Posture correction method based on electric lifting table

Abstract

The application provides a sitting posture correction method based on an electric lifting table, which comprises the following steps: S100, acquiring user sitting posture data through a sitting posture recognition unit, and judging whether the user sitting posture conforms to a preset standard sitting posture; if the user sitting posture does not conform to the standard sitting posture, executing step S200; S200, executing a correction step, reminding the user according to a preset correction action, and judging whether the correction is successful; if the correction is successful, the execution frequency of step S200 is cleared, the electric lifting table is restored to a state before the correction step is executed, and the method returns to step S100; if the correction fails, executing step S300; S300, judging whether the execution frequency of step S200 reaches a preset threshold value; if the execution frequency of step S200 does not reach the preset threshold value, returning to step S200; if the execution frequency of step S200 has reached the preset threshold value, locking step S200, and waiting for manual unlocking; wherein the intensity of the correction action increases with the increase of the execution frequency of step S200, and the correction action comprises at least one of light prompting, voice prompting and table board lifting.

Description

Technical Field

[0001] This invention relates to the field of electric height-adjustable desks, and more particularly to a posture correction method based on an electric height-adjustable desk. Background Technology

[0002] In the field of furniture and office equipment technology, electric height-adjustable desks have gradually become common equipment in office and home settings because they can adjust the desktop height according to the user's height or usage needs, improving comfort during long periods of sitting. Traditional electric height-adjustable desks typically include a desk body, retractable legs, and a drive mechanism. Users control the drive mechanism via buttons on the control panel to extend and retract the retractable legs, thereby adjusting the desktop height to suit their usage habits. For example, the desktop can be raised when the user is standing and lowered when sitting.

[0003] In actual use, users are prone to poor posture when sitting for long periods, such as slouching, looking down, and excessive forward head tilting. They often fail to notice these postures in real time, which can lead to health problems such as cervical strain, lumbar spine deformation, and myopia over time. To monitor user posture in real time, existing electric height-adjustable desks are equipped with posture recognition units. These units collect user posture data, and the control unit determines whether the user's posture conforms to standard posture. If the user's posture is deemed incorrect, a light or sound prompt is given to remind them to adjust their posture. However, existing solutions only attempt to evoke self-correction through a single, fixed-intensity prompt. Users often ignore these prompts due to focus on work or study, resulting in a low success rate of posture correction. Summary of the Invention

[0004] To address the technical problem that existing posture correction methods for electric height-adjustable desks only provide single prompts of fixed intensity, resulting in poor correction effects, this invention provides a posture correction method based on an electric height-adjustable desk. By executing the correction action with progressively increasing intensity, the intensity of the correction reminder can be gradually increased, thereby effectively improving the user's perception of poor posture and guiding the user to instinctively adjust back to the standard posture, thus improving the success rate of correction.

[0005] To achieve the above-mentioned technical objectives, the present invention provides a posture correction method based on an electrically adjustable desk, wherein the electrically adjustable desk includes: Tabletop; At least one retractable table leg driven by a motor is used to raise and lower the tabletop; The posture recognition unit is used to collect the user's posture data; The posture correction method includes: Step S100: Obtain user sitting posture data through the sitting posture recognition unit and determine whether the user sitting posture conforms to the preset standard sitting posture. If the user's sitting posture does not conform to the standard sitting posture, proceed to step S200; Step S200: Perform the correction steps, remind the user according to the preset correction actions, and determine whether the correction is successful; If the correction is successful, the execution count of step S200 is reset to zero, the electric lifting table is restored to the state before the correction step was executed, and the process returns to step S100. If the correction fails, proceed to step S300; Step S300: Determine whether the number of executions in step S200 has reached a preset threshold; If the number of times step S200 is executed does not reach the preset threshold, return to step S200; If the number of times step S200 has been executed has reached a preset threshold, step S200 is locked and the process waits for manual unlocking. The intensity of the corrective action increases with the number of times step S200 is executed, and the corrective action includes at least one of light prompts, voice prompts, and table lifting.

