Intelligent office table self-adaptive control method and system based on artificial intelligence

Abstract

The invention discloses an intelligent office table self-adaptive control method and system based on artificial intelligence. The method comprises the following steps: S1, multi-modal data acquisition; s2, data preprocessing; s3, feature extraction; s4, data fusion and state identification; and S5, executing an intervention strategy. Three types of data including visual information, physiological signals and pressure distribution are fused to perform multi-dimensional state sensing of multi-modal data fusion, sitting posture key points are extracted based on a convolutional neural network (CNN), real-time recognition of bad sitting postures is realized, time sequence characteristics of pressure distribution are analyzed through a long-short-term memory network (LSTM), behavior modes such as sedentariness and heeling are recognized, and the sitting posture recognition accuracy is improved. According to the method, the association between operation habits and fatigue is mined, HRV frequency domain features are extracted by using Fourier transform, and the fatigue state is comprehensively evaluated in combination with multiple indexes, so that three-dimensional description of different states of different users is realized, and whether the users are fatigued or not is judged to provide more refined health management.

Description

Technical Field

[0001] This invention belongs to the field of intelligent office technology, and in particular relates to an adaptive control method and system for intelligent office desks based on artificial intelligence. Background Technology

[0002] Smart desks are office furniture that combines modern technology with ergonomic design, aiming to improve work efficiency, enhance the work environment, and prioritize user health and comfort. With the development of office automation, health problems caused by prolonged desk work are receiving increasing attention.

[0003] Traditional office desks have limited functionality and cannot monitor users' physical condition and work behavior in real time, making it difficult to provide effective health protection. While some existing smart office desks have some monitoring functions, they are insufficient in terms of the comprehensiveness of data collection, the accuracy of multimodal data fusion, and the personalization of adaptive intervention strategies, thus failing to achieve efficient and precise health intervention. Summary of the Invention

[0004] To address the problems in the prior art, the present invention proposes the following technical solution:

[0005] An AI-based adaptive control method for smart office desks includes the following steps:

[0006] S1, Multimodal data acquisition;

[0007] Multimodal sensor modules are deployed on smart desks to collect data synchronously at different sampling frequencies, acquiring visual information data, physiological signal data, and pressure data of the parts in contact with the desk surface during the user's work.

[0008] S2, Data Preprocessing;

[0009] The edge computing terminal preprocesses the collected raw data to reduce the impact of external factors on the data. It prioritizes left and right hands and uses different preprocessing parameters and algorithms based on the different usage habits and physiological characteristics of the left and right hands.

[0010] S3, Feature Extraction;

[0011] Different algorithms are used to extract features from the preprocessed data. For visual data, convolutional neural networks (CNNs) are used to extract key points of sitting posture, and the user's sitting posture is analyzed through information such as neck angle and head tilt angle. For stress data, including overall desktop stress data and left and right hand stress data, the temporal characteristics of stress distribution are analyzed based on long short-term memory networks (LSTMs) to identify behaviors such as prolonged sitting and tilting, as well as different operating behaviors of the left and right hands. For physiological data, HRV frequency domain features are extracted through Fourier transform, and combined with other physiological indicators to comprehensively assess the user's physical fatigue status.

[0012] S4. Data fusion and status recognition;

[0013] The Dynamic Time Warping (DTW) algorithm is used to align time series data from multiple sources, fusing visual data features, stress data features, and physiological data features. State confidence is calculated using evidence theory, and the user's current state label is output.

[0014] S5. Implementation of intervention strategies;

[0015] Based on the status label output by the status recognition, the preset control strategy is invoked, and three-level treatment plans for mild fatigue, moderate fatigue and severe fatigue are set and processed according to different fatigue levels.

[0016] As a preferred embodiment of the above technical solution, it also includes: S6, intervention effect monitoring and parameter optimization;

[0017] Real-time monitoring of intervention effects includes monitoring changes in physiological indicators such as whether HRV recovers, and using visual and stress data to determine whether the user's posture has improved.

[0018] As a preferred embodiment of the above technical solution, it also includes: S7, model update;

[0019] The system stores user response data in real time and triggers fine-tuning of intervention strategies based on user behavior, reducing or increasing the processing time of contingency plans.

[0020] As a preferred embodiment of the above technical solution, the three-level processing plan in step S5 includes the following:

[0021] For mild fatigue, i.e., HRV decreases by 10%-20% and SCL increases by <5μS, the lighting system is triggered to provide a warm color tone reminder, reducing the color temperature to 3000K to create a comfortable office atmosphere and relieve fatigue.

