Method and device for adjusting headrest of electric wheelchair

Abstract

The invention provides an electric wheelchair headrest adjusting method and device, and relates to the technical field of wheelchair headrest adjusting.According to the technical scheme, when it is detected that a user triggers a headrest adjusting instruction through one-button operation, current positioning data of an electric wheelchair, head pressure data of the user, headrest angle data and use time data are obtained; identifying a current use scene according to the acquired data; determining a target position of the headrest according to the identified current use scene and pre-stored personalized parameters; the position difference between the current position and the target position of the headrest is calculated, and a headrest adjusting track is generated based on the position difference; and controlling the headrest to execute an adjusting action according to the headrest adjusting track, simultaneously monitoring force feedback and acceleration in the adjusting process, and automatically stopping or reducing the adjusting speed when abnormity is detected. The method and the device for adjusting the headrest of the electric wheelchair have the advantages of simplicity and convenience in operation and intelligent scene recognition.

Description

Technical Field

[0001] The present application relates to the technical field of wheelchair headrest adjustment, and in particular to a method and device for adjusting a headrest of an electric wheelchair. Background Art

[0002] In the existing electric wheelchair headrest adjustment technology, manual or simple electric adjustment methods are usually adopted. However, these methods are inconvenient for elderly users with limited hand mobility or mild cognitive impairment. For example, an elderly user who has been using an electric wheelchair for a long time needs to frequently adjust the position of the headrest in daily life. The user's family has set up multiple preset scene modes for him, including eating, watching TV, and resting. However, due to the limited hand mobility of the user, the complex control panel cannot be accurately operated. In addition, the user also suffers from mild cognitive impairment and has difficulty remembering the specific settings corresponding to different scenes.

[0003] In this case, users need to quickly switch the position of the headrest between different scenarios, while also considering the need for emergency adjustments in unexpected situations. More specifically, when users switch from eating mode to watching TV mode, the headrest needs to be adjusted to the appropriate angle and height quickly and smoothly while ensuring comfort. In addition, considering the safety of the user, the head needs to be prevented from being subjected to sudden shaking or pressure during the adjustment process.

[0004] Existing technologies are difficult to meet these special needs. The main problems are: complicated operation, not suitable for users with limited hand mobility; lack of intelligent scene recognition function, unable to automatically adapt to different usage scenarios; lack of stability and safety guarantee in the adjustment process; unable to meet personalized needs and emergency adjustment requirements. These problems seriously affect the comfort and safety of electric wheelchair users, especially for those with special needs, who are in urgent need of a more intelligent, convenient and safe headrest adjustment method.

[0005] In view of the above problems, the existing technology needs to be improved urgently. Summary of the invention

[0006] The purpose of the present application is to provide a method and device for adjusting the headrest of an electric wheelchair, which has the advantages of easy operation, intelligent scene recognition, smooth and safe adjustment, and meeting personalized needs.

[0007] The present application provides a method for adjusting the headrest of an electric wheelchair, and the technical solution is as follows: The method includes: when it is detected that a user triggers a headrest adjustment instruction through a one-touch operation, obtaining the current positioning data of the electric wheelchair, the user's head pressure data, the headrest angle data and the usage time data; identifying the current usage scenario according to the positioning data, the user's head pressure data, the headrest angle data and the usage time data; determining the target position of the headrest according to the identified current usage scenario and pre-stored personalized parameters; calculating the position difference between the current position and the target position of the headrest, and generating a headrest adjustment trajectory based on the position difference; controlling the headrest to perform an adjustment action according to the headrest adjustment trajectory, while monitoring the force feedback and acceleration during the adjustment process, and automatically stopping or reducing the adjustment speed when an abnormality is detected.

[0008] Furthermore, the present application also proposes that the step of identifying the current usage scenario based on the positioning data, the user's head pressure data, the headrest angle data and the usage time data includes: constructing a multidimensional feature space, and mapping the acquired positioning data, head pressure data, headrest angle data and usage time data in the multidimensional feature space as a feature vector, wherein each dimension corresponds to a data type; analyzing the feature vector to obtain a continuous scene description value characterizing the user's current state; and dynamically determining the current usage scenario based on the continuous scene description value, combined with the stored user historical usage data and the real-time collected physiological state data.

[0009] Furthermore, the present application also proposes that the step of analyzing the feature vector to obtain a continuous scene description value characterizing the current state of the user includes: when a change in the user state is detected, analyzing the feature vector to obtain an initial continuous scene description value; based on the changing trend of the historical continuous scene description value, performing time series correction on the initial continuous scene description value; dynamically adjusting the weight of each feature according to the importance of different features in the current time period, wherein the feature includes at least one of the head pressure distribution, the headrest angle and the usage time; and optimizing the corrected continuous scene description value in combination with the adjusted feature weight to obtain the final continuous scene description value.

[0010] Furthermore, the present application also proposes that the step of performing time series correction on the initial continuous scene description value based on the changing trend of the historical continuous scene description value includes: obtaining a sequence of historical continuous scene description values ​​within a specified time window based on the changing trend of the historical continuous scene description value, and establishing a state transition probability matrix, wherein the state transition probability matrix reflects the user's transition rules between various preset scenes; calculating the confidence of the initial continuous scene description value according to the state transition probability matrix, and when the confidence is lower than a preset threshold, correcting the initial continuous scene description value based on the state transition probability matrix to obtain a corrected continuous scene description value.

[0011] Furthermore, the present application also proposes that the step of optimizing the corrected continuous scene description value in combination with the adjusted feature weight to obtain the final continuous scene description value includes: after obtaining the user's physiological state data, calculating the physiological state matching degree of each preset scene based on the preset scene-physiological state correspondence relationship; using the physiological state matching degree as an additional feature, and fusing it with the adjusted feature weight to obtain a comprehensive feature weight; using the comprehensive feature weight to perform weighted averaging on the corrected continuous scene description value to obtain the final continuous scene description value.

[0012] Furthermore, the present application also proposes that the step of correcting the initial continuous scene description value based on the state transition probability matrix includes: after obtaining the user's historical scene conversion data and current physiological state data, calculating the user's scene conversion frequency in different time periods based on the historical scene conversion data, and dynamically adjusting the preset threshold according to the scene conversion frequency and the current physiological state data; when the confidence of the initial continuous scene description value is lower than the dynamically adjusted preset threshold, correcting the initial continuous scene description value based on the state transition probability matrix to obtain a corrected continuous scene description value.

[0013] Furthermore, the present application also proposes that the step of calculating the position difference between the current position and the target position of the headrest and generating a headrest adjustment trajectory based on the position difference includes: after obtaining the current position coordinates and the target position coordinates of the headrest, calculating the position difference vector based on the current position coordinates and the target position coordinates, and dynamically adjusting the maximum speed and maximum acceleration parameters in combination with the user's physiological state data; generating an S-shaped speed curve that takes into account the user's neck bearing capacity according to the size of the position difference vector and the adjusted maximum speed and maximum acceleration parameters, wherein the S-shaped speed curve includes an initial slow acceleration section, an intermediate uniform speed section and a final slow deceleration section; based on the S-shaped speed curve and the position difference vector, calculating the intermediate position coordinate sequence of the headrest during the adjustment process to form a smooth headrest adjustment trajectory that adapts to the user's physiological characteristics.

[0014] Furthermore, the present application also proposes that the step of generating an S-shaped velocity curve that takes into account the user's neck tolerance according to the size of the position difference vector and the adjusted maximum speed and maximum acceleration parameters includes: after obtaining the user's neck physiological parameters, calculating the maximum safe acceleration of the user's neck based on the neck physiological parameters, and comparing it with the preset maximum acceleration parameters, and selecting the smaller value as the actual maximum acceleration parameter; according to the actual maximum acceleration parameter and the size of the position difference vector, calculating the time and distance of the acceleration segment and the deceleration segment, and generating an S-shaped velocity curve including an initial slow acceleration segment, an intermediate uniform speed segment and a final slow deceleration segment.

