abruptness.

[0131] Example 5: Continuous Optimization of Machine Learning Models

[0132] To adapt to individual differences and changes in preferences over long-term use, reinforcement learning mechanisms can be introduced:

[0133] Current posture, pressure distribution in each area, historical massage patterns, and passenger manual adjustment records;

[0134] Choose or adjust massage strategies (such as mode, intensity, and timing);

[0135] The calculation is based on a combination of factors, including a decrease in subsequent pressure fluctuations, a decrease in muscle activation (reflecting fatigue relief), and passenger manual adjustment behaviors (such as increasing intensity as a positive reward and turning it off early as a negative reward).

[0136] We employ a deep Q-network (DQN) for online learning and regularly update the model parameters.

[0137] Through this mechanism, the seat can continuously adapt to the passenger's preferences during use, achieving a personalized experience that becomes more and more intuitive with each use.

[0138] The present invention has been described above by way of example with reference to the accompanying drawings. Obviously, the implementation of the present invention is not limited to the above-described manner. Any improvements made using the inventive concept and technical solution of the present invention, or the direct application of the inventive concept and technical solution of the present invention to other occasions without modification, are all within the protection scope of the present invention.

Claims

1. A posture recognition method based on multi-region pressure sensing, characterized in that, Includes the following steps: A flexible thin-film pressure sensor matrix is ​​installed on the headrest, backrest, seat cushion, and leg rest of the seat to collect pressure distribution data of various parts of the passenger in real time; wherein, the backrest is divided into the upper backrest area and the lumbar backrest area, and the seat cushion is divided into the buttock area and the thigh area. Based on the pressure distribution data, extract the pressure statistical characteristics of each region; Using a preset posture recognition model, the passenger's current posture is determined based on the pressure statistical characteristics. The posture includes at least sitting posture, supine posture, and side-lying posture. Based on the identified posture, corresponding control commands are generated to control the working status of the massage execution units distributed in various areas of the seat.

2. The attitude recognition method based on multi-region pressure sensing according to claim 1, characterized in that: The step of determining the passenger's current posture further includes: First, a coarse classification is performed between sitting and lying down. When the first discrimination condition is met, it is determined to be a sitting posture, and when the second discrimination condition is met, it is determined to be a lying posture. Once the position is determined to be supine, further fine classification of supine and lateral positions is performed, and supine and lateral positions are distinguished according to the third discrimination condition. The third criterion includes: calculating the lateral deviation of the pressure center in the upper backrest area and the pressure ratio of the left and right halves, and combining the pressure ratio of the left and right halves of the buttock area to comprehensively determine whether it is a side-lying posture.

3. The attitude recognition method based on multi-region pressure sensing according to claim 2, characterized in that, The second discrimination condition includes: The pressure in the upper backrest area, lumbar backrest area, hip area, thigh area, leg support area, and headrest area all exceeded their respective preset thresholds. The distribution range of the pressure center along the length of the seat exceeds the preset length; The variance of the pressure distribution in the thigh region is less than the variance threshold, exhibiting a planar support characteristic.

4. The posture recognition method based on multi-region pressure sensing according to claim 2, characterized in that, The third discrimination condition also includes: Calculate the principal axis direction angle of the pressure image in the upper region of the backrest. If the principal axis direction angle is greater than the second angle threshold, it is used to help determine the side-lying posture.

5. A smart high-speed rail seat, employing the attitude recognition method based on multi-region pressure sensing as described in any one of claims 1 to 4, characterized in that, include: The multi-zone pressure sensing module includes a flexible thin-film pressure sensor matrix laid on the headrest, upper backrest, lumbar backrest, buttocks, thighs, and leg rest. The posture recognition module is electrically connected to the multi-area pressure sensing module and identifies the passenger's current posture based on the pressure distribution data. The intelligent massage execution module includes multiple independently controllable execution units distributed in various areas, each execution unit including an airbag and / or a massage motor; The adaptive control module is electrically connected to the posture recognition module and the intelligent massage execution module, and generates corresponding control commands based on the recognized posture.

6. The intelligent high-speed rail seat according to claim 5, characterized in that: The adaptive control module is configured as follows: When the user is identified as sitting, the massage execution unit in the thigh area is disabled, while the massage execution unit in the backrest and lumbar area is allowed to operate. When the patient is identified as lying supine, the massage execution units in all areas are allowed to work, and a multi-point coordinated massage mode for the whole body is activated. When the patient is identified as lying on their side, the regular massage execution units in all areas are paused, and a micro-pressure migration mode is activated. The execution units corresponding to the high-pressure areas are controlled to perform micro-lifting movements in a mode lower than the first amplitude threshold and the first frequency threshold.

7. The intelligent high-speed rail seat according to claim 6, characterized in that: The full-body multi-point synergistic massage mode includes: Gentle vibration in the headrest area; Wave-like pressure transfer is performed in coordination between the upper backrest area and the lumbar backrest area; Alternating lifting of the hip and thigh areas; Periodic lifting of the leg support area; The coordinated movement sequence in each area creates a gentle, flowing motion throughout the body from head to toe.

8. The intelligent high-speed rail seat according to claim 6, characterized in that: The micro-pressure migration pattern includes: Identify high-pressure areas in a side-lying position, including the upper backrest area corresponding to the shoulder and / or the buttock area corresponding to the hip. The actuator corresponding to the high-pressure area is controlled to slowly inflate and deflate the gas in a manner with a lifting amplitude of less than 0.5 cm and an action frequency of not less than once every 2 minutes, so that the pressure point is slowly transferred.

9. The intelligent high-speed rail seat according to claim 5, characterized in that: The adaptive control module is also configured to perform smooth transition processing when a change in attitude is detected: The currently performed massage movements gradually decrease until they stop within the first time period; the current massage progress is saved. The massage movements in the new posture gradually increase to the target intensity during the second time period.

10. The intelligent high-speed rail seat according to claim 5, characterized in that: It also includes a reinforcement learning module, which uses the current posture, pressure distribution in each area, historical massage patterns and passenger manual adjustment records as states, the selection and adjustment of massage strategies as actions, and passenger comfort feedback as rewards to optimize massage strategies online.

Images