[0022] For moderate fatigue, when HRV decreases by 20%-30% and is accompanied by head tilting forward >15°, activate the desktop vibration reminder and give a "suggestion to get up and move around" via voice.

[0023] For severe fatigue, when HRV decreases by more than 30% and the pressure sensor detects that the user has been sitting for more than 90 minutes, the desktop is forcibly locked and automatically adjusted to standing mode, raising the height to 110cm to force the user to change their working posture and relieve physical fatigue.

[0024] As a preferred embodiment of the above technical solution, the preprocessing of the raw data in step S2 includes the following:

[0025] S2a. For visual image data, advanced image denoising algorithms are used to remove noise interference and improve image quality.

[0026] S2b. For physiological signal data, perform baseline correction to eliminate interference factors such as baseline drift in the signal;

[0027] S2c: For pressure signal data, filter the data to remove high-frequency noise.

[0028] The system of an AI-based intelligent office desk adaptive control method includes: a multimodal sensor module and a desktop. The multimodal sensor module includes a visual sensor, a pressure sensor matrix, and a physiological indicator sensor. The visual sensor is a camera set above or on both sides of the desktop to collect image data such as the user's sitting posture, head posture, and hand movements. The pressure sensor matrix is ​​a thin-film pressure sensor embedded in the desktop and chair surfaces to monitor the user's contact pressure distribution and determine the shift of the body's center of gravity and the duration of sitting. The physiological indicator sensor is a non-contact heart rate monitoring module integrated at the edge of the desktop to collect fatigue-related physiological signals such as heart rate (HR) and heart rate variability (HRV) in real time.

[0029] As a preferred embodiment of the above technical solution, an electric lifting table frame is provided at the bottom of the desktop, which is used to raise the height of the desktop and control the vibration of the desktop. A lighting system is provided above and on both sides of the desktop, which is used to control the light.

[0030] As a preferred embodiment of the above technical solution, it further includes: an edge computing terminal, wherein the edge computing terminal is equipped with an integrated convolutional neural network (CNN), long short-term memory network (LSTM) model and dynamic time warping (DTW) and Fourier transform algorithm.

[0031] The beneficial effects of this invention are as follows:

[0032] 1. This system integrates visual information, physiological signals, and pressure distribution data for multimodal data fusion and multidimensional state perception. During preprocessing, it differentiates between left and right hands during work, enabling more accurate acquisition of user work behavior and physical state information. Based on a convolutional neural network (CNN), it extracts key points of sitting posture to achieve real-time identification of poor posture. By analyzing the temporal characteristics of pressure distribution through a long short-term memory network (LSTM), it identifies behavioral patterns such as prolonged sitting and tilting, uncovers the correlation between operating habits and fatigue, and utilizes Fourier transform to extract HRV frequency domain features. Combined with multiple indicators, it comprehensively assesses fatigue status, thereby achieving a three-dimensional characterization of different users' states and determining whether users are fatigued to provide more refined health management.

[0033] 2. Combining advanced data processing and fusion algorithms, it achieves accurate identification of the user's work status and then progressively implements adaptive health intervention strategies based on the identification results, from environmental adjustment (lighting), tactile reminders (vibration), to physical coercion (standing mode). The intervention intensity is matched according to the fatigue level, which avoids excessive disturbance when the user is mildly fatigued, and can forcibly protect the user's health when the user is severely fatigued. It can effectively relieve the user's physical fatigue, correct bad posture, and protect the user's physical health. Attached Figure Description

[0034] Figure 1 The diagram shown is a schematic representation of the workflow of an embodiment. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments and the accompanying drawings.

[0036] Example

[0037] Figure 1 The adaptive control method for an intelligent office desk based on artificial intelligence includes the following steps:

[0038] S1, Multimodal data acquisition;

[0039] Multimodal sensor modules are deployed on smart desks to collect data synchronously at different sampling frequencies, acquiring visual information data, physiological signal data, and pressure data of the parts in contact with the desk surface during the user's work.

[0040] S2, Data Preprocessing;

[0041] The edge computing terminal preprocesses the collected raw data to reduce the impact of external factors on the data. It prioritizes left and right hands and uses different preprocessing parameters and algorithms based on the different usage habits and physiological characteristics of the left and right hands.