[0015] Furthermore, the present application also proposes that the step of optimizing the corrected continuous scene description value in combination with the adjusted feature weight to obtain the final continuous scene description value includes: obtaining the corrected continuous scene description value X=(x1, x2, ..., xn) and the adjusted feature weight W=(w1, w2, ..., wn), where xi represents the description value of the i-th preset scene, wi represents the weight of the i-th feature, and n is the number of preset scenes, and collecting the user's physiological state data P=(p1, p2, ..., pm) in real time, where pj represents the j-th physiological state parameter, and m is the number of physiological state parameters; based on the pre-established scene-physiological state correspondence matrix S, calculating the physiological state matching degree M=S·P, where S is an n×m matrix, and fusing the feature weight and the physiological state matching degree to obtain the comprehensive feature weight W'=α·W+(1-α)·normalize(M), where α is a balance parameter, 0≤α≤1, and normalize is a normalization function; calculating the final continuous scene description value Y=(y1,y2, ..., yn), where Y = softmax(β·X+(1-β)·W'), β is a balance parameter, 0≤β≤1.

[0016] Furthermore, the present application also proposes a headrest adjustment device for an electric wheelchair, the device comprising: an acquisition module, for acquiring the current positioning data of the electric wheelchair, the user's head pressure data, the headrest angle data and the usage time data when it is detected that the user triggers the headrest adjustment instruction through a one-button operation; an identification module, for identifying the current usage scenario based on the positioning data, the user's head pressure data, the headrest angle data and the usage time data; a determination module, for determining the target position of the headrest based on the identified current usage scenario and pre-stored personalized parameters; a generation module, for calculating the position difference between the current position and the target position of the headrest, and generating a headrest adjustment trajectory based on the position difference; a control module, for controlling the headrest to perform an adjustment action according to the headrest adjustment trajectory, while monitoring the force feedback and acceleration during the adjustment process, and automatically stopping or reducing the adjustment speed when an abnormality is detected.

[0017] From the above, it can be seen that the electric wheelchair headrest adjustment method and device provided in the present application simplifies the operation process through one-button operation and intelligent scene recognition, and is suitable for users with limited hand movement ability; through multi-dimensional data analysis and dynamic adjustment, intelligent scene recognition and personalized adjustment are realized; by generating a smooth adjustment trajectory and real-time monitoring, the smoothness and safety of the adjustment process are guaranteed, and it has the advantages of simple operation, intelligent scene recognition, smooth and safe adjustment, and meeting personalized needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1A flow chart of a method for adjusting the headrest of an electric wheelchair provided in this application.

[0019] Figure 2 This is a schematic structural diagram of an electric wheelchair headrest adjustment device provided in this application.

[0020] In the figure: 210, acquisition module; 220, identification module; 230, determination module; 240, generation module; 250, control module. DETAILED DESCRIPTION

[0021] The technical solutions in the present application will be clearly and completely described below in conjunction with the drawings in the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the present application for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0022] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0023] In practical applications, the electric wheelchair headrest adjustment system faces technical problems such as low intelligence, insufficient safety and comfort. Specifically, existing systems usually use manual or simple electric adjustment methods, which is inconvenient for elderly users with limited hand mobility or mild cognitive impairment. The system lacks intelligent scene recognition function and cannot automatically adapt to different usage scenarios, resulting in users frequently manually adjusting the headrest position. In addition, the adjustment process lacks smoothness and safety guarantees, which may cause discomfort or potential safety hazards to users. These problems seriously affect the system's ease of use, intelligence level, and ability to ensure user comfort and safety.

[0024] In this regard, refer to Figure 1 , the present application proposes a method for adjusting the headrest of an electric wheelchair, the method comprising: S110, when it is detected that the user triggers a headrest adjustment instruction through a one-touch operation, obtaining current positioning data of the electric wheelchair, the user's head pressure data, the headrest angle data, and the usage time data; S120, identifying the current usage scenario according to the positioning data, the user's head pressure data, the headrest angle data, and the usage time data; S130, determining a target position of the headrest according to the identified current usage scenario and pre-stored personalized parameters; S140, calculating a position difference between a current position of the headrest and a target position, and generating a headrest adjustment trajectory based on the position difference; S150, controlling the headrest to perform adjustment actions according to the headrest adjustment trajectory, while monitoring force feedback and acceleration during the adjustment process, and automatically stopping or reducing the adjustment speed when an abnormality is detected.

[0025] Among them, one-button operation means that the user can trigger the headrest adjustment command through a simple single operation. It can be implemented by using technologies such as touch buttons, voice commands or gesture recognition. This simplified operation method reduces the difficulty of operation for the user and is particularly suitable for users with limited hand movement ability.

[0026] Among them, positioning data refers to the information about the current location of the electric wheelchair, which can be obtained by GPS, indoor positioning system or other positioning technologies. These data help the system determine the user's environment and thus more accurately identify the current usage scenario.

[0027] Among them, head pressure data refers to the pressure distribution information generated when the user's head contacts the headrest, which can be collected using a pressure sensor array. These data reflect the user's head posture and support requirements, and are crucial for judging the user's comfort and adjustment needs.

[0028] Among them, the headrest angle data refers to the current tilt angle of the headrest, which can be measured by an angle sensor or an encoder. These data directly reflect the current state of the headrest and are an important reference for adjusting the position of the headrest.

[0029] Among them, usage time data refers to the continuous usage time of the user in the current state, which can be recorded by a timer or timestamp. These data help the system determine whether the user needs to adjust the headrest to avoid discomfort caused by maintaining the same posture for a long time.

[0030] Among them, scene recognition refers to automatically judging the user's current usage based on the acquired multi-dimensional data. It can be implemented by using preset algorithms or machine learning algorithms such as decision trees or neural networks. This intelligent recognition technology enables the system to automatically adapt to different usage scenarios without the user having to manually select a mode.

[0031] Among them, personalized parameters refer to the headrest adjustment preferences pre-set for specific users, which may include information such as the ideal headrest angle and height in different scenarios. These parameters enable the system to make more precise adjustments based on the user's personal needs.

[0032] The headrest adjustment trajectory refers to the path of the headrest moving from the current position to the target position, which can be generated by curve interpolation or trajectory planning algorithm. This smooth adjustment trajectory ensures the comfort and safety of the adjustment process.

[0033] Among them, force feedback refers to the resistance or pressure changes detected during the headrest adjustment process, which can be measured by force sensors. These data help the system to detect abnormal situations during the adjustment process in a timely manner.

[0034] Among them, acceleration refers to the rate of change of the headrest's movement speed during the adjustment process, which can be measured by an acceleration sensor. These data are used to monitor the smoothness of the adjustment process and prevent sudden shaking from causing discomfort to the user.

[0035] The core innovation of this application is to realize automatic judgment of the user's usage status through data collection and intelligent scene recognition technology. Combined with the pre-stored personalized parameters, the system can automatically determine the target position of the headrest that best suits the user. During the adjustment process, the safety and comfort of the adjustment are ensured by generating a smooth adjustment trajectory and monitoring force feedback and acceleration in real time.

[0036] The working principle of this application can be divided into the following key steps: First, when the user triggers the headrest adjustment command through a one-button operation, the system will start the data collection process, including obtaining the current positioning data of the electric wheelchair, the user's head pressure data, the headrest angle data and the usage time data.

[0037] Next, these multi-dimensional data are input into a pre-trained scene recognition algorithm. The algorithm can be based on machine learning techniques, such as decision trees or neural networks, and can automatically determine the current usage scenario based on the input data features, such as eating, watching TV, or resting.

[0038] After identifying the current scene, the system will determine the target position of the headrest in combination with the pre-stored personalized parameters. These personalized parameters may include information such as the user's preferred headrest angle and height in different scenes, which can be obtained through the user's usage history or manual settings, that is, different scene templates can be preset.

[0039] The system then calculates the position difference between the current and target positions of the headrest. Based on this position difference, the system generates a smooth adjustment trajectory in order to ensure smooth movement of the headrest.

[0040] When performing the adjustment action, the system controls the movement of the headrest according to the generated trajectory. At the same time, the force sensor and acceleration sensor continuously monitor the force feedback and acceleration changes during the adjustment process. If an abnormal situation is detected, such as a sudden increase in force feedback or acceleration exceeding the preset threshold, the system will respond immediately and automatically stop or reduce the adjustment speed to ensure the safety and comfort of the user.

