Multi-user cooperative interaction method of intelligent sofa

Abstract

The application provides a multi-user cooperative interaction method of an intelligent sofa, comprising the following steps: acquiring position distribution information and pressure distribution data of multiple users on the sofa in real time through a pressure sensor array arranged on the surface of the sofa, including pressure distribution of each part of the body and position coordinates of the users on the sofa, and processing the collected data; continuously monitoring the area range and current posture of each user, and determining that the posture and position of the user have changed when detecting that the area range and the posture of the user have changed; if the change occurs, analyzing the posture and position relationship of the multiple users, judging whether there is an encroachment behavior between the users through a space collision detection algorithm, and triggering deformation intervention of the sofa and marking the user as an encroacher if the encroachment behavior is detected.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology, and in particular to a multi-user collaborative interaction method of an intelligent sofa. BACKGROUND

[0002] In the multi-user collaborative interaction scene, the intelligent sofa faces the technical challenge of meeting the comfort needs of multiple users in a limited space. Since each user has a different definition of comfortable posture and space, when one user adjusts the posture, it may encroach on the comfortable area of other users who have already adjusted. At this time, the sofa needs to change shape to prevent this encroachment and move the encroacher back to the previous area to maintain the comfort experience of multiple users. The key technical problem here is how the sofa accurately perceives the posture and position changes of multiple users and judges the occurrence of encroachment. At the same time, when performing shape intervention, it also needs to consider how to accurately control in a limited space, both moving the encroacher back to the original place and avoiding affecting other users in the comfortable area. SUMMARY

[0003] The present application provides a multi-user collaborative interaction method of an intelligent sofa, mainly including:

[0004] Through the pressure sensor array arranged on the surface of the sofa, real-time acquisition of the position distribution information and pressure distribution data of multiple users on the sofa is performed, including the pressure distribution of each part of the body and the position coordinates of the users on the sofa, and the collected data is processed;

[0005] Based on the pressure distribution data, in combination with the position distribution information, the range of the area occupied by each user is identified, including the contour boundary and the center of gravity position of each user, and through analysis of the shape and intensity of the pressure distribution, the current posture of multiple users is obtained, including whether to cross the legs and whether to tilt the body;

[0006] The range of the area of each user and the current posture are continuously monitored, and when it is detected that the range of the area of a user changes and the posture changes, it is determined that the posture and position of the user have changed;

[0007] If the change occurs, the posture and position relationship of multiple users is analyzed, and through a space collision detection algorithm, it is judged whether there is an encroachment behavior between the users, and if the encroachment behavior is detected, the shape intervention of the sofa is triggered, and the user is marked as an encroacher;

[0008] In the shape intervention, according to the specific position of the encroachment and the encroachment degree in the encroachment behavior, wherein the encroachment degree includes the encroachment area and the depth, in combination with the distribution position of the air bag and the servo motor inside the sofa, the target shape of the shape intervention is calculated, including the area height that needs to be raised, the inclination angle, and the coordination change parameters of the surrounding area, a control instruction is generated;

[0009] The control instructions are distributed to the execution units inside the sofa, synchronously driving the plurality of air bags and servo motors to realize local deformation, bulging the sofa surface in the occupied area, exerting resistance to the occupant, deforming the sofa surface in the area where the occupant originally stays, and forming a comfortable concave area around the occupant through contraction and expansion of the air bags to induce the occupant to naturally move back to the original position.

[0010] During the deformation intervention process, the posture and position changes of the user are continuously tracked, the moving speed and direction of the user's center of gravity and the change trend of the pressure of each part are monitored, and if it is detected that the user does not move as expected or a new uncomfortable posture appears, the deformation is adjusted in real time, including air bag pressure adjustment, motor drive angle correction and surface hardness adjustment.

[0011] The technical solution provided by the embodiment of the present application can include the following beneficial effects:

[0012] The present application discloses a multi-user collaborative interaction method of intelligent sofa, which acquires multi-user position and pressure distribution data in real time through a pressure sensor array, identifies user areas and postures, and continuously monitors changes. When detecting user occupation behavior, a deformation intervention mechanism is triggered, the target shape is calculated according to the occupation position and degree, and air bags and servo motors are driven to realize local deformation, exert resistance to the occupant and induce the occupant to return to the original position. During this process, the system continuously tracks user reactions and adjusts parameters in real time. The present application effectively solves the problem of space occupation when multiple people share a sofa through intelligent sensing and active intervention, improves user comfort and use experience, and realizes the intelligence and humanization of the sofa. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 The present application discloses a multi-user collaborative interaction method of intelligent sofa, which acquires multi-user position and pressure distribution data in real time through a pressure sensor array, identifies user areas and postures, and continuously monitors changes. When detecting user occupation behavior, a deformation intervention mechanism is triggered, the target shape is calculated according to the occupation position and degree, and air bags and servo motors are driven to realize local deformation, exert resistance to the occupant and induce the occupant to return to the original position. During this process, the system continuously tracks user reactions and adjusts parameters in real time. The present application effectively solves the problem of space occupation when multiple people share a sofa through intelligent sensing and active intervention, improves user comfort and use experience, and realizes the intelligence and humanization of the sofa. DETAILED DESCRIPTION

[0014] In order for those skilled in the art to better understand the technical solutions in the present specification, the technical solutions in the present specification will be described in detail below in conjunction with the drawings in the present specification. Obviously, the described embodiments are only part of the embodiments of the present specification, not all. Based on the embodiments in the present specification, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present specification.

[0015] As Figure 1 , the multi-user collaborative interaction method of intelligent sofa of the present embodiment can specifically include:

[0016] In step S101, the position distribution information and pressure distribution data of multiple users on the sofa are acquired in real time by a pressure sensor array arranged on the surface of the sofa, including the pressure distribution of each part of the body and the position coordinates of the users on the sofa, and the collected data is processed.

[0017] The pressure values of each point on the surface of the sofa are collected by the pressure sensor array, the original pressure data is smoothed after noise reduction to obtain smoothed pressure data, the smoothed pressure data is subjected to bilinear interpolation to obtain a pressure distribution matrix. According to the pressure distribution matrix, the user body contour information is extracted, and the position coordinates of the user on the sofa and the pressure values of each part of the body are calculated in combination with a pre-trained support vector regression model. The time series change of the pressure distribution matrix is acquired, a clustering algorithm is used to classify the pressure change mode, and the number of users currently on the sofa is identified, and a unique identifier is assigned to each user. The pressure data, user position coordinates, body part pressure values, user number and unique identifier data are stored in a time series database InfluxDB, and the data is indexed according to the timestamp and user identifier.