[0042] S3, Feature Extraction;

[0043] Different algorithms were used to extract features from the preprocessed data. For visual data, convolutional neural networks (CNNs) were used to extract key points of sitting posture. The user's sitting posture was analyzed by information such as neck angle and head tilt angle. Based on the user's left and right hand preferences, some habits of left-handed users were input, and specific sitting postures of left-handed users were not used as key points of sitting posture. For stress data, including overall desktop stress data and left and right hand stress data, the temporal characteristics of stress distribution were analyzed based on long short-term memory networks (LSTM) to identify behaviors such as prolonged sitting and tilting, as well as different operating behaviors of the left and right hands. For physiological data, HRV frequency domain features were extracted by Fourier transform and combined with other physiological indicators to comprehensively assess the user's physical fatigue status.

[0044] S4. Data fusion and status recognition;

[0045] The Dynamic Time Warping (DTW) algorithm is used to align time series data from multiple sources, fusing visual data features, stress data features, and physiological data features. State confidence is calculated using evidence theory, and the user's current state label is output.

[0046] S5. Implementation of intervention strategies;

[0047] Based on the status label output by the status recognition, the preset control strategy is invoked, and three-level treatment plans for mild fatigue, moderate fatigue and severe fatigue are set and processed according to different fatigue levels.

[0048] For right-handed users, right-hand pressure signal data may contain more high-frequency pressure changes caused by fine operations, so a filter more suitable for extracting high-frequency features should be used during filtering. For left-handed pressure signal data, a corresponding preprocessing method should be used based on its general pressure distribution and change characteristics, while the method should be reversed for left-handed users, so as to accurately distinguish the pressure state and behavior pattern of the left and right hands during work.

[0049] This system integrates visual information, physiological signals, and pressure distribution data for multimodal data fusion and multidimensional state perception. It also incorporates differentiated processing of left and right hands during work, enabling more accurate acquisition of user work behavior and physical state information. Based on a convolutional neural network (CNN), it extracts key points of sitting posture for real-time identification of poor posture. By analyzing the temporal characteristics of pressure distribution through a long short-term memory network (LSTM), it identifies behavioral patterns such as prolonged sitting and tilting, uncovers the correlation between operating habits and fatigue, and utilizes Fourier transform to extract HRV frequency domain features. Combined with multiple indicators, it comprehensively assesses fatigue status, thereby achieving a three-dimensional characterization of different users' states and determining whether users are fatigued to provide more refined health management.

[0050] It also includes: S6, monitoring of intervention effects;

[0051] The intervention effect is monitored in real time, including changes in physiological indicators such as whether HRV rebounds, and whether the user's posture has improved through visual and stress data. At the same time, an incremental learning algorithm is used to dynamically adjust the intervention thresholds at each level based on the user's historical data. Taking into account the differences in physiological characteristics of different groups, personalized intervention threshold settings are achieved to improve the effectiveness of the intervention strategy.

[0052] Also includes: S7, model update;

[0053] The system stores user response data in real time and triggers fine-tuning of intervention strategies based on user behavior, reducing or increasing the processing time of contingency plans.

[0054] By storing user response data in real time, the system can automatically optimize intervention strategy parameters and achieve self-evolution.

[0055] In step S5, the three-level processing plan includes the following:

[0056] For mild fatigue, i.e., HRV decreases by 10%-20% and SCL increases by <5μS, the lighting system is triggered to provide a warm color tone reminder, reducing the color temperature to 3000K to create a comfortable office atmosphere and relieve fatigue.

[0057] For moderate fatigue, when HRV decreases by 20%-30% and is accompanied by head tilting forward >15°, activate the desktop vibration reminder and give a "suggestion to get up and move around" via voice.

[0058] For severe fatigue, when HRV decreases by more than 30% and the pressure sensor detects that the user has been sitting for more than 90 minutes, the desktop is forcibly locked and automatically adjusted to standing mode, raising the height to 110cm to force the user to change their working posture and relieve physical fatigue.

[0059] Combining advanced data processing and fusion algorithms, it achieves accurate identification of the user's work status and then progressively implements adaptive health intervention strategies based on the identification results, from environmental adjustment (lighting), tactile reminders (vibration), to physical coercion (standing mode). The intervention intensity is matched according to the fatigue level, which avoids excessive disturbance when the user is mildly fatigued, and can forcibly protect the user's health when the user is severely fatigued. It can effectively relieve the user's physical fatigue, correct bad posture, and protect the user's physical health.