[0041] Specifically, in some specific implementations, the relationship between the positioning data, the user's head pressure data, the headrest angle data, the usage time data and the current usage scenario is as follows: Scenario Positioning data Head pressure data Headrest angle data Using time data Possible headrest adjustments Eating Kitchen / Dining Room The headrest has low pressure and the head is tilted forward Angle 0-15° Short time Adjust the headrest to a low angle to avoid interfering with eating watch TV living room Even headrest pressure Angle 30-45° Longer time The headrest is adjusted to a comfortable angle and provides slow fine adjustment Lunch break Bedroom or living room The headrest has high pressure and is evenly distributed Angle 60-90° Time gradually increases The headrest angle automatically adjusts to a relaxing angle to prevent muscle tension read Living room / study room The headrest pressure is concentrated in the lower part Angle 15-30° Moderate The headrest is slightly tilted forward to provide good support Sleep bedroom Headrest pressure is stable and high Angle 75-90° No change for a long time The headrest enters a long-term static mode to avoid frequent adjustments In this way, the present application realizes an intelligent, safe and comfortable electric wheelchair headrest adjustment method, which effectively solves the problems of complex operation, lack of intelligent scene recognition, and lack of safety assurance in the adjustment process in traditional methods.

[0042] In some of the above embodiments of the present application, a step of identifying the current usage scenario based on positioning data, the user's head pressure data, the headrest angle data and the usage time data is proposed to determine the current usage scenario. However, in this process, how to achieve more accurate, continuous and personalized scene recognition for the user during the adjustment of the electric wheelchair headrest to adapt to the user's transition state between different scenes and improve the accuracy and comfort of the headrest adjustment becomes a problem.

[0043] In this regard, the present application further proposes the steps of identifying the current usage scenario based on the positioning data, the user's head pressure data, headrest angle data and usage time data, including: constructing a multidimensional feature space, and mapping the acquired positioning data, head pressure data, headrest angle data and usage time data in the multidimensional feature space as a feature vector, where each dimension corresponds to a data type; analyzing the feature vector to obtain a continuous scene description value that characterizes the user's current state; based on the continuous scene description value, combined with the stored user historical usage data and the real-time collected physiological state data, dynamically determine the current usage scenario.

[0044] This application realizes the unified processing and analysis of multidimensional data by constructing a multidimensional feature space and mapping various data types into feature vectors. This method improves data processing efficiency and provides a more comprehensive information basis for subsequent scene recognition. The analysis of the feature vector obtains a continuous scene description value. Compared with discrete scene classification, this continuous value representation method can better reflect the subtle changes in user status and improve the accuracy and sensitivity of scene recognition. Finally, by combining the user's historical usage data and real-time physiological status data, the current usage scenario is dynamically determined. This method takes into account the user's personalized characteristics and real-time status, making the scene recognition results more accurate and adaptable.

[0045] The technical solution of this application effectively solves the problems of low data processing efficiency and low scene recognition accuracy through comprehensive analysis of multi-dimensional data and dynamic scene recognition. By introducing continuous scene description values ​​and considering user historical data and real-time physiological status, this application can more accurately capture the dynamic changes of user status and provide a more accurate and adaptable scene recognition basis for subsequent headrest adjustment. This method not only improves the accuracy of scene recognition, but also enhances the system's ability to adapt to user personalized needs, thereby achieving smarter and more accurate headrest adjustment.

[0046] In this application, constructing a multidimensional feature space is a key step. This feature space can be implemented in a variety of ways. For example, you can use principal component analysis (PCA) or t-SNE and other dimensionality reduction techniques to construct a low-dimensional feature space, or you can directly use the original data dimensions to construct a high-dimensional feature space. Each dimension corresponds to a data type. For example, positioning data may include longitude and latitude coordinates, head pressure data may include a pressure distribution matrix, headrest angle data may include pitch angle and roll angle, and usage time data may include the current timestamp and continuous usage time.

[0047] When mapping the acquired data to the feature space, different normalization or standardization methods can be used to ensure that different types of data are comparable in the feature space. For example, Z-score normalization or Min-Max scaling can be used to handle data of different dimensions.

[0048] A variety of machine learning or deep learning algorithms can be used to analyze feature vectors. For example, classification algorithms such as support vector machines (SVM) or random forests can be used to obtain discrete scene classification results, which are then converted into continuous scene description values ​​through a softmax function. Alternatively, neural network models such as multi-layer perceptrons (MLPs) or long short-term memory networks (LSTMs) can be used directly to learn the mapping relationship between feature vectors and continuous scene description values.

[0049] The introduction of continuous scene description values ​​enables the application to capture subtle changes in user status. For example, if a user gradually changes from a resting state to a TV watching state, the continuous scene description value can smoothly reflect this process instead of suddenly jumping from one discrete state to another. This continuity makes the headrest adjustment smoother and more natural, improving the user's comfort.

[0050] The current usage scenario is dynamically determined by combining the user's historical usage data and real-time physiological status data, further improving the accuracy and personalization of scene recognition. Historical usage data can reflect the user's habits and preferences, while real-time physiological status data (such as heart rate, breathing rate, etc.) can reflect the user's immediate needs. The combination of these two types of data makes scene recognition more comprehensive and accurate.

[0051] In the process of solving the problem of accurate, continuous and personalized scene recognition, the technical solution of this application realizes the unified processing of multi-source heterogeneous data through the construction of multi-dimensional feature space, thereby improving the efficiency of data utilization. The introduction of continuous scene description values ​​enables the system to capture subtle changes in user status and adapt to the transition state between different scenes. The dynamic scene determination mechanism that combines historical data and real-time physiological status data further improves the accuracy and personalization of scene recognition.

[0052] Compared with the traditional discrete scene classification method, this method can better adapt to the dynamic changes of user status. For example, when the user gradually changes from a resting state to a TV watching state, the headrest can be smoothly adjusted according to the continuously changing scene description value instead of suddenly jumping from one position to another. This not only improves the user's comfort, but also reduces the discomfort or safety risks that may be caused by sudden adjustments.

[0053] In addition, the method of the present application achieves a higher degree of personalization by considering the user's historical usage data and real-time physiological state. For example, the system can learn that a certain user is more inclined to use a certain scene mode in a certain time period, or requires a specific headrest adjustment in a certain physiological state. This personalized scene recognition and adjustment greatly improves the system's adaptability and user experience.

[0054] As a preferred implementation, the present application can be implemented through the following steps: First, a 5-dimensional feature space is constructed, corresponding to the positioning data (longitude and latitude), head pressure data (pressure center coordinates), headrest angle data (pitch angle), and usage time data (current timestamp).

[0055] The acquired data is then preprocessed and normalized. For example, longitude and latitude are converted to offsets relative to the user's usual location, head pressure data is converted to offsets relative to the center of the headrest, headrest angle data is converted to angles relative to the horizontal position, and usage time data is converted to percentage of time in a day.

[0056] Next, the standardized feature vector is analyzed using a pre-trained deep neural network model to output a continuous scene description value, which can be a number between 0 and 1, where 0 represents a completely resting state, 1 represents a fully active state, and the intermediate value represents a transition state.

[0057] Finally, the continuous scene description value is fine-tuned by combining the user's historical usage data (such as usage patterns at similar times and locations in the past week) and real-time physiological state data (such as heart rate and breathing rate). For example, if historical data shows that the user is usually resting at this time, but real-time heart rate data shows that the user may be active, the system will appropriately increase the scene description value.

[0058] In this way, the present application achieves accurate, continuous and personalized scene recognition, providing a reliable basis for subsequent headrest adjustment, thereby improving the accuracy and comfort of electric wheelchair headrest adjustment.

[0059] In some of the above embodiments of the present application, it is proposed to analyze the feature vector to obtain a continuous scene description value that characterizes the user's current state to identify the current usage scenario. However, the following problems may exist in this process: first, the initial continuous scene description value may not be accurate enough to fully reflect the dynamic changes in the user's state; second, the importance of different features may change in different time periods and require dynamic adjustment; finally, relying solely on feature vector analysis may ignore the user's real-time physiological state information, resulting in scene recognition results that are not comprehensive and accurate enough.

[0060] In this regard, the present application further proposes to analyze the feature vector when a change in user status is detected to obtain an initial continuous scene description value; perform time correction on the initial continuous scene description value based on the change trend of the historical continuous scene description value; dynamically adjust the weight of each feature according to the importance of different features in the current time period, wherein the features include at least one of the head pressure distribution, the headrest angle and the usage time; and optimize the corrected continuous scene description value in combination with the adjusted feature weight to obtain the final continuous scene description value.

[0061] The technical solution of the present application effectively improves the accuracy and adaptability of scene recognition by introducing mechanisms such as timing correction, dynamic weight adjustment, and continuous scene description value optimization. Specifically, the timing correction mechanism takes into account the changing trend of historical continuous scene description values, avoids sudden changes in scene recognition results, and improves the stability of the system. The dynamic weight adjustment mechanism adjusts the weights according to the importance of each feature in different time periods, so that the system can better adapt to the changing needs of users at different times of the day. The use of continuous scene description values ​​can more finely capture the user's transition state between different scenes, achieving smoother and more comfortable headrest adjustment.