[0018] Specifically, the pressure sensor array collects pressure values at each point on the sofa surface and transmits the data to the central control unit through the low-power Bluetooth module. The original pressure data is denoised using a sliding median filter, and then a Gaussian smoothing algorithm is used to eliminate data fluctuations to obtain stable pressure values. The smoothed pressure data is bilinearly interpolated to improve the resolution of the pressure distribution map and generate a high-precision pressure distribution matrix. The user's body contour information is extracted from the pressure distribution matrix using the Canny edge detection algorithm, and the pre-trained support vector regression model is used to calculate the user's specific position coordinates and body part pressure values on the sofa. According to the time series change of the pressure distribution matrix, a pressure threshold is set to determine whether the user enters or leaves the sofa. When the pressure value of a certain area changes from below the threshold to above the threshold, it is determined that a new user has entered; otherwise, it is determined that the user has left. The K-means clustering algorithm is used to classify the pressure change patterns to identify the number of users currently on the sofa and assign each user a unique identifier, enabling simultaneous online monitoring of multiple users. The pressure sensor array continuously collects sofa surface pressure data and transmits it to the central control unit, which periodically analyzes the data according to the pre-set time interval. The Euclidean distance of the pressure distribution between two consecutive samples is calculated, and when the distance exceeds the pre-set threshold, the real-time analysis process is triggered to determine the user's posture change. For detected posture changes, the change time, pressure distribution difference before and after the change, and user identifier are recorded. The central control unit uses the time series database InfluxDB to store pressure data, including original pressure values, smoothed pressure values, high-resolution pressure distribution maps, user position coordinates, body part pressure values, user entry and exit records, and posture change information. The stored data is indexed by timestamp and user identifier for subsequent query and analysis. A retention policy is set for the stored pressure data to periodically clean up expired data, ensuring effective use of database performance and storage space. The intelligent sofa system uses a 16x16 high-sensitivity pressure sensor array, with each sensor capable of detecting a pressure range of 0-100kg and a sensitivity of 0.1kg. The sensors collect data at a frequency of 100Hz and transmit it to the central control unit through the low-power Bluetooth 5.0 module. The original data is first filtered using a sliding median filter with a window size of 5 to remove outliers, and then a Gaussian smoothing algorithm with a standard deviation of 1.5 is applied to eliminate data fluctuations. The 16x16 pressure matrix after smoothing is bilinearly interpolated to obtain a 64x64 high-resolution pressure distribution map. The Canny edge detection algorithm uses a low threshold of 30 and a high threshold of 90 to extract the user's body contour, and combines a pre-trained support vector regression model to calculate the user's position coordinates and body part pressure values. The pressure threshold is set to 2kg, and when the pressure of a certain area changes from below the threshold to above the threshold for three consecutive sampling periods, it is determined that a new user has entered; otherwise, it is determined that the user has left.The K-means clustering algorithm is initially set to have a maximum number of users of 4, and the optimal number of clusters is evaluated by a silhouette coefficient to achieve multi-user identification. Periodic analysis is performed every 0.5 seconds, and the Euclidean distance of the pressure distribution of adjacent two samples is calculated. When the distance exceeds 10 kg, real-time analysis is triggered. The posture change record includes the change timestamp, the difference matrix of the pressure distribution before and after the change, and the user identifier. InfluxDB is used to store the pressure data, and the data retention policy is set to 30 days. The timestamp and user identifier are used as index fields. Data queries use InfluxQL language, which supports flexible queries based on time range, user identifier, pressure threshold, and other conditions.

[0019] In step S102, based on the pressure distribution data, combined with the position distribution information, the area range occupied by each user is identified, including the contour boundary and the center of gravity position of each user. By analyzing the shape and intensity of the pressure distribution, the current posture of multiple users is obtained, including whether to cross the legs and whether to tilt the body.

[0020] The pressure distribution data is image segmented, and seed points are identified according to local pressure peak points. The watershed algorithm is performed on the seed points to obtain the preliminary area range of each user. The contour boundary of each user is extracted according to the preliminary area range. The center of gravity position is obtained by calculating the average value of the coordinates of all points within the contour boundary. It is judged whether the area where the center of gravity position is located is a valid area. The shape feature of the pressure distribution is calculated for the valid area. The shape feature is compared. The current posture of the user is determined according to the comparison result. The current posture of the user is analyzed, and the offset distance between the pressure center of gravity and the geometric center of gravity in the user area is calculated. If the offset distance exceeds a predetermined threshold, it is determined as a tilted body posture. The Kalman filtering method is used to track the changes of the center of gravity position and the posture parameters. The posture transition probability matrix is established according to the changes. The next posture state of the user is predicted through the posture transition probability matrix.

[0021] Specifically, the central control unit receives the pressure distribution data, and uses a watershed algorithm to perform image segmentation on the data, with a pressure threshold of 5 Newtons per square centimeter. First, identify the pressure peak points in the local 3x3 region as seed points, and start the watershed algorithm from these seed points until a point below the threshold is encountered or contact with another region is made, to obtain the preliminary range of each user's region. According to the preliminary range of the region, the Moore neighborhood tracking algorithm is used to extract the contour boundary of each user, and the barycentric position is obtained by calculating the average value of the coordinates of all points in the contour. When judging the effective region, in addition to considering the area, the pressure gradient of adjacent points in the region is also calculated, and the gradient threshold is set to 2 Newtons per square centimeter per pixel. If the area of the region is less than 1000 square centimeters or the pressure gradient is less than the threshold, it is determined to be an invalid region and is excluded. For each valid user region, the shape features of the pressure distribution are calculated, including the aspect ratio, circularity, and rectangularity of the region. A database containing different posture features is established, and a shape matching algorithm based on Fourier descriptors is used for comparison to determine the current posture of the user. If the aspect ratio is greater than 1.5 and the circularity is less than 0.6, and the pressure distribution shows obvious asymmetry with a left-right pressure difference of more than 20%, it is determined to be a cross-legged posture. The intensity variation of the pressure distribution is analyzed, and the offset distance between the pressure barycenter and the geometric barycenter in the user region is calculated. If the offset distance exceeds 20% of the length of the region, it is determined to be a tilted body posture, and the tilt direction and angle are recorded. For the case of multiple users tilting at the same time, the tilt state is calculated separately for each user region, and the tilt direction relative to the overall pressure distribution is recorded. The Kalman filter is used to track the changes in the user's barycentric position and posture parameters, and a posture transition probability matrix is established to predict the user's next possible posture state. The intelligent sofa system uses a 64x64 high-precision pressure sensor array, each sensor can detect a pressure range of 0-100N, with a sensitivity of 0.1N. After the central control unit receives the pressure distribution data, it first applies a 3x3 Gaussian filter to smooth the data and eliminate noise. Then use the watershed algorithm for image segmentation, set the pressure threshold to 5N / cm 2 . The algorithm identifies the pressure peak points in the local 3x3 region as seed points, and starts the watershed process from these points. Each region obtained by segmentation is extracted by the Moore neighborhood tracking algorithm to extract the contour, and the barycentric position is obtained by calculating the average value of the coordinates of all points in the contour. When judging the effective region, set the area threshold to 1000cm 2 , and the pressure gradient threshold to 2N / cm 2 / pixel. For the valid user area, shape features are calculated: aspect ratio, circularity (4π x area / perimeter2) and rectangularity (area / (length x width)). A posture feature database is built, containing Fourier descriptors of 10 common postures. The descriptor of current posture is calculated using fast Fourier transform, compared with descriptors in the database using Euclidean distance, and the closest posture is selected. If the aspect ratio > 1.5, circularity < 0.6, and the left-right pressure difference > 20%, the crossed legs posture is determined. The offset of pressure barycenter and geometric barycenter is calculated, and if it exceeds 20% of the area length, the leaning posture is determined. A Kalman filter is used to track the barycenter position changes, with the state vector containing x, y coordinates and their velocities, the measurement noise covariance set to 0.1, and the process noise covariance set to 0.01. A 5x5 posture transition probability matrix is built based on historical data, used to predict the next posture state. The analysis results are updated every 0.1 seconds, providing real-time user behavior data for smart home control.