[0060] The preprocessing of the raw data in step S2 includes the following:

[0061] S2a. For visual image data, advanced image denoising algorithms are used to remove noise interference and improve image quality.

[0062] S2b. For physiological signal data, perform baseline correction to eliminate interference factors such as baseline drift in the signal;

[0063] S2c: For pressure signal data, filter the data to remove high-frequency noise.

[0064] The system of an AI-based intelligent office desk adaptive control method includes: a multimodal sensor module and a desktop. The multimodal sensor module includes a visual sensor, a pressure sensor matrix, and a physiological indicator sensor. The visual sensor is a camera set above or on both sides of the desktop to collect image data such as the user's sitting posture, head posture, and hand movements. The pressure sensor matrix is ​​a thin-film pressure sensor embedded in the desktop and chair surfaces to monitor the user's contact pressure distribution and determine the shift of the body's center of gravity and the duration of sitting. The physiological indicator sensor is a non-contact heart rate monitoring module integrated at the edge of the desktop to collect fatigue-related physiological signals such as heart rate (HR) and heart rate variability (HRV) in real time.

[0065] The bottom of the desktop is equipped with an electric lifting table frame, which is used to raise the height of the desktop and control the vibration of the desktop. A lighting system is provided above and on both sides of the desktop, which is used to control the light.

[0066] It also includes: an edge computing terminal, which is equipped with an integrated convolutional neural network (CNN), long short-term memory network (LSTM) model and dynamic time warping (DTW) and Fourier transform algorithm.

[0067] Example

[0068] When a user is working on a weekday morning, the smart desk's multimodal sensors begin to collect data synchronously. The visual sensor captures images of the user's work environment at 30fps, the physiological signal sensor collects physiological signals such as heart rate, HRV, and SCL at 100Hz, and the pressure signal sensor collects overall desktop pressure and left and right hand pressure data at 10Hz.

[0069] The edge computing terminal preprocesses the collected raw data. For visual images, noise is removed using denoising algorithms to make the images clearer; for physiological signals, baseline correction is performed to ensure signal accuracy; for pressure signals, the overall pressure on the desktop and the pressure from the left and right hands are filtered separately, using different filtering parameters and algorithms based on the different characteristics of the left and right hands to accurately distinguish the pressure states of the left and right hands. For example, if a user is right-handed, the right-hand pressure signal contains high-frequency key pressure changes during keyboard input. These high-frequency features are extracted using an appropriate filter, and the left-hand pressure signal is processed accordingly to obtain accurate pressure information of the left and right hands during operation.

[0070] In the feature extraction stage, CNN is used to extract key points of the user's sitting posture, such as the shoulder and neck angle and head tilt angle, from the visual data; LSTM is used to analyze the temporal features of the overall desktop pressure and left and right hand pressure data to identify the user's sitting posture and left and right hand operation behavior; and Fourier transform is used to extract HRV frequency domain features from physiological data to assess the user's physical fatigue state.

[0071] Then, the DTW algorithm was used to align the multi-source data time series, and the extracted features were fused. The confidence level of the state was calculated using evidence theory to determine that the user was currently in a "focused work" state with a confidence level of 95%. As time progressed, based on changes in the user's physiological and stress data, it was determined that the user had entered a state of mild fatigue, at which point HRV decreased by 15% and SCL increased by 3μS. The smart desk, according to a preset control strategy, triggered a warm-toned LED light reminder, lowering the color temperature to 3000K. For 30 minutes after the adjustment, the system continuously monitored changes in the user's HRV and other physiological indicators, as well as posture. Reinforcement learning was used to optimize subsequent reminder frequencies and other control parameters, and an incremental learning algorithm was used to dynamically adjust the fatigue intervention threshold based on the user's historical data to achieve more personalized and effective health intervention.

[0072] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it.