[0062] Furthermore, the technical solution of this application focuses on parameters directly related to the adjustment of the electric wheelchair headrest, such as head pressure distribution, headrest angle and usage time, ensuring a high correlation between the scene recognition results and the actual headrest adjustment requirements. This targeted parameter selection and optimization strategy enables the system to more accurately and timely capture the user's state changes during the use of the electric wheelchair, providing a more reliable basis for subsequent headrest adjustment.

[0063] In specific implementation, the acquisition of the initial continuous scene description value can be achieved in a variety of ways. For example, a pre-trained neural network model can be used to take the feature vector as input and output a multi-dimensional continuous scene description value. Another method is to use a clustering algorithm to map the feature vector to a predefined scene space and obtain the probability distribution of each preset scene as the initial continuous scene description value.

[0064] The time series correction step can use the sliding window method combined with the exponentially weighted moving average (EWMA) algorithm. Specifically, a fixed-size queue of historical continuous scene description values ​​can be maintained. When a new initial continuous scene description value is generated, it is combined with the historical value in the queue for EWMA calculation to obtain the corrected continuous scene description value. This method not only takes into account historical trends, but also can timely reflect changes in the current state.

[0065] Dynamic weight adjustment can be implemented based on time periods and user behavior patterns. For example, the feature importance matrix for different time periods can be pre-defined, and the weights of each feature can be dynamically adjusted in combination with the user's historical usage data. In addition, reinforcement learning algorithms can be introduced to optimize the feature weight allocation strategy by continuously learning user feedback.

[0066] The optimization process of the final continuous scene description value can adopt weighted average or more complex fusion algorithm. For example, a soft voting mechanism can be used to fuse the corrected continuous scene description value and the scene probability distribution calculated based on the feature weight to obtain the final continuous scene description value. This method retains the stability of timing correction and incorporates the flexibility of dynamic weight adjustment.

[0067] In a specific embodiment, it is assumed that the electric wheelchair is equipped with a pressure sensor array, an angle sensor and a timer, which are used to collect head pressure distribution, headrest angle and usage time data respectively. The system collects data every 5 seconds and constructs a feature vector.

[0068] First, the pre-trained convolutional neural network model is used to process the feature vector to obtain a 5-dimensional initial continuous scene description value, which corresponds to the five preset scenes of resting, eating, watching TV, reading and talking. Then, the system maintains a sliding window containing the data of the last 10 minutes and uses the EWMA algorithm (α=0.7) to perform time series correction on the initial value.

[0069] Next, the system adjusts the feature weights based on the current time period (for example, breakfast time period 8:00-9:00). During this time period, the weights of eating-related features (such as head forward tilt angle) will be increased. Assume that the adjusted weights are: head pressure distribution 0.4, headrest angle 0.3, and usage time 0.3.

[0070] Finally, the system combines the corrected continuous scene description value and the adjusted feature weights and uses the weighted average method to calculate the final continuous scene description value. For example, if the corrected value is [0.1, 0.6, 0.2, 0.05, 0.05], it may be adjusted to [0.08, 0.65, 0.18, 0.04, 0.05] after considering the feature weights, which more accurately reflects the state of the user who may be in a eating scene.

[0071] Through this method, the technical solution of the present application can more accurately and timely capture the user's state changes during the use of the electric wheelchair. For special user groups who find it difficult to accurately express their needs, this improvement significantly improves the accuracy and comfort of the headrest adjustment. For example, when the user switches from eating to watching TV, the system can smoothly adjust the position of the headrest to avoid sudden changes causing discomfort to the user. In addition, since time factors and historical trends are taken into account, the system can better predict and adapt to changes in user needs, reduce the number of unnecessary adjustments, and improve user experience.

[0072] In some of the above embodiments of the present application, a step of performing time-series correction on the initial continuous scene description value based on the changing trend of the historical continuous scene description value is proposed to improve the accuracy of scene recognition. However, in this process, there may be a situation where the initial continuous scene description value does not match the actual scene, resulting in an inaccurate correction result. In addition, relying solely on historical data for correction may not reflect the sudden change of the user status in a timely manner, affecting the real-time and adaptability of scene recognition.

[0073] In this regard, the present application further proposes to obtain a sequence of historical continuous scene description values ​​within a specified time window based on the changing trend of the historical continuous scene description values, and establish a state transition probability matrix, which reflects the user's transition rules between various preset scenes; calculate the confidence of the initial continuous scene description value according to the state transition probability matrix, and when the confidence is lower than a preset threshold, correct the initial continuous scene description value based on the state transition probability matrix to obtain a corrected continuous scene description value.

[0074] This application collects time-correlated historical scene description values ​​by setting a suitable time window to provide a data basis for subsequent analysis. This method is particularly suitable for elderly users with fixed work and rest schedules. By establishing a state transition probability matrix, the transition rules of elderly users between daily scenes such as eating, watching TV, and resting are modeled. This method can capture the regularity of elderly users' behavior patterns and improve the accuracy of scene recognition.

[0075] By calculating the confidence of the initial continuous scene description value and comparing it with the dynamically adjusted preset threshold, the possible erroneous scene judgments can be identified. The setting of the dynamic threshold takes into account the cognitive impairment of elderly users, so that the system can better adapt to the characteristics of different users. When the confidence of the recognition result is low, the state transition probability matrix is ​​used to make corrections so that the scene recognition result is more in line with the actual usage habits of elderly users. This mechanism can effectively prevent misjudgments caused by sudden changes in sensor data or environmental interference.

[0076] The technical solution of the present application can be implemented in the following manner: First, based on the change trend of the historical continuous scene description value, an appropriate time window is set, such as the last 24 hours or the last week. Within this time window, the system collects and stores the user's historical continuous scene description value sequence. These data can be stored in the database in the form of timestamps and corresponding scene description values.

[0077] Next, based on the collected historical data, a state transition probability matrix is ​​established. Assuming that the preset scenes include n scenes such as eating, watching TV, and resting, the state transition probability matrix will be an n×n matrix. Each element P(i,j) in the matrix represents the probability of transitioning from scene i to scene j. This probability can be calculated by counting the frequency of scene transitions in historical data. For example, if in the historical data, the number of times the user transitions from the eating scene to the watching TV scene accounts for 30% of all the times the user transitions out of the eating scene, then P(eating, watching TV)=0.3.

[0078] After obtaining the state transition probability matrix, the system can use this matrix to calculate the confidence of the initial continuous scene description value. One possible calculation method is to use the forward algorithm. Specifically, assuming that the initial continuous scene description value at the current moment is S_t, and the scene at the previous moment is S_(t-1), then the confidence of S_t can be expressed as: Confidence(S_t) = P(S_t | S_(t-1)) * Confidence(S_(t-1)); Where P(S_t | S_(t-1)) can be directly obtained from the state transition probability matrix.

[0079] The system also needs to set a preset threshold to determine whether the initial continuous scene description value is credible. This threshold can be dynamically adjusted according to the characteristics of the user. For example, for users with mild cognitive impairment, a higher threshold, such as 0.8, can be set; while for users with severe cognitive impairment, a lower threshold, such as 0.6, can be set.

[0080] When the calculated confidence is lower than the preset threshold, the system will start the correction mechanism. The correction process can use the maximum likelihood estimation method. Specifically, the system will calculate the most likely scene sequence under the current state transition probability matrix. This can be achieved through the Viterbi algorithm. The algorithm will consider the historical scene sequence and the currently observed data, find the most likely scene sequence, and replace the initial continuous scene description value with the last scene in this sequence.

[0081] In this way, the technical solution of the present application can effectively improve the accuracy and stability of scene recognition and reduce the impact of misjudgment on the use experience of elderly users. Especially for elderly users with limited hand mobility and mild cognitive impairment, this scene recognition method based on historical data and behavioral patterns can better meet their needs, thereby providing a more reliable basis for the precise adjustment of the headrest and improving the comfort and safety of electric wheelchair use.

[0082] As a preferred implementation, the technical solution of the present application can be implemented in an intelligent electric wheelchair system. The system includes a central processing unit, multiple sensors (such as pressure sensors, angle sensors, GPS positioning modules, etc.), a data storage unit and a headrest adjustment actuator.