[0022] In step S103, the area range and current posture of each user are continuously monitored. When a change in the area range and posture of a certain user is detected, it is determined that the posture and position of the user have changed.

[0023] The pressure sensor array data is obtained, and based on the data, the pressure difference value of each sensor point from the previous frame is calculated to obtain the average change rate of the pressure difference value. If the average change rate exceeds a preset threshold, edge detection processing is triggered. The Canny edge detection algorithm is used to process the pressure distribution map to extract the user area boundary contour in the pressure distribution map. The area and perimeter of the user area are calculated based on the boundary contour, and the Hu moment is extracted as a shape descriptor. If the change rate of the area, perimeter, or shape descriptor exceeds the corresponding preset threshold, it is determined that the user area range has changed. Principal component analysis algorithm is applied to the user area to extract the main feature vector of the pressure distribution. The local binary pattern feature of the pressure distribution is calculated. The main feature vector and the local binary pattern feature are combined to form a comprehensive posture feature descriptor. The similarity between the comprehensive posture feature descriptor sequence and a preset standard posture sequence is calculated. If the similarity is lower than a preset threshold, it is determined that the user posture has changed. A Kalman filter is used to track the change trend of the comprehensive posture feature descriptor. If the deviation between the feature value predicted by the Kalman filter and the actual observed value exceeds a preset threshold, it is confirmed whether the posture has changed. User merging or separation events are identified by monitoring the distance change between adjacent user areas.

[0024] Specifically, the pressure sensor array data is collected in real time, with one frame of data collected every 50 milliseconds. For each sensor point, the pressure difference with the previous frame is calculated, and then the average rate of change of the entire region is calculated to obtain the real-time pressure rate of change. The pressure rate of change threshold is set to 5%, and if it exceeds this threshold, the subsequent processing flow is triggered. The Canny edge detection algorithm is used to process each frame of pressure distribution map to extract the boundary contour of the user area. The area, perimeter, and Hu moments are calculated as shape descriptors. The area change rate threshold is set to 10%, the perimeter change rate threshold is set to 5%, and the shape similarity threshold is set to 0.9. If any of the indicators exceeds the threshold, it is determined that the user area range has changed. Each detected user area is assigned a unique identifier to enable multi-user parallel processing. Principal component analysis algorithm is applied to each user area to extract the main feature vector of the pressure distribution. At the same time, the local binary pattern feature is calculated to capture the texture information of the pressure distribution. The principal component analysis result and the local binary pattern feature are combined to form a comprehensive gesture feature descriptor. A feature sequence is constructed, containing the feature descriptors of the last 10 frames. The dynamic time warping algorithm is introduced to calculate the similarity between the current gesture feature sequence and the preset standard gesture sequence. The similarity threshold is set to 0.8, and if it is lower than the threshold, it is determined that the user gesture has changed. The Kalman filter is used to track the trend of the gesture feature change, and the filter state vector contains the components of the feature descriptor. If the deviation between the predicted feature value of the Kalman filter and the actual observed value exceeds the preset threshold, the gesture change is further confirmed. At the same time, an interaction detection mechanism between users is established to identify possible user merging or separation events by monitoring the distance change of adjacent user areas. The intelligent sofa system uses a 64x64 high-precision pressure sensor array, and collects one frame of data every 50 milliseconds. The system calculates the pressure difference of each sensor point with the previous frame in real time, and obtains the average rate of change of the entire region. When the rate of change exceeds 5%, the Canny edge detection algorithm is triggered, with the low threshold set to 30 and the high threshold set to 90. After extracting the user area contour, the area, perimeter, and 7 Hu moments are calculated as shape descriptors. The conditions for determining area change are area change rate > 10%, or perimeter change rate > 5%, or shape similarity < 0.9. Each user area is assigned a unique 8-bit hexadecimal identifier. For the detected user area, the principal component analysis algorithm is applied to extract the first 5 principal components, and the local binary pattern feature of a 3x3 neighborhood is calculated to obtain a 256-dimensional feature vector. The two features are combined to form a comprehensive gesture feature descriptor, and the last 10 frames of feature sequence are saved. The dynamic time warping algorithm is used to calculate the similarity between the current feature sequence and the preset standard gesture sequence, with the similarity threshold set to 0.8. The Kalman filter is used to track the change of the gesture feature, with the state vector containing 261 components (5 principal components + 256-dimensional LBP feature), the measurement noise covariance set to 0.1, and the process noise covariance set to 0.01.If the Euclidean distance between the filter prediction and the observation exceeds 0.5, the posture change is confirmed. The distance change of adjacent user areas is also monitored, and when the distance between the edges of two user areas is less than 10 cm, merging detection is triggered; when a single user area is obviously separated, separation detection is triggered.

[0025] In step S104, if a change occurs, the posture and position relationship of multiple users is analyzed, and whether there is an invasion behavior between users is judged by a spatial collision detection algorithm. If an invasion behavior is detected, the sofa deformation intervention is triggered, and the user is marked as an invader.

[0026] Obtain pressure sensor array data, extract a posture contour according to the pressure sensor array data, and calculate the centroid coordinates and area through the posture contour. Determine the minimum circumscribed rectangle by using the rotation lock method, generate a bounding box according to the minimum circumscribed rectangle, and the bounding box includes vertex coordinates and edge length. Divide the sofa surface into spatial units, and the size of the spatial unit does not exceed a preset threshold. Mark the occupied spatial unit according to the position of the bounding box, and obtain the spatial overlap area by comparing the occupied units of adjacent users. Judge whether the area proportion of the spatial overlap area exceeds a preset invasion threshold, if the area proportion exceeds the preset invasion threshold, it is determined that an invasion behavior occurs. If it is determined that an invasion behavior occurs, activate the airbag inflation mechanism corresponding to the spatial overlap area.