[0006] The technical effects of this invention are as follows: Unlike existing technologies that use a fixed-intensity, single-time prompt, this invention increases the intensity of the corrective action as the number of times step S200 is executed. This gradually increases the intensity of the corrective reminder, thus gradually overcoming the perceptual blockage caused by the user's focus on work or study. The corrective reminder changes from being "easily ignored" to "not easily ignored," effectively guiding the user to instinctively adjust back to a standard sitting posture and improving the success rate of correction. Furthermore, the graded increasing intensity reminder mechanism avoids excessive interference from a single high-intensity stimulus while ensuring that even when the user's awareness is weak, the progressively increasing reminder intensity ultimately achieves the corrective goal, effectively improving the success rate. Finally, after successful correction, this invention restores the electric height-adjustable desk to its state before the correction step, ensuring that the user's original desktop height usage habits are not changed. After multiple failed correction attempts in step S200, step S200 is locked and awaits manual release. This design avoids the interference of infinite loop correction on the user experience and prevents excessive wear on the motor and other actuators due to frequent ineffective actions, improving the safety and reliability of the system operation.

[0007] In the above-mentioned posture correction method, the intensity of the corrective action is the output intensity of a single corrective action. Increasing the output intensity of a single corrective action includes at least one of the following: Increase the lifting stroke and / or lifting speed of the tabletop; Increase the output intensity of the light; Enhance the output strength of the speech.

[0008] Under the same type of corrective action, by continuously increasing the output intensity of a single physical quantity, stimulation can be applied in a progressive manner that is perceptible to the user, so as to gradually improve the user's perception of the corrective action reminder, thereby effectively guiding the user to adjust to the standard sitting posture. Specifically, regarding the increased lifting range of the tabletop, gradually raising the tabletop forces the user's elbows and forearms to be raised progressively, causing discomfort in the arms and guiding the user to adjust to a standard sitting posture. Regarding the increased lifting speed, increasing the lifting speed allows the tabletop to complete the lifting action in a shorter time, producing a more noticeable displacement sensation, thus enhancing the user's tactile feedback and making it easier for them to perceive changes in the tabletop, guiding them to adjust to a standard sitting posture. Regarding lighting cues, gradually increasing the light output intensity changes the light from soft to bright, creating an increasingly strong visual signal in the user's peripheral vision, gradually breaking through the visual barrier of a focused state, making the user feel glare, and thus guiding them to adjust to a standard sitting posture. Regarding voice cues, gradually increasing the voice output intensity changes the cues from soft to clear, gradually penetrating the auditory barrier of a focused state, thus guiding the user to adjust to a standard sitting posture.

[0009] In the above-described posture correction method, the intensity of the corrective action is determined by the number of combinations of corrective actions. As the number of executions of step S200 increases, more types of corrective actions are superimposed. This design, by superimposing multiple types of corrective actions to achieve incremental intensity, utilizes the synergistic effect of multimodal stimulation, significantly enhancing the user's perception probability. Specifically, visual stimuli (light), auditory stimuli (voice), and somatosensory stimuli (table lifting) act on different human sensory channels. When a single-channel stimulus is ignored by the user, multi-channel superimposed stimuli can overcome the user's perceptual barriers. For example, initially, only light cues are used, relying on the user's visual attention; if the user ignores the light due to focusing on the screen, a second audio cues are superimposed, providing simultaneous visual and auditory cues; if the user ignores the audio due to wearing headphones or a noisy environment, a third, table lifting is further superimposed, directly intervening through the visual, auditory, and tactile channels. The synergistic effect of multimodal superposition is far greater than the simple summation of single-channel intensity. It can achieve a higher overall perception rate with a lower single-channel stimulus intensity, avoiding the interference and discomfort caused to users or surrounding people by over-enhancing a single stimulus (such as extremely bright light or extremely loud voice).