[0083] In practical applications, the system can set a 24-hour sliding time window. Every 5 minutes, the system will record the user's scene description value and store this data in the data storage unit. Assume that 5 scenes are preset: eating, watching TV, resting, reading, and talking. Then the state transition probability matrix will be a 5×5 matrix.

[0084] The system will continuously update this matrix. For example, if in the last 24 hours, the user switched from the eating scene to the watching TV scene 20 times, and switched out of the eating scene 50 times in total, then P(eating, watching TV) = 20 / 50 = 0.4.

[0085] When the user triggers the headrest adjustment command, the system first obtains an initial continuous scene description value based on the current sensor data. Assume that this value indicates that the user is most likely watching TV (with a probability of 0.7). The system calculates the confidence of this initial continuous scene description value. If the user was eating at the previous moment, the confidence may be: Confidence = 0.7 * P(watching TV|eating) = 0.7 * 0.4 = 0.28; Assume that the system sets a preset threshold of 0.6 for this elderly user with mild cognitive impairment. Since 0.28<0.6, the system will start the correction mechanism. Through the Viterbi algorithm, the system may find that, considering historical data and current observations, the most likely state sequence of the user is: eating->resting->resting. Therefore, the system will correct the initial "watching TV" judgment to "resting".

[0086] This correction mechanism can effectively prevent misjudgments caused by sudden interference and improve the accuracy and stability of scene recognition. At the same time, because the system continuously updates the state transition probability matrix, it can also adapt to the slow changes in user behavior patterns and maintain long-term recognition accuracy. This can significantly improve the user experience and comfort for elderly users who need to use electric wheelchairs for a long time.

[0087] In some of the above embodiments of the present application, a step of optimizing the corrected continuous scene description value in combination with the adjusted feature weight to obtain the final continuous scene description value is proposed to optimize the continuous scene description value. However, in this process, only considering the feature weight may not fully reflect the user's real-time physiological state, resulting in insufficient accuracy in scene recognition. In addition, a single feature weight adjustment may not be able to adapt to complex and changing usage environments, affecting the accuracy and personalization of the headrest adjustment.

[0088] In this regard, the present application further proposes that after obtaining the user's physiological state data, the physiological state matching degree of each preset scene is calculated based on the preset scene-physiological state correspondence relationship; the physiological state matching degree is used as an additional feature and is integrated with the adjusted feature weight to obtain a comprehensive feature weight; the comprehensive feature weight is used to perform weighted averaging on the corrected continuous scene description value to obtain the final continuous scene description value.

[0089] The technical solution of this application realizes the integration of multi-dimensional information by introducing physiological state data and scene-physiological state correspondence, thereby improving the accuracy and personalization of scene recognition. Specifically, the solution first obtains the user's physiological state data, such as heart rate, respiratory rate, and body temperature. These data can reflect the user's real-time physiological state and provide additional personalized information for scene recognition.

[0090] Next, based on the pre-established scene-physiological state correspondence, the physiological state matching degree of each preset scene is calculated. This step takes into account the typical physiological characteristics of users in different scenes and enhances the pertinence of scene recognition. For example, eating scenes may be associated with specific heart rate and breathing patterns, while resting scenes may correspond to different physiological indicators.

[0091] The calculated physiological state matching degree is used as a new feature and fused with the previously adjusted feature weight to obtain a comprehensive feature weight. This fusion method integrates multi-dimensional information and makes scene recognition more comprehensive and accurate. The fusion process can use weighted averaging or other more complex algorithms to ensure that the physiological state information and other feature information are reasonably balanced.

[0092] Finally, the obtained comprehensive feature weights are used to perform weighted averaging on the corrected continuous scene description values ​​to obtain the final continuous scene description value. This step comprehensively considers multiple information sources and improves the accuracy and reliability of the scene description.

[0093] The technical solution of this application significantly improves the accuracy and personalization of scene recognition by introducing physiological state data and scene-physiological state correspondence. This method is particularly suitable for elderly users who have difficulty expressing their needs accurately. The system can assist in judging the current scene through changes in their physiological state. For example, for an elderly user with mild cognitive impairment, even if he cannot clearly express his needs, the system can more accurately judge whether the user is eating, watching TV or resting by monitoring changes in physiological indicators such as heart rate and respiratory rate, combined with current time and environmental information.

[0094] In specific implementation, machine learning algorithms can be used to establish and optimize the scene-physiological state correspondence. For example, algorithms such as support vector machines (SVM) or random forests can be used to train models based on a large amount of user data to establish mapping relationships between different scenes and physiological states. This model can be updated regularly to adapt to long-term changes in the user's physiological state.

[0095] When calculating the physiological state matching degree, methods such as cosine similarity or Euclidean distance can be used to compare the similarity between the current user's physiological state vector and the standard physiological state vector of the preset scene. For example, suppose the system presets three scenes: eating, watching TV, and resting, and each scene has a corresponding standard physiological state vector. After obtaining the user's real-time physiological state data, calculate the similarity between this data and the three standard vectors to obtain the matching degree of each scene.

[0096] When fusing feature weights and physiological state matching, an adaptive weight algorithm can be used. For example, the weight of feature weights and physiological state matching in the final decision can be dynamically adjusted based on the user's historical data. If it is found that the physiological state change of a user is highly correlated with the scene change, the system will increase the weight of the physiological state matching; conversely, if it is found that the physiological state change is not obvious, the system will reduce its weight accordingly.

[0097] In this way, the technical solution of the present application can more comprehensively capture the state information of elderly users during the use of electric wheelchairs. This not only improves the accuracy of scene recognition, but also enhances the system's ability to adapt to individual differences of users. Ultimately, this optimized scene description value can provide a more reliable basis for the precise adjustment of the electric wheelchair headrest, thereby improving the comfort and safety of elderly users using electric wheelchairs in daily scenes such as eating, watching TV and resting.

[0098] Through this implementation, the system can flexibly adjust the weight of physiological state information and other feature information in the final decision, so as to adapt to the individual differences of different users and the special needs of different scenarios. For example, for a user with heart disease, the system may increase the weight of heart rate parameters to more sensitively capture possible abnormal conditions. For a user with sleep disorders, the system may pay more attention to breathing rate and body movement to more accurately identify the resting state.

[0099] This approach not only improves the accuracy of scene recognition, but also enhances the adaptability and personalization of the system. By taking into account multiple information sources, the system can better understand the user's actual needs and provide more precise headrest adjustment services. This is particularly important for elderly users with limited hand mobility and possible cognitive impairments, and can significantly improve their comfort and safety when using electric wheelchairs.

[0100] In some of the above embodiments of the present application, a step of correcting the initial continuous scene description value based on the state transition probability matrix is ​​proposed to improve the accuracy of scene recognition. However, in this process, the fixedness of the preset threshold may cause the system to be insufficiently adaptable to changes in user behavior patterns, especially when the user's physiological state changes, which may affect the sensitivity and accuracy of scene recognition.

[0101] In this regard, the present application further proposes that after obtaining the user's historical scene conversion data and current physiological state data, the user's scene conversion frequency in different time periods is calculated based on the historical scene conversion data, and the preset threshold is dynamically adjusted according to the scene conversion frequency and the current physiological state data; when the confidence of the initial continuous scene description value is lower than the dynamically adjusted preset threshold, the initial continuous scene description value is corrected based on the state transition probability matrix to obtain a corrected continuous scene description value.

[0102] The technical solution of the present application improves the adaptability and accuracy of scene recognition by introducing a dynamic threshold adjustment mechanism. Specifically, the system first obtains the user's historical scene conversion data and current physiological state data. The historical scene conversion data may include the user's scene switching records in the past period of time (such as a week or a month), and the current physiological state data may include indicators such as heart rate, blood pressure, and body temperature.

[0103] Based on historical scene transition data, the system calculates the scene transition frequency of users in different time periods. This step can be achieved through statistical analysis methods, such as using time series analysis or sliding window technology to calculate the number of scene transitions in each time period (such as every hour or every half day). The calculation of scene transition frequency takes into account the time correlation of user behavior and can reflect the activity patterns of users at different times of the day.

[0104] Next, the system dynamically adjusts the preset threshold based on the calculated scene transition frequency and current physiological state data. This process can be achieved by designing a weight function that takes the scene transition frequency and physiological state data as input and outputs an adjustment coefficient. For example, when the scene transition frequency is high, the system may lower the threshold to increase sensitivity to scene changes; when the user's physiological state indicators show fatigue or discomfort, the system may increase the threshold to reduce misjudgment.