[0027] Specifically, according to the pressure sensor array data, the Canny edge detection algorithm is used to extract the posture contour of each user, and the centroid coordinates and area of the contour are calculated. The posture features are represented by shape context descriptors, and the contour changes between adjacent time frames are compared to determine whether the user's posture has changed significantly. The area change threshold is set to 10%, the centroid displacement threshold is 5 cm, and the posture feature similarity threshold is 0.8. The rotating jigsaw method is used to calculate the minimum bounding rectangle, first find the extreme points of the contour, then traverse all possible rectangular directions, and calculate the minimum area rectangle. A bounding box is generated for each user's posture contour, recording the vertex coordinates and edge length of the bounding box, constructing a space occupation model, and recording the spatial position and size information of each user. The quadtree algorithm is applied to divide the sofa surface into multiple spatial units. Starting from the root node, recursively divide the space into four sub-regions until the preset minimum unit size of 5x5 cm or only one user is included in the node. According to the position of the user's bounding box, mark the occupied space unit, and detect the spatial overlap area by comparing the occupied units of adjacent users. If spatial overlap is detected, trigger the intrusion behavior judgment logic. Calculate the area ratio of the overlap area, and set the intrusion threshold to 20%. Record the duration of the intrusion state, and set the time threshold to 3 seconds. When the intrusion state lasts longer than the time threshold, calculate the intrusion degree of each user, which is the ratio of the overlap area to the total area, and select the user with the highest intrusion degree as the main intruder. At the same time, record the intrusion judgment results in the past 30 seconds, if there is frequent state switching (more than 5 times), the intrusion threshold is temporarily increased to 30%. After confirming the intrusion behavior, activate the airbag inflation mechanism in the corresponding area to realize local deformation intervention. The intelligent sofa system uses a 64x64 high-precision pressure sensor array, and collects a frame of data every 50 milliseconds. The system uses the Canny edge detection algorithm to extract the user contour, and the parameter settings are low threshold 30 and high threshold 90. Calculate the shape context descriptors of 60 equally spaced sampling points for each contour, forming a 300-dimensional feature vector. Compare the feature vectors of adjacent frames to determine posture changes using cosine similarity, with a threshold of 0.8. At the same time, calculate the contour area and centroid position, with an area change threshold of 10% and a centroid displacement threshold of 5 cm. The rotating jigsaw method is used to calculate the minimum bounding rectangle, taking 100 angles to find the smallest rectangle. The quadtree space division starts from a root node of 128x128 cm, and recursively divides until the leaf node size is 5x5 cm or only one user is included. Spatial overlap detection is achieved by comparing the occupancy status of the leaf nodes, with an overlap area ratio threshold of 20%. A 30-second sliding time window is maintained to record the intrusion state every 100 milliseconds. If the intrusion is detected continuously for 3 seconds and the state switching does not exceed 5 times within 30 seconds, the intrusion behavior is confirmed. Calculate the intrusion degree of each overlapping user (overlap area / user total area), and select the highest one as the intruder.The air bag inflation mechanism is distributed on the surface of the sofa, and each 5*5 cm area corresponds to an independently controlled air bag, the inflation pressure range is 0-50 kPa, the response time is less than 100 ms, and precise local deformation intervention is realized.

[0028] According to the posture and position relationship of multiple users, a bounding box is established for each user, a suspected encroachment behavior is identified, and whether the encroachment behavior occurs is judged through an encroachment range threshold and an encroachment time threshold. If it is determined that the encroachment behavior occurs, encroachment information is output, including the identity of the involved user, the location of the encroachment area and the time.

[0029] According to the pressure sensor array data, the main features of the pressure distribution are extracted, and the minimum circumscribed rectangle is calculated by combining the rotating calipers method to obtain the vertex coordinates, length, width and pressure center point of the user bounding box; the generalized collision detection is performed on the user bounding box, if the calculated nearest point pair distance is less than a preset threshold, it is determined that a collision occurs; for the user pair that the collision occurs, the intersection area is calculated, if the proportion of the intersection area to the area of the respective bounding box exceeds a preset threshold, it is determined that a suspected encroachment behavior occurs; for the suspected encroachment behavior, a sliding time window monitoring mechanism is started, if the continuous encroachment state duration in the sliding time window exceeds a preset threshold, it is determined that an encroachment behavior occurs; the k-nearest neighbor algorithm is used to match the identity of the user pressure distribution features, to obtain the unique identifiers of the encroacher and the encroached, and to output the encroachment information containing the unique identifiers, the encroachment area coordinate range and the encroachment start time.

[0030] Specifically, according to the pressure sensor array data, the principal component analysis algorithm is used to extract the main features of the pressure distribution, and the minimum enclosing rectangle is calculated by the rotating calipers method to establish the bounding box for each user. The coordinates of the top point, length, width and pressure center point of the bounding box are recorded. The changes of the bounding box and the principal component features of adjacent time frames are compared to determine the user posture and position changes. The position change threshold is set to 5 cm, and the feature similarity threshold is set to 0.9. The Gilbert-Johnson-Keerthi algorithm is used for generalized collision detection, which is suitable for user postures that may form non-convex shapes. Collision judgment is performed on each pair of user bounding boxes, and the distance between the nearest point pairs is calculated. If the distance is less than the preset threshold of 1 cm, it is determined that a collision has occurred, and the colliding user pair is recorded. For the user pairs detected to collide, the scan line algorithm is used to calculate the intersection area. Starting from the minimum value of the y-axis, the intersection points are determined by scanning line by line to construct the intersection polygon. The area of the intersection region and its proportion of the area of each bounding box are calculated, and the encroachment range threshold is set to 20%. If the threshold is exceeded, it is marked as a suspected encroachment behavior. For suspected encroachment behavior, a sliding time window monitoring mechanism is started, with a window size of 10 seconds and a step size of 1 second. The encroachment time threshold is set to 3 seconds. If the encroachment state persists for 3 seconds continuously within the window, it is determined to be an encroachment behavior. An encroachment relationship graph is constructed, with nodes representing users and edges representing encroachment relationships. Through the connectivity analysis of the graph, complex multi-person encroachment situations are identified. The k-nearest neighbor algorithm is used for user pressure distribution feature identification, and the output encroachment information includes the unique identifiers of the encroacher and the encroached, the coordinate range of the encroached area and the start time of the encroachment. The intelligent sofa system uses a 64x64 high-precision pressure sensor array, and collects a frame of data every 50 milliseconds. The system first applies the principal component analysis algorithm to the pressure data, and extracts the first five principal components as the posture features. At the same time, the minimum enclosing rectangle is calculated using the rotating calipers method, and the area of the smallest rectangle is found by sampling 100 angles. The system compares the principal component features of adjacent frames, and uses cosine similarity to judge the posture change, with a threshold of 0.9. Position change is achieved by calculating the Euclidean distance of the pressure center point, with a threshold of 5 cm. The Gilbert-Johnson-Keerthi algorithm is used for collision detection, with a minimum separation distance of 1 cm. After detecting a collision, the scan line algorithm is used to calculate the intersection area, with a scanning accuracy of 1 mm. The system maintains a 10-second sliding time window, which is updated every second, and records the encroachment state. If encroachment is detected for 3 seconds continuously, and the intersection area exceeds 20% of the area of any user bounding box, the encroachment behavior is confirmed. The adjacency matrix is used to represent the encroachment relationship graph, with a matrix size of n x n, where n is the current number of users. Through the depth-first search algorithm, the connectivity of the graph is analyzed to identify complex multi-person encroachment situations. The user identity recognition uses the k-nearest neighbor algorithm, with a k value of 3. The feature vector includes the principal components of the pressure distribution and the bounding box size, with a total of 7 dimensions.The system maintains a user feature database with a capacity of 100 for identity matching. The final output is the intrusion information, including a 10-character unique identifier of the user, the coordinates of the intrusion area in centimeters, and the intrusion start timestamp accurate to milliseconds.