[0010] In the aforementioned posture correction method, the intensity of the corrective action is the perceived intensity of the action itself. The corrective actions include light cues, voice cues, and table lifting, with the intensity of the light cues being less than the intensity of the voice cues, which in turn is less than the intensity of the table lifting. This design, by hierarchically sorting different corrective actions according to the user's perceived intensity and activating them sequentially, achieves a progressive correction logic "from weak to strong, from passive to active." Specifically, light cues, as the weakest perceived correction action, are passive visual stimuli with the lowest degree of interruption to the user's current operation, making them suitable as initial reminders. Voice cues, as a medium perceived correction action, are auditory stimuli, and their directness and enforceability in information delivery are superior to light cues, making them suitable as secondary reminders. Table lifting, as the strongest perceived correction action, is an active intervention at the physical level, passively guiding the user to adjust their posture by changing the spatial position of the user's limb support surface, making it suitable as a final measure. This sequence aligns with the human perception priority of multi-sensory stimuli, and can achieve the corrective goal while minimizing disruption to the user's work, thereby enhancing the user experience, based on the principle of "minimum necessary intervention".

[0011] In the above-mentioned posture correction method, step S200, determining whether the correction is successful, includes: If the user's sitting posture returns to the preset standard sitting posture during or after the correction steps are performed, the correction is considered successful. If the user's sitting posture does not return to the preset standard sitting posture after the correction steps are completed, the electric height-adjustable desk will be restored to the state before the correction steps were executed, and a correction non-response event will be recorded. The correction steps will be repeated. When the cumulative number of correction non-response events reaches the preset number K, the correction is judged to have failed.

[0012] This design allows for real-time monitoring of posture recovery during the correction process, enabling users to proactively adjust their posture and terminate the correction before it is completed. This reflects the system's encouragement of user cooperation and improved response efficiency. Furthermore, the system re-verifies the posture after the correction is finished, ensuring the authenticity and reliability of the correction effect. Failure is only determined when the cumulative number of non-responding events reaches a preset K, rather than a single failure. This avoids misjudging system locks due to brief user movements (such as stretching, turning to look at a phone, picking up items, looking down at a book, drinking water, or picking up a pen), improving the robustness of the judgment. In addition, the table is restored to its original state after each correction failure, ensuring the traceability and repeatability of the system state. This ensures that subsequent correction actions always start from the same baseline state, guaranteeing the consistency of the intensity increment logic.

[0013] In the above-described posture correction method, in step S200, when the correction is successful, a first cooling phase is executed before returning to step S100. This design avoids frequent execution of correction steps, which could interfere with the user.

[0014] In the aforementioned posture correction method, in step S300, if it is determined that the number of times step S200 has been executed has not reached a preset threshold, a second cooling-off phase is executed before returning to step S200. This design avoids frequent execution of correction steps, which could interfere with the user. Simultaneously, the second cooling-off phase provides the user with a window of time to adjust their posture independently. The user may become aware of the posture problem and actively adjust it during the cooling-off period between the end of the correction action and the start of the next action, thereby achieving the correction goal with lower system intervention costs and enhancing the user's sense of control over the system.

[0015] In the above posture correction method, the cooling time of the second cooling stage increases with the number of times step S200 is executed. This design, where the cooling time of the second cooling stage is extended with each subsequent execution of step S200, is based on the following logic: when a user fails to respond to the correction action multiple times consecutively, it indicates that the user may be in a state of deep concentration (such as writing an important document or participating in a video conference) or has briefly left their seat but was misjudged by the system as still seated. In this case, the probability of the user responding to the correction action is low, and frequent and intensive correction interventions are not only unlikely to be effective but will also severely disrupt the user experience and even cause user resentment. By increasing the cooling time, the system automatically reduces the intervention frequency after multiple failures, extending the waiting time and providing the user with a more sufficient response window, while reducing the number of ineffective interventions. This protects the user experience and reduces system power consumption and the frequency of action of the execution components.

[0016] In the aforementioned posture correction method, the electric height-adjustable desk also includes an input device that can be manually operated by the user