[0105] After dynamically adjusting the threshold, the system compares the confidence of the initial continuous scene description value with the new threshold. When the confidence of the initial continuous scene description value is lower than the preset threshold after dynamic adjustment, the system will correct the initial continuous scene description value based on the state transition probability matrix. This correction process can be achieved through matrix operation, multiplying the initial continuous scene description value with the state transition probability matrix to obtain the corrected scene description value.

[0106] This combination of dynamic threshold adjustment and scene description value correction has multiple advantages. First, it improves the system's ability to adapt to changes in user behavior patterns. By considering the frequency of scene transitions in different time periods, the system is better able to capture regular changes in users' daily activities. Second, taking current physiological state data into account allows the system to adjust recognition sensitivity based on the user's real-time condition, which is particularly important for elderly users or users with changeable physical conditions. Finally, by combining dynamic threshold adjustment with probability matrix-based correction, the system is able to effectively reduce misjudgments caused by environmental interference or sensor fluctuations while maintaining sensitivity.

[0107] In some of the above embodiments of the present application, a step of calculating the position difference between the current position of the headrest and the target position, and generating a headrest adjustment trajectory based on the position difference is proposed to achieve the adjustment of the headrest. However, in this process, how to ensure the smoothness and safety of the adjustment process while taking into account the physiological characteristics and comfort of the user has become an urgent problem to be solved. Especially for elderly users with limited hand mobility and mild cognitive impairment, how to ensure the adjustment effect while avoiding sudden shaking or pressure on the head has become a key technical challenge.

[0108] In this regard, the present application further proposes that after obtaining the current position coordinates and target position coordinates of the headrest, a position difference vector is calculated based on the current position coordinates and the target position coordinates, and the maximum speed and maximum acceleration parameters are dynamically adjusted in combination with the user's physiological state data; based on the size of the position difference vector and the adjusted maximum speed and maximum acceleration parameters, an S-shaped speed curve that takes into account the user's neck bearing capacity is generated, and the S-shaped speed curve includes an initial slow acceleration segment, an intermediate uniform speed segment and a final slow deceleration segment; based on the S-shaped speed curve and the position difference vector, a sequence of intermediate position coordinates of the headrest during the adjustment process is calculated to form a smooth headrest adjustment trajectory that adapts to the user's physiological characteristics.

[0109] The technical solution of this application provides a safe and comfortable headrest adjustment experience for elderly users with limited hand mobility and mild cognitive impairment through dynamic parameter adjustment, segmented speed control, smooth trajectory generation and safety assurance mechanism. This method is particularly suitable for daily scenes such as eating, watching TV and resting. The system can generate the most suitable adjustment trajectory according to the user's real-time status, which not only ensures the accuracy of the adjustment, but also avoids unnecessary stimulation or harm to the user.

[0110] Specifically, the technical solution of this application includes the following key features: First, the dynamic parameter adjustment mechanism dynamically adjusts the maximum speed and maximum acceleration parameters by acquiring the user's physiological state data in real time. This can be achieved in a variety of ways, such as using wearable devices to collect the user's heart rate, blood pressure and other physiological data, or using the pressure sensor built into the headrest to detect the user's head posture and neck muscle tension. Based on this data, the system can use fuzzy logic or machine learning algorithms to determine the most suitable speed and acceleration parameters.

[0111] Secondly, the generation of the S-shaped speed curve takes into account the tolerance of the user's neck. In specific implementation, the entire adjustment process can be divided into three stages: the initial slow acceleration stage, the intermediate uniform speed stage, and the final slow deceleration stage. The time and distance of each stage can be dynamically calculated based on the size of the position difference vector and the user's physiological state. For example, for users with fragile necks, the system can extend the time of the acceleration and deceleration stages and shorten the time of the uniform speed stage to provide a smoother adjustment experience.

[0112] Finally, the calculation of the intermediate position coordinate sequence is based on the S-shaped velocity curve and the position difference vector. This can be achieved by numerical integration method, integrating the velocity curve in the time dimension to obtain the position of the headrest at each time point. In order to improve the computational efficiency, the Runge-Kutta method with adaptive step size can be used, using smaller step size at key points (such as acceleration change points) and larger step size in the uniform speed section.

[0113] Dynamic parameter adjustment directly affects the generation of the S-shaped speed curve, which in turn determines the calculation results of the intermediate position coordinate sequence. This design enables the entire adjustment process to accurately adapt to the user's real-time status and needs.

[0114] Through this design, the technical solution of the present application can effectively solve the problems existing in the prior art. Compared with simple linear adjustment, the S-shaped speed curve provides a smoother acceleration and deceleration process, significantly reducing the impact on the user's neck. The dynamic parameter adjustment mechanism ensures that the adjustment process can adapt to the changes in the user's physiological state in real time, further improving safety and comfort.

[0115] In some of the above embodiments of the present application, it is proposed to generate an S-shaped velocity curve that takes into account the user's neck bearing capacity according to the size of the position difference vector and the adjusted maximum velocity and maximum acceleration parameters to achieve smooth adjustment of the headrest. However, in this process, there is still the problem of how to more accurately consider the physiological characteristics of the user's neck and how to protect the user's neck safety to the greatest extent while ensuring the adjustment efficiency. Especially for some users with fragile necks, how to avoid possible discomfort or injury during the headrest adjustment process while achieving effective position adjustment is a technical problem that needs to be solved urgently.

[0116] In this regard, the present application further proposes that after obtaining the user's neck physiological parameters, the maximum safe acceleration of the user's neck is calculated based on the neck physiological parameters, and compared with the preset maximum acceleration parameters, and the smaller value is selected as the actual maximum acceleration parameter; according to the actual maximum acceleration parameter and the size of the position difference vector, the time and distance of the acceleration segment and the deceleration segment are calculated, and an S-shaped speed curve including an initial slow acceleration segment, an intermediate uniform speed segment and a final slow deceleration segment is generated.

[0117] The technical solution of this application obtains the user's neck physiological parameters and calculates the personalized maximum safe acceleration, taking into account the differences between individual elderly users, ensuring that the speed curve generated subsequently does not exceed the tolerance of the user's neck. This method not only takes into account the user's safety, but also can adapt to the physiological characteristics of different users and provide a more personalized adjustment experience.

[0118] Specifically, this application first obtains the user's neck physiological parameters. These parameters may include factors such as neck muscle strength, cervical flexibility, and age. Based on these parameters, the system can use a preset algorithm or query a pre-established correspondence table to calculate the maximum safe acceleration of the user's neck. For example, for a 65-year-old user with moderate neck muscle strength and good cervical flexibility, the system may calculate that the maximum safe acceleration is 0.5m / s².

[0119] Furthermore, the system compares the calculated maximum safe acceleration with the preset maximum acceleration parameter. The preset maximum acceleration parameter may be set based on the general population, such as 0.8m / s². The system will select the smaller of the two values ​​as the actual maximum acceleration parameter. In this example, the system will select 0.5m / s² as the actual maximum acceleration parameter. This dynamic adjustment mechanism ensures that the system always puts user safety first while ensuring adjustment efficiency.

[0120] According to the actual maximum acceleration parameters and the size of the position difference vector, the system will accurately calculate the time and distance of the acceleration and deceleration sections. For example, if the total distance the headrest needs to move is 30cm, the system may calculate that the acceleration and deceleration sections each require a distance of 5cm and a time of 1 second, and the intermediate uniform speed section requires a distance of 20cm and a time of 2 seconds. This precise planning ensures the smoothness and controllability of the headrest adjustment process.

[0121] Based on the above calculations, the system generates an S-shaped speed curve that includes an initial slow acceleration segment, an intermediate uniform speed segment, and a final slow deceleration segment. This curve design minimizes the impact on the neck of elderly users and provides a more comfortable adjustment experience.

[0122] The technical solution of this application can provide a safe, comfortable and personalized headrest adjustment experience for elderly users with limited hand mobility and mild cognitive impairment through such fine speed control. Especially in daily scenes such as eating, watching TV and resting, the system can generate the most suitable adjustment speed curve according to the individual characteristics of the user, which not only ensures the accuracy and efficiency of the adjustment, but also avoids unnecessary stimulation or damage to the user's neck.

[0123] As a preferred embodiment, the system can further combine the real-time monitoring force feedback and acceleration data to dynamically adjust the S-shaped speed curve. For example, if the user's neck muscles are detected to be abnormally tense during the adjustment process, the system can immediately reduce the current acceleration or speed, or even pause the adjustment process. This real-time feedback mechanism further improves the safety and comfort of the adjustment process.