[0031] In step S105, in the deformation intervention, according to the specific intrusion position and the intrusion degree in the intrusion behavior, wherein the intrusion degree includes the intrusion area and the depth, combined with the distribution positions of the air bags and the servo motors inside the sofa, the target shape of the deformation intervention is calculated, including the region height to be raised, the inclination angle, and the coordination change parameters of the surrounding region, to generate the control instruction.

[0032] The intrusion behavior data carrying the intrusion position coordinates and the intrusion degree parameters are received; according to the intrusion position coordinates and the intrusion degree parameters, the k-nearest neighbor algorithm combined with the bilinear interpolation is used to obtain the affected execution unit list, which contains the grid coordinates of the air bags and the servo motors inside the sofa; based on the intrusion degree parameters, the preset fuzzy rule set is used to calculate the strength coefficient of the deformation intervention, which is used to adjust the subsequent deformation parameters; the strength coefficient and the intrusion position coordinates are used to generate the target surface of the deformation intervention by the cubic spline interpolation algorithm, the target surface takes the intrusion position as the center, selects the surrounding control points and sets the natural boundary condition; based on the strength coefficient of the deformation intervention, the region height to be raised and the inclination angle are determined,

[0033] h=k*sin(θ)

[0034] In the formula, h represents the height of the region, k represents the strength coefficient, and θ represents the inclination angle,

[0035]

[0036] In the formula, θ represents the inclination angle, h represents the height, and l represents the horizontal distance; the target surface is discretized into a grid point set, which is mapped to the air bag and servo motor positions in the execution unit list, and the target state of each execution unit is calculated, including the air bag inflation amount and the servo motor rotation angle; a specific control instruction sequence is generated, which includes the unique identifier of the execution unit, the target state value, and the execution timestamp.

[0037] Specifically, the central control unit receives the encroachment data, including the encroachment position coordinates and the encroachment degree parameters. The encroachment degree parameters include the encroachment area percentage and the average pressure value. Using the k-nearest neighbor algorithm combined with bilinear interpolation, the encroachment position is accurately mapped into the grid coordinate system of the airbags and servo motors inside the sofa, obtaining the list of affected execution units. According to the encroachment area percentage and the average pressure value, the strength coefficient of the deformation intervention is calculated using a fuzzy logic controller. The controller uses a triangular membership function to define three levels of "low", "medium", and "high". The input variables are derived through a pre-set fuzzy rule set, such as "if the encroachment area is large and the pressure is high, then the strength is high", to derive a strength coefficient between 0 and 1, which is used to adjust the subsequent deformation parameters. The cubic spline interpolation algorithm is used to generate the target surface of the deformation intervention. Selecting the surrounding control points with the encroachment position as the center, the natural boundary condition is set. By solving the cubic polynomial coefficients, a smooth three-dimensional surface model is generated. The surface height and slope are adjusted by the strength coefficient to determine the height and inclination angle of the area that needs to be raised. The generated three-dimensional surface model is discretized into a grid point set, corresponding to the positions of the airbags and servo motors inside the sofa. The target state of each execution unit is calculated, including the airbag inflation volume and the servo motor rotation angle. Considering the physical limitations of the execution units, such as the maximum airbag inflation volume of 500 milliliters and the maximum servo motor rotation angle of 180 degrees. A control instruction sequence is designed for step-by-step execution, with each step execution time not less than 100 milliseconds. A specific control instruction sequence is generated, which contains the unique identifier of the execution unit, the target state value, and the execution timestamp. The intelligent sofa system uses a 16x16 execution unit grid, each unit containing an airbag and a servo motor. After the central control unit receives the encroachment data, it first locates the nearest execution unit using the 3-nearest neighbor algorithm, and then applies bilinear interpolation to accurately calculate the encroachment position. For example, for an encroachment position with coordinates (7.3, 8.6), it is located at (7, 8), (7, 9), (8, 8), and (8, 9) four adjacent units, and the accurate mapping is obtained through interpolation. The fuzzy logic controller uses a triangular membership function to divide the encroachment area percentage and the average pressure value into three levels: low 0-30%, medium 20-70%, and high 60-100%. Through 9 fuzzy rules, such as "if the area is high and the pressure is high, then the strength is 0.9", the strength coefficient is calculated. The cubic spline interpolation algorithm selects a 5x5 control point grid and sets the natural boundary condition, where the second derivative at the boundary (curve) is 0. By solving 16 cubic polynomial equation systems, a smooth surface is generated. For example, the center point height is set to 10 cm x strength coefficient, and the edge point height is 0, generating a raised area with a maximum height of 9 cm. The physical limitations of the execution units consider the maximum airbag inflation volume of 500 ml and the maximum servo motor rotation angle of 180 degrees. The surface is discretized into a 16x16 grid corresponding to the execution units.The control instruction sequence is executed step by step with an increment of 10%, and each step interval is 100 ms, and the total execution time is about 1 second. Each instruction contains a 4-bit unit ID, a 3-bit inflation amount (ml), and a 3-bit angle value.

[0038] In step S106, the control instructions are distributed to the execution units inside the sofa, and the plurality of air bags and servo motors are synchronously driven to realize local deformation, the sofa surface of the occupied area is raised, resistance is applied to the occupant, the sofa surface of the original position of the occupant is deformed, and a comfortable concave area is formed around the occupant through the contraction and expansion of the air bag to induce the occupant to naturally move back to the original position.

[0039] The control instructions sent by the central control unit through the CAN bus are obtained, and the control instructions include target air bag inflation amount and servo motor rotation angle; according to the control instructions, the execution unit adopts a precise clock synchronization protocol to complete instruction execution within a preset millisecond synchronization time window; if the execution unit of the occupied area receives the control instruction, the air bag inflation pump is controlled to increase the inflation amount to a preset air pressure, and the servo motor is driven to adjust the angle of the support structure to a preset angle; if the execution unit of the original position area receives the deformation instruction, the air bag is controlled to periodically contract and expand, wherein the contraction time is a preset value, and the expansion time is a preset value; it is judged whether there are multiple occupants, if there are, a weighted method based on the occupation degree is used to allocate resources, wherein the higher the occupation degree, the more execution units are allocated.