Claims

1. An adaptive control method for an intelligent office desk based on artificial intelligence, characterized in that: Includes the following steps: S1, Multimodal data acquisition; Multimodal sensor modules are deployed on smart desks to collect data synchronously at different sampling frequencies, acquiring visual information data, physiological signal data, and pressure data of the parts in contact with the desk surface during the user's work. S2, Data Preprocessing; The edge computing terminal preprocesses the collected raw data to reduce the impact of external factors on the data. It prioritizes left and right hands and uses different preprocessing parameters and algorithms based on the different usage habits and physiological characteristics of the left and right hands. S3, Feature Extraction; Different algorithms are used to extract features from the preprocessed data. For visual data, convolutional neural networks (CNNs) are used to extract key points of sitting posture, and the user's sitting posture is analyzed through information such as neck angle and head tilt angle. For stress data, including overall desktop stress data and left and right hand stress data, the temporal characteristics of stress distribution are analyzed based on long short-term memory networks (LSTMs) to identify behaviors such as prolonged sitting and tilting, as well as different operating behaviors of the left and right hands. For physiological data, HRV frequency domain features are extracted through Fourier transform, and combined with other physiological indicators to comprehensively assess the user's physical fatigue status. S4. Data fusion and status recognition; The Dynamic Time Warping (DTW) algorithm is used to align time series data from multiple sources, fusing visual data features, stress data features, and physiological data features. State confidence is calculated using evidence theory, and the user's current state label is output. S5. Implementation of intervention strategies; Based on the status label output by the status recognition, the preset control strategy is invoked, and three-level treatment plans for mild fatigue, moderate fatigue and severe fatigue are set and processed according to different fatigue levels.

2. The adaptive control method for an intelligent office desk based on artificial intelligence according to claim 1, characterized in that, Also includes: S6. Monitoring the intervention effect; Real-time monitoring of intervention effects includes monitoring changes in physiological indicators such as whether HRV recovers, and using visual and stress data to determine whether the user's posture has improved.

3. The adaptive control method for an intelligent office desk based on artificial intelligence according to claim 1, characterized in that, Also includes: S7, Model Update; The system stores user response data in real time and triggers fine-tuning of intervention strategies based on user behavior, reducing or increasing the processing time of contingency plans.

4. The adaptive control method for an intelligent office desk based on artificial intelligence according to claim 1, characterized in that, In step S5, the three-level processing plan includes the following: For mild fatigue, i.e., HRV decreases by 10%-20% and SCL increases by <5μS, the lighting system is triggered to provide a warm color tone reminder, reducing the color temperature to 3000K to create a comfortable office atmosphere and relieve fatigue. For moderate fatigue, when HRV decreases by 20%-30% and is accompanied by head tilting forward >15°, activate the desktop vibration reminder and give a "suggestion to get up and move around" via voice. For severe fatigue, when HRV decreases by more than 30% and the pressure sensor detects that the user has been sitting for more than 90 minutes, the desktop is forcibly locked and automatically adjusted to standing mode, raising the height to 110cm to force the user to change their working posture and relieve physical fatigue.

5. The adaptive control method for an intelligent office desk based on artificial intelligence according to claim 1, characterized in that, The preprocessing of the raw data in step S2 includes the following: S2a. For visual image data, advanced image denoising algorithms are used to remove noise interference and improve image quality. S2b. For physiological signal data, perform baseline correction to eliminate interference factors such as baseline drift in the signal; S2c: For pressure signal data, filter the data to remove high-frequency noise.

6. A system for adaptive control of an intelligent office desk based on artificial intelligence, characterized in that, include: The multimodal sensor module and desktop are described. The multimodal sensor module includes a visual sensor, a pressure sensor matrix, and a physiological indicator sensor. The visual sensor is a camera positioned above or on both sides of the desktop to collect image data such as the user's sitting posture, head posture, and hand movements. The pressure sensor matrix is ​​a thin-film pressure sensor embedded in the desktop and seat surface to monitor the user's contact pressure distribution and determine the shift of the body's center of gravity and the duration of sitting. The physiological indicator sensor is a non-contact heart rate monitoring module integrated at the edge of the desktop to collect fatigue-related physiological signals such as heart rate (HR) and heart rate variability (HRV) in real time.

7. The system of the intelligent office desk adaptive control method based on artificial intelligence according to claim 6, characterized in that, The bottom of the desktop is equipped with an electric lifting table frame, which is used to raise the height of the desktop and control the vibration of the desktop. A lighting system is provided above and on both sides of the desktop, which is used to control the light.

8. The system of the intelligent office desk adaptive control method based on artificial intelligence according to claim 6, characterized in that, Also includes: The edge computing terminal is equipped with an integrated convolutional neural network (CNN), long short-term memory network (LSTM) model, dynamic time warping (DTW), and Fourier transform algorithm.

Images