[0124] Another possible implementation method is that the system can gradually optimize and adjust the calculation model of the maximum safe acceleration according to the user's usage habits and preferences. For example, if the system finds that a user can withstand higher acceleration without discomfort during multiple adjustments, the system can appropriately increase the user's maximum safe acceleration threshold, thereby improving the adjustment efficiency while ensuring safety.

[0125] The technical solution of the present application effectively solves the problem of how to protect the user's neck safety to the greatest extent while ensuring the adjustment efficiency by accurately considering the physiological characteristics of the user's neck, dynamically adjusting the maximum acceleration parameters, and generating a personalized S-shaped speed curve. This method not only improves the safety and comfort of headrest adjustment, but also can adapt to the individual differences of different users, providing a more considerate user experience for elderly users. Through this refined control, the technical solution of the present application can significantly reduce neck discomfort or injury that may be caused by improper adjustment of the headrest in daily use, thereby improving the comfort and safety of elderly users using electric wheelchairs.

[0126] In some of the above embodiments of the present application, a step of optimizing the corrected continuous scene description value in combination with the adjusted feature weight to obtain the final continuous scene description value is proposed to optimize the continuous scene description value. However, in this process, only considering the feature weight may not fully reflect the user's real-time physiological state, resulting in insufficient accuracy of scene recognition. In addition, a simple weighted average may not be able to effectively handle the complex relationship between features, affecting the accuracy of the final scene description value.

[0127] In this regard, the present application further proposes obtaining a corrected continuous scene description value X=(x1, x2, ..., xn) and an adjusted feature weight W=(w1, w2, ..., wn), where xi represents the description value of the i-th preset scene, wi represents the weight of the i-th feature, and n is the number of preset scenes; real-time collection of the user's physiological state data P=(p1, p2, ...,pm), where pj represents the j-th physiological state parameter, and m is the number of physiological state parameters; based on the pre-established scene-physiological state correspondence matrix S, the physiological state matching degree M=S·P is calculated, where S is an n×m matrix, and the feature weight and the physiological state matching degree are integrated to obtain a comprehensive feature weight W'=α·W+(1-α)·normalize(M), where α is a balance parameter, 0≤α≤1, and normalize is a normalization function; the final continuous scene description value Y=(y1, y2, ..., yn), where Y = softmax(β·X+(1-β)·W'), β is a balance parameter, 0≤β≤1.

[0128] The technical solution of this application introduces several key technical features, including real-time collection of user physiological state data, calculation of physiological state matching degree through scene-physiological state correspondence matrix, use of balance parameter α to fuse feature weight and physiological state matching degree to obtain comprehensive feature weight, and introduction of softmax function and balance parameter β to calculate the final continuous scene description value. These technical features work together to effectively solve the problems existing in previous solutions.

[0129] By introducing physiological state data and the scene-physiological state correspondence matrix, the system can more comprehensively consider the user's real-time state and improve the accuracy of scene recognition. The combination of balance parameters and softmax function enables the system to handle the relationship between features more flexibly, improving the accuracy of the final scene description value.

[0130] The technical solution of the present application has the following innovations: directly incorporating user physiological state data into the scene recognition process to achieve more personalized and dynamic scene recognition; efficiently integrating multiple data sources through matrix operations and normalization processing to improve computing efficiency; introducing multiple balance parameters to make the system more adaptable and robust, and able to be optimized according to different users and scenarios.

[0131] In the technical solution of the present application, firstly, real-time collection of the user's physiological state data P=(p1, p2, ..., pm) is the basis of the entire solution. These data can be obtained through a variety of sensors, such as heart rate sensors, blood pressure sensors, body temperature sensors, etc. The data type and quantity collected can be adjusted according to actual needs to adapt to different application scenarios.

[0132] The scenario-physiological state correspondence matrix S is pre-established, which reflects the relationship between different scenarios and the user's physiological state. This matrix can be obtained by training a machine learning algorithm based on a large amount of historical data, or it can be constructed by an expert system based on medical knowledge and experience. Each row of the matrix S corresponds to a preset scenario, and each column corresponds to a physiological state parameter. The matrix elements represent the contribution of a physiological state parameter to a scenario.

[0133] The calculation of physiological state matching degree M=S·P is achieved through matrix multiplication. This step matches the user's real-time physiological state with the preset scene to obtain the matching degree of each scene. The higher the matching degree, the more the user's current physiological state matches the scene.

[0134] One of the key steps of this scheme is to fuse the feature weight and the physiological state matching degree to obtain the comprehensive feature weight W'=α·W+(1-α)·normalize(M). The balance parameter α is introduced here to adjust the relative importance of the feature weight W and the physiological state matching degree M. The normalize function is used to normalize M to make it on the same order of magnitude as W. The value of α can be dynamically adjusted according to the actual application scenario. For example, when the user's physiological state fluctuates greatly, the α value can be increased to increase the impact of the physiological state matching degree.

[0135] Finally, the final continuous scene description value Y = softmax (β · X + (1-β) · W') is calculated by introducing the softmax function and another balancing parameter β. The use of the softmax function ensures that the output Y is a probability distribution with the sum of each element being 1, which is convenient for subsequent processing and decision-making. The β parameter is used to balance the influence of the corrected continuous scene description value X and the comprehensive feature weight W'.

[0136] There are close correlations and interactions between these features. For example, the physiological state data P is converted into the physiological state matching degree M through the scene-physiological state correspondence matrix S, and then combined with the feature weight W to form the comprehensive feature weight W'. This multi-level data fusion enables the system to capture user status and scene characteristics more comprehensively. At the same time, by introducing the balance parameters α and β, the system has the ability to dynamically adjust the influence of various factors and can adaptively adjust the strategy according to different situations.

[0137] In practical applications, the parameter range can be set according to specific needs. For example, α and β can be adjusted between 0.3 and 0.7 to balance the influence of different factors. The dimension of the matrix S can be determined according to the number of preset scenarios and the number of monitored physiological parameters. For example, a 10×5 matrix can represent 10 preset scenarios and 5 physiological parameters.

[0138] The technical solution of this application solves the problem of optimizing continuous scene description values ​​by collecting the user's physiological state data in real time, so that the system can capture the changes in the user's state in a timely manner. These data are combined with the pre-established scene-physiological state correspondence matrix to calculate the physiological state matching degree, thereby directly incorporating the user's real-time physiological state into the scene recognition process.

[0139] Next, the system obtains the comprehensive feature weight by fusing the feature weight and the physiological state matching degree. The key to this step is to introduce the balance parameter α, which enables the system to dynamically adjust the relative importance of feature weight and physiological state matching degree according to different situations. For example, when the user is in a stable state, he may be more inclined to rely on historical feature weights; when the user's state changes significantly, the system can consider the real-time physiological state matching degree more.

[0140] When calculating the final continuous scene description value, the system further introduces a softmax function and a balance parameter β. The use of the softmax function ensures that the output result is a probability distribution, which not only makes the result easier to interpret, but also provides a better basis for subsequent decision-making. The introduction of the balance parameter β allows the system to find the best balance between the corrected continuous scene description value and the comprehensive feature weight.

[0141] Compared with simple weighted average, this solution can better handle the complex relationship between features and improve the accuracy and stability of scene recognition. At the same time, by introducing multiple adjustable parameters, the system has stronger adaptability and flexibility, and can be adaptively adjusted according to the needs of different users, different environments and different time periods.

[0142] Compared with previous methods, the technical solution of this application introduces real-time physiological status data, so that the system can more accurately reflect the user's current condition and improve the accuracy of scene recognition. The introduced balance parameters α and β enable the system to dynamically adjust the weights of different factors according to actual conditions, enhancing the adaptability of the system. The softmax function is used to process the final output, so that the system can better handle the nonlinear relationship between features. Through matrix operations and normalization processing, efficient data fusion and processing are achieved. The system can be optimized according to the characteristics of different users and scenarios to provide more personalized headrest adjustment services.

[0143] Second, refer to Figure 2 The present application further proposes a headrest adjustment device for an electric wheelchair, the device comprising: The acquisition module 210 is used to acquire the current positioning data of the electric wheelchair, the user's head pressure data, the headrest angle data and the usage time data when it is detected that the user triggers the headrest adjustment instruction through a one-button operation; An identification module 220, for identifying a current usage scenario based on positioning data, user's head pressure data, headrest angle data, and usage time data; A determination module 230, for determining a target position of the headrest according to the identified current usage scenario and pre-stored personalized parameters; A generating module 240, for calculating a position difference between a current position of the headrest and a target position, and generating a headrest adjustment trajectory based on the position difference; The control module 250 is used to control the headrest to perform adjustment actions according to the headrest adjustment trajectory, while monitoring the force feedback and acceleration during the adjustment process, and automatically stop or reduce the adjustment speed when an abnormality is detected.