[0040] Specifically, the central control unit distributes control commands to the execution units inside the sofa using a CAN bus. Each execution unit is equipped with an independent microcontroller that receives and interprets the commands. The commands contain the target airbag inflation volume and the servo motor rotation angle. The execution units are connected to the airbag inflation pump and servo motor through a local bus, enabling precise control. The execution units use a precise clock synchronization protocol to ensure the synchronization of multiple airbags and servo motors. A 50-millisecond synchronization time window is set, and the execution of the command is completed within this window. The execution unit sends an acknowledgment signal immediately after receiving the command, and if the acknowledgment is not received within the specified time, the central control unit will resend the command. The execution speed is adjusted in real-time through feedback signals to ensure the smoothness of the deformation process. The execution units in the intrusion area control the airbag inflation pump to increase the inflation volume based on the received command, while driving the servo motor to adjust the angle of the support structure, achieving local bulging. The bulging height is linearly mapped to the intrusion level, with a maximum bulging height of 10 centimeters. The stiffness is controlled by the airbag pressure and the servo motor angle, with a pressure range of 0.5 to 2 atmospheres and an angle range of 0 to 45 degrees, forming a progressive resistance to the intruder. The execution units in the original position area receive the deformation command and control the airbag to contract and dilate periodically. The contraction time is set to 1 second, and the dilation time is 2 seconds. The airbag capacity varies between 80% and 120% of the baseline value, and the output of the inflation pump is accurately adjusted by a PID controller. The depth of the concave changes sinusoidally over time, with a period of 3 seconds and an amplitude of 2 centimeters. At the same time, the user's response is monitored in real-time by a pressure sensor, and if a sudden pressure change is detected, the deformation parameters are immediately adjusted to ensure user comfort. For multiple intruders, a weighted algorithm based on the intrusion level is used to allocate resources, with higher intrusion levels receiving more execution units. The intelligent sofa system uses a 16x16 grid of execution units, with communication through a 500kbps CAN bus. Each execution unit is equipped with a 32-bit ARM Cortex-M4 microcontroller running at 80MHz. The central control unit uses the IEEE 1588 precision clock synchronization protocol, with a synchronization accuracy of 1 microsecond. The control command contains an 8-bit unit ID, a 12-bit inflation volume, and a 9-bit angle value. The execution unit sends an acknowledgment signal within 10ms of receiving the command, and if the acknowledgment is not received within 15ms, the command is resent. The bulging height in the intrusion area is calculated using linear interpolation, with an intrusion level of 0-100% corresponding to a height of 0-10cm. The airbag pressure control uses a proportional valve with a resolution of 0.01 atmospheres. The servo motor uses PWM control with a resolution of 0.1 degrees. The airbag capacity variation in the original position area uses a PID controller with Kp=0.5, Ki=0.1, and Kd=0.05, with a control period of 10ms. The sine wave generation uses a lookup table method with a table size of 1000 points. The pressure sensor has a sampling rate of 100Hz and a resolution of 0.1N. The sudden change detection threshold is set to a pressure change of more than 20N within 1 second. The multi-intruder resource allocation uses a weighted round-robin algorithm, with the weight proportional to the intrusion level.The system response time is no more than 200 ms from detection to execution completion, ensuring real-time user experience. The intelligent sofa can quickly and smoothly respond to various complex intrusion situations and provide precise local deformation adjustment.

[0041] In step S107, during the deformation intervention process, the user's posture and position changes are continuously tracked, the movement speed and direction of the user's center of gravity are monitored, and the change trend of the pressure of each part is monitored. If it is detected that the user does not move as expected or a new uncomfortable posture appears, the deformation is adjusted in real time, including air bag pressure adjustment, motor drive angle correction, and surface hardness adjustment.

[0042] The user body pressure distribution data is collected in real time using a pressure sensor array, and the user's center of gravity position is calculated using a centroid calculation algorithm. According to the user body pressure distribution data, the user's posture is classified using a support vector machine algorithm to obtain the current posture type of the user. A posture change threshold and a center of gravity movement threshold are set. If it is detected that the user's posture change amplitude exceeds the preset threshold or the center of gravity movement speed is lower than the expected value, the deformation parameter adjustment is triggered. A decision tree is used to store the deformation strategy, and the nodes of the decision tree include posture type, center of gravity position, and pressure distribution characteristics. According to the current posture type, center of gravity position, and pressure distribution characteristics of the user, the decision tree is traversed to obtain the deformation strategy. According to the deformation strategy, a control instruction sequence is generated, which includes air bag pressure adjustment amount, motor drive angle correction value, and surface hardness adjustment parameter. The control instruction sequence is sent to the relevant execution unit, and the execution unit executes the adjustment after receiving the control instruction sequence and returns the execution result.

[0043] Specifically, high-precision pressure sensor array is used to collect user body pressure distribution data in real time, and the user's center of gravity position is calculated by the center of mass calculation algorithm. The particle filter is used to smooth the center of gravity position to obtain the moving speed and direction of the center of gravity. At the same time, the exponential weighted moving average calculation is performed on the pressure values of each part to obtain the pressure change trend, and the sampling period is 20 milliseconds. Based on the pressure distribution data, the support vector machine algorithm is used to classify the user's posture, and the pre-trained model contains 10 common postures such as normal sitting posture, inclination and cross-legged posture. Posture recognition is performed every 200 milliseconds, and the recognition result is compared with the expected moving direction. The mode of the continuous 5 recognition results is calculated as the current stable posture. The posture change threshold and the center of gravity movement threshold are set. If the user's posture change amplitude exceeds the preset threshold or the center of gravity moving speed is lower than 50% of the expected value, the morphing parameter adjustment process is triggered. The decision tree is used to store the morphing strategy, and the nodes of the tree contain the posture type, the center of gravity position and the pressure distribution characteristics. The leaf nodes store the corresponding morphing parameters. According to the current state, the decision tree is traversed to obtain the suitable morphing strategy. According to the obtained morphing strategy, a new control instruction sequence is generated, including the air bag pressure adjustment amount, the motor drive angle correction value and the surface hardness adjustment parameter. The priority queue is used to process multiple adjustment instructions, and the priority is determined based on the morphing urgency. The instruction is sent to the relevant execution unit through the CAN bus, and the execution unit executes the adjustment immediately after receiving the instruction and returns the execution result. The gradual adjustment is realized by using the tamping function to avoid discomfort caused by sudden changes. At the same time, the pressure change rate is monitored. If it exceeds the preset threshold, the adjustment is suspended and the last comfortable state is returned. The intelligent sofa system uses a 64x64 high-precision pressure sensor array with a sampling rate of 50Hz. The weighted average method is used for center of mass calculation, and the particle filter uses 1000 particles for state estimation. The decay factor of the exponential weighted moving average is set to 0.2. The support vector machine uses the RBF kernel function, C=1.0, gamma=0.1, and the accuracy rate of 93% is obtained after training on 10000 samples. The posture change threshold is set to 3 times of the continuous recognition result, and the center of gravity movement threshold is 2cm / s. The depth of the decision tree is 5, containing 32 leaf nodes, each node storing 5 key morphing parameters. The morphing strategy priority is divided into 3 levels, and the priority queue is realized by using the heap. The baud rate of the CAN bus is 1Mbps, and the single communication delay is not more than 2ms. The air bag pressure adjustment range of the execution unit is 0-500kPa, with an accuracy of 0.5kPa; the motor angle adjustment range is ±30°, with an accuracy of 0.1°; the hardness adjustment range is 10-100N / cm, with an accuracy of 1N / cm. The tamping function uses a cubic Bezier curve, and the adjustment time is 500ms-2s. The pressure change rate threshold is set to 20% / s, and the rollback mechanism is triggered when it exceeds.

[0044] It will be apparent to those skilled in the art that the application is not limited to the details of the above-exemplified embodiments and that the present application can be implemented in other particular forms without departing from the spirit or essential characteristics thereof. The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the application being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. No reference herein to any prior art is to be taken as an admission that the application is not entitled to antedate such prior art by virtue of prior application. Any reference to a "patent" or "patent document" is not intended as an admission that the patent or patent document is prior art with respect to the present application. Any reference to a document or other item of prior art is not intended as an admission that the document or item was at the priority date either known to be relevant to an issue on sale, novelty or non-obviousness under the patent laws or considered important to an invention covered by the application.

Claims

1. A multi-user collaborative interaction method for a smart sofa, characterized in that: The method comprises: Through the pressure sensor array arranged on the sofa surface, the position distribution information and pressure distribution data of multiple users on the sofa are obtained in real time, including the pressure distribution of various parts of the body and the position coordinates of the users on the sofa, and the collected data is processed; Based on the pressure distribution data and position distribution information, the area occupied by each user is identified, including the outline boundary and center of gravity of each user. By analyzing the shape and intensity of the pressure distribution, the current posture of multiple users is obtained, including whether they cross their legs or lean to the side. Continuously monitor the area range and current posture of each user. When a change in the area range and posture of a user is detected, it is determined that the user's posture and position have changed; If there is a change, the posture and position relationship of multiple users is analyzed, and a spatial collision detection algorithm is used to determine whether there is any encroachment between users. If encroachment is detected, the sofa deformation intervention is triggered and the user is marked as an encroacher. During deformation intervention, the target shape of the deformation intervention is calculated based on the specific location and degree of encroachment (the degree of encroachment includes the area and depth), combined with the distribution of the airbags and servo motors inside the sofa. This includes the height and tilt angle of the area to be raised, as well as the coordinated change parameters of the surrounding area, and generates control instructions. Control commands are distributed to the execution unit inside the sofa, which synchronously drives multiple airbags and servo motors to achieve local deformation. The sofa surface in the occupied area is raised to apply resistance to the invader, and the sofa surface in the area where the invader was originally located is deformed. Through the contraction and expansion of the airbags, a comfortable concave area is formed around the invader, inducing them to naturally move back to their original position. During the deformation intervention process, the system continuously tracks the user's posture and position changes, monitors the speed and direction of the user's center of gravity, and the changing trends of pressure in various parts of the body. If it detects that the user does not move as expected or develops a new uncomfortable posture, the deformation is adjusted in real time, including airbag pressure adjustment, motor drive angle correction, and surface hardness adjustment.

2. The method according to claim 1, characterized in that The pressure sensor array arranged on the sofa surface is used to obtain position distribution information and pressure distribution data of multiple users on the sofa in real time, including pressure distribution of various parts of the body and position coordinates of the users on the sofa, and the collected data is processed, including: The pressure value of each point on the sofa surface is collected through the pressure sensor array; The original pressure data is denoised and then smoothed to obtain smoothed pressure data; Perform bilinear interpolation on the smoothed pressure data to obtain the pressure distribution matrix; According to the pressure distribution matrix, the user's body contour information is extracted and combined with the pre-trained support vector regression model to calculate the user's position coordinates on the sofa and the pressure values ​​of various parts of the body; Obtaining the time series changes of the pressure distribution matrix, classifying the pressure change patterns using a clustering algorithm, identifying the number of users currently on the sofa, and assigning a unique identifier to each user; The time series database InfluxDB is used to store pressure data, user location coordinates, pressure values ​​of various body parts, number of users, and unique identifier data. The data is indexed by timestamp and user identifier.

3. The method according to claim 1, characterized in that Based on the pressure distribution data and combined with the position distribution information, the area occupied by each user is identified, including the outline boundary and center of gravity of each user. By analyzing the shape and intensity of the pressure distribution, the current posture of multiple users, including whether they cross their legs and whether they lean their bodies, is obtained. Performing image segmentation on the pressure distribution data, and identifying seed points based on local pressure peak points; Execute a watershed algorithm on the seed points to obtain a preliminary area range for each user; Extracting the contour boundary of each user according to the preliminary area range; The center of gravity position is obtained by calculating the average coordinates of all points within the contour boundary; Determining whether the area where the center of gravity is located is a valid area; Calculating shape characteristics of the pressure distribution for the effective area; comparing the shape features; Determine the user's current posture based on the comparison results; Analyzing the user's current posture and calculating the offset distance between the pressure center of gravity and the geometric center of gravity in the user area; If the offset distance exceeds a preset threshold, it is determined to be a tilted body posture; The Kalman filter method is used to track the changes of center of gravity position and posture parameters; establishing a posture transition probability matrix based on the changes; The user's next posture state is predicted using the posture transition probability matrix.

4. The method according to claim 1, wherein The continuously monitoring the area range and current posture of each user, and when detecting that a user's area range and posture have changed, determining that the user's posture and position have changed, includes: Acquire pressure sensor array data, calculate the pressure difference between each sensor point and the previous frame based on the data, and obtain the average change rate of the pressure difference; If the average rate of change exceeds a preset threshold, edge detection processing is triggered; Using the Canny edge detection algorithm to process the pressure distribution map and extract the user area boundary contour in the pressure distribution map; Calculating the area and perimeter of the region according to the boundary contour, and extracting Hu moments as shape descriptors; If the rate of change of the area and perimeter or shape descriptor of the region exceeds the corresponding preset threshold, it is determined that the user area range has changed; Apply the principal component analysis algorithm to the user area to extract the main eigenvectors of the pressure distribution; calculating a local binary pattern feature of the pressure distribution; combining the main feature vector with the local binary pattern feature to form a comprehensive posture feature descriptor; Calculating the similarity between the comprehensive posture feature descriptor sequence and the preset standard posture sequence; If the similarity is lower than a preset threshold, it is determined that the user's posture has changed; Tracking the change trend of the comprehensive posture feature descriptor using a Kalman filter; If the deviation between the eigenvalue predicted by the Kalman filter and the actual observed value exceeds a preset threshold, confirming whether the posture has changed; By monitoring the distance changes between adjacent user areas, user merging or separation events can be identified.