[0144] Through one-button operation and intelligent scene recognition, the operation process is simplified and suitable for users with limited hand movement ability; through multi-dimensional data analysis and dynamic adjustment, intelligent scene recognition and personalized adjustment are realized; through the generation of smooth adjustment trajectory and real-time monitoring, the smoothness and safety of the adjustment process are guaranteed. It has the advantages of easy operation, intelligent scene recognition, smooth and safe adjustment, and meeting personalized needs.

[0145] In addition, in some preferred embodiments, an electric wheelchair headrest adjustment device proposed in the present application can perform any one of the steps in the above method.

[0146] The above description is only an embodiment of the present application and is not intended to limit the protection scope of the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for adjusting a headrest of an electric wheelchair, characterized in that: The method includes: When it is detected that the user triggers the headrest adjustment instruction through a one-touch operation, the current positioning data of the electric wheelchair, the user's head pressure data, the headrest angle data and the usage time data are obtained; Identify the current usage scenario based on positioning data, the user's head pressure data, headrest angle data, and usage time data; determining a target position of the headrest based on the identified current usage scenario and pre-stored personalized parameters; Calculating a position difference between a current position of the headrest and a target position, and generating a headrest adjustment trajectory based on the position difference; The headrest is controlled to perform adjustment actions according to the headrest adjustment trajectory, and force feedback and acceleration during the adjustment process are monitored. When an abnormality is detected, the adjustment speed is automatically stopped or reduced.

2. The method for adjusting the headrest of an electric wheelchair according to claim 1, characterized in that: The step of identifying the current usage scenario according to the positioning data, the user's head pressure data, the headrest angle data and the usage time data comprises: Construct a multidimensional feature space, and map the acquired positioning data, head pressure data, headrest angle data, and usage time data in the multidimensional feature space as a feature vector, where each dimension corresponds to a data type; Analyze the feature vector to obtain a continuous scene description value representing the current state of the user; The current usage scenario is dynamically determined based on the continuous scenario description value in combination with the stored historical usage data of the user and the physiological status data collected in real time.

3. The method for adjusting the headrest of an electric wheelchair according to claim 2, characterized in that: The step of analyzing the feature vector to obtain a continuous scene description value representing the current state of the user includes: When a change in the user state is detected, the feature vector is analyzed to obtain an initial continuous scene description value; Based on the changing trend of the historical continuous scene description values, the initial continuous scene description values ​​are time-series corrected; Dynamically adjust the weight of each feature according to the importance of different features in the current time period, wherein the feature includes at least one of head pressure distribution, headrest angle and usage time; Combined with the adjusted feature weights, the corrected continuous scene description value is optimized to obtain the final continuous scene description value.

4. The method for adjusting the headrest of an electric wheelchair according to claim 3, characterized in that: The step of performing time series correction on the initial continuous scene description values ​​based on the change trend of the historical continuous scene description values ​​comprises: Based on the change trend of the historical continuous scene description value, a sequence of historical continuous scene description values ​​within a specified time window is obtained, and a state transition probability matrix is ​​established, wherein the state transition probability matrix reflects the user's transition rules between various preset scenes; The confidence of the initial continuous scene description value is calculated according to the state transition probability matrix. When the confidence is lower than a preset threshold, the initial continuous scene description value is corrected based on the state transition probability matrix to obtain a corrected continuous scene description value.

5. The method for adjusting the headrest of an electric wheelchair according to claim 3, characterized in that: The step of optimizing the corrected continuous scene description value in combination with the adjusted feature weight to obtain the final continuous scene description value comprises: After obtaining the user's physiological state data, the physiological state matching degree of each preset scene is calculated based on the preset scene-physiological state correspondence relationship; The physiological state matching degree is used as an additional feature and is integrated with the adjusted feature weight to obtain a comprehensive feature weight; The corrected continuous scene description values ​​are weighted averaged using the comprehensive feature weights to obtain a final continuous scene description value.

6. The method for adjusting the headrest of an electric wheelchair according to claim 4, characterized in that: The step of correcting the initial continuous scene description value based on the state transition probability matrix comprises: After obtaining the user's historical scene conversion data and current physiological state data, the scene conversion frequency of the user in different time periods is calculated based on the historical scene conversion data, and the preset threshold is dynamically adjusted according to the scene conversion frequency and the current physiological state data; When the confidence of the initial continuous scene description value is lower than the preset threshold value after dynamic adjustment, the initial continuous scene description value is corrected based on the state transition probability matrix to obtain a corrected continuous scene description value.

7. The method for adjusting the headrest of an electric wheelchair according to claim 1, characterized in that: The step of calculating the position difference between the current position of the headrest and the target position, and generating the headrest adjustment trajectory based on the position difference comprises: After obtaining the current position coordinates and the target position coordinates of the headrest, the position difference vector is calculated based on the current position coordinates and the target position coordinates, and the maximum speed and maximum acceleration parameters are dynamically adjusted in combination with the user's physiological state data; According to the size of the position difference vector and the adjusted maximum speed and maximum acceleration parameters, an S-shaped speed curve is generated that takes into account the endurance of the user's neck, wherein the S-shaped speed curve includes an initial slow acceleration segment, an intermediate uniform speed segment, and a final slow deceleration segment; Based on the S-shaped velocity curve and the position difference vector, the intermediate position coordinate sequence of the headrest during the adjustment process is calculated to form a smooth headrest adjustment trajectory that adapts to the user's physiological characteristics.

8. The method for adjusting the headrest of an electric wheelchair according to claim 7, characterized in that: The step of generating an S-shaped speed curve considering the user's neck tolerance according to the size of the position difference vector and the adjusted maximum speed and maximum acceleration parameters comprises: After obtaining the user's neck physiological parameters, calculating the maximum safe acceleration of the user's neck based on the neck physiological parameters, and comparing the maximum safe acceleration with the preset maximum acceleration parameter, and selecting the smaller value as the actual maximum acceleration parameter; According to the actual maximum acceleration parameters and the size of the position difference vector, the time and distance of the acceleration and deceleration segments are calculated, and an S-shaped velocity curve including an initial slow acceleration segment, an intermediate uniform speed segment and a final slow deceleration segment is generated.

9. The method for adjusting the headrest of an electric wheelchair according to claim 3, characterized in that: The step of optimizing the corrected continuous scene description value in combination with the adjusted feature weight to obtain the final continuous scene description value comprises: Obtain the corrected continuous scene description value X=(x1, x2, ..., xn) and the adjusted feature weight W=(w1, w2, ..., wn), where xi represents the description value of the i-th preset scene, wi represents the weight of the i-th feature, n is the number of preset scenes, and collect the user's physiological state data P=(p1, p2, ..., pm) in real time, where pj represents the j-th physiological state parameter, and m is the number of physiological state parameters; Based on the pre-established scene-physiological state correspondence matrix S, the physiological state matching degree M=S·P is calculated, where S is an n×m matrix, and the feature weight and physiological state matching degree are fused to obtain the comprehensive feature weight W'=α·W+(1-α)·normalize(M), where α is the balance parameter, 0≤α≤1, and normalize is the normalization function; Calculate the final continuous scene description value Y=(y1, y2, ..., yn), where Y=softmax(β·X+(1-β)·W'), β is the balance parameter, 0≤β≤1.

10. An electric wheelchair headrest adjustment device, characterized in that: The device includes: An acquisition module is used to acquire the current positioning data of the electric wheelchair, the user's head pressure data, the headrest angle data and the usage time data when it is detected that the user triggers the headrest adjustment instruction through a one-button operation; An identification module, used to identify the current usage scenario based on positioning data, user's head pressure data, headrest angle data and usage time data; a determination module, for determining a target position of the headrest according to the identified current usage scenario and pre-stored personalized parameters; A generating module, configured to calculate a position difference between a current position of the headrest and a target position, and to generate a headrest adjustment trajectory based on the position difference; The control module is used to control the headrest to perform adjustment actions according to the headrest adjustment trajectory, while monitoring the force feedback and acceleration during the adjustment process, and automatically stop or reduce the adjustment speed when an abnormality is detected.

Images