5. The method according to claim 1, wherein If there is a change, the posture and position relationship of multiple users is analyzed, and a spatial collision detection algorithm is used to determine whether there is any encroachment between users. If encroachment is detected, deformation intervention of the sofa is triggered, and the user is marked as an encroacher, including: Acquiring pressure sensor array data, extracting a posture profile based on the pressure sensor array data, and calculating the center of mass coordinates and area based on the posture profile; Determine a minimum enclosing rectangle using a rotating caliper method, and generate a bounding box based on the minimum enclosing rectangle, wherein the bounding box includes vertex coordinates and side lengths; Dividing the sofa surface into spatial units, wherein the size of the spatial units does not exceed a preset threshold; Mark the occupied spatial units according to the positions of the bounding boxes, and obtain the spatial overlapping area by comparing the occupied units of adjacent users; Determining whether the area ratio of the spatial overlapping region exceeds a preset encroachment threshold, and if the area ratio exceeds the preset encroachment threshold, determining that an encroachment behavior has occurred; If it is determined that an encroachment has occurred, activating an airbag inflation mechanism corresponding to the spatial overlap area; It also includes: establishing a bounding box for each user based on the posture and position relationship of multiple users, identifying suspected encroachment behavior, and judging whether encroachment behavior occurs through the encroachment range threshold and encroachment time threshold. If it is determined to be an encroachment behavior, the encroachment information is output, including the identity of the users involved, the location and time of the encroachment area.

6. The method according to claim 5, characterized in that Based on the posture and position relationship of multiple users, a bounding box is established for each user to identify suspected encroachment behavior, and an encroachment range threshold and an encroachment time threshold are used to determine whether encroachment behavior occurs. If encroachment behavior is determined, encroachment information is output, including the identity of the user involved, the location of the encroachment area, and the time, including: Based on the pressure sensor array data, the main features of the pressure distribution are extracted. The rotating calculus method is used to calculate the minimum bounding rectangle (MBR) to obtain the vertex coordinates, length, width, and pressure center of the user's bounding box. Generalized collision detection is performed on the user's bounding box. If the calculated distance between the closest points is less than a preset threshold, a collision is determined. For the colliding user pairs, the intersection area is calculated. If the ratio of the intersection area to the area of ​​each bounding box exceeds a preset threshold, it is determined to be a suspected encroachment. For the suspected encroachment behavior, a sliding time window monitoring mechanism is activated. If the duration of continuous encroachment within the sliding time window exceeds a preset threshold, it is determined to be an encroachment behavior; the k-nearest neighbor algorithm is used to perform identity matching on the user's pressure distribution characteristics to obtain the unique identifiers of the encroacher and the encroached, and the encroachment information including the unique identifier, the coordinate range of the encroachment area and the start time of the encroachment is output.

7. The method according to claim 1, characterized in that In the deformation intervention, the target shape of the deformation intervention is calculated based on the specific location and degree of the encroachment in the encroachment behavior, where the encroachment degree includes the encroachment area and depth, combined with the distribution of the airbags and servo motors inside the sofa, including the height and inclination angle of the area to be raised, and the coordinated change parameters of the surrounding area, and the control instructions are generated, including: Receiving encroachment behavior data carrying encroachment location coordinates and encroachment degree parameters; According to the occupied position coordinates and the occupied degree parameters, a k-nearest neighbor algorithm combined with bilinear interpolation is used to obtain a list of affected execution units, wherein the list of execution units includes the grid coordinates of the airbags and servo motors inside the sofa; Based on the encroachment degree parameter, a preset fuzzy rule set is used to calculate an intensity coefficient of the deformation intervention, wherein the intensity coefficient is used to adjust subsequent deformation parameters; Using the intensity coefficient and the coordinates of the occupied position, a target surface for deformation intervention is generated through a cubic spline interpolation algorithm. The target surface is centered on the occupied position, and surrounding control points are selected and natural boundary conditions are set. Determine the height and inclination angle of the area that needs to be raised based on the strength coefficient of deformation intervention. h=k·sin(θ) Where h represents the height of the region, k represents the intensity coefficient, and θ represents the tilt angle. Where θ represents the tilt angle, h represents the height, and l represents the horizontal distance; Discretize the target surface into a grid point set corresponding to the airbag and servo motor positions in the execution unit list, and calculate the target state of each execution unit, wherein the target state includes the inflation amount of the airbag and the rotation angle of the servo motor; A specific control instruction sequence is generated, where the control instruction sequence includes a unique identifier of the execution unit, a target state value, and an execution timestamp.

8. The method according to claim 1, characterized in that The control instructions are distributed to the execution unit inside the sofa, which synchronously drives multiple airbags and servo motors to achieve local deformation, thereby raising the sofa surface in the occupied area to apply resistance to the invader, deforming the sofa surface in the area where the invader was originally located, and forming a comfortable concave area around the invader through the contraction and expansion of the airbags to induce the invader to naturally move back to his original position, including: Obtaining a control instruction sent by the central control unit via the CAN bus, wherein the control instruction includes a target airbag inflation volume and a servo motor rotation angle; According to the control instruction, the execution unit uses a precision clock synchronization protocol to complete the instruction execution within a preset synchronization time window of milliseconds; If the execution unit in the occupied area receives the control instruction, it controls the airbag inflation pump to increase the inflation volume to the preset air pressure, and at the same time drives the servo motor to adjust the angle of the support structure to the preset angle; If the execution unit in the original position area receives a deformation instruction, the airbag is controlled to perform periodic contraction and expansion, wherein the contraction time is a preset value and the expansion time is a preset value; Determine whether there are multiple encroachers. If so, allocate resources using a weighted method based on the degree of encroachment, where the higher the degree of encroachment, the more execution units are allocated.

9. The method according to claim 1, characterized in that During the deformation intervention process, the user's posture and position changes are continuously tracked, the speed and direction of the user's center of gravity are monitored, and the pressure change trend of each part is monitored. If it is detected that the user does not move as expected or a new uncomfortable posture occurs, the deformation is adjusted in real time, including airbag pressure adjustment, motor drive angle correction and surface hardness adjustment, including: A pressure sensor array is used to collect the user's body pressure distribution data in real time, and the user's center of gravity is calculated using a center of mass calculation algorithm; Based on the user's body pressure distribution data, the support vector machine algorithm is used to classify the user's posture and obtain the user's current posture type; Set posture change thresholds and center of gravity movement thresholds. If the user's posture change amplitude exceeds the preset threshold or the center of gravity movement speed is lower than the expected value, the deformation parameter adjustment is triggered; Using a decision tree to store deformation strategies, wherein nodes of the decision tree include posture type, center of gravity position, and pressure distribution characteristics, and traversing the decision tree to obtain deformation strategies based on the user's current posture type, center of gravity position, and pressure distribution characteristics; generating a control instruction sequence according to the deformation strategy, wherein the control instruction sequence includes an airbag pressure adjustment amount, a motor drive angle correction value, and a surface hardness adjustment parameter; The control instruction sequence is sent to the relevant execution unit, and the execution unit performs adjustments after receiving the control instruction sequence and returns the execution result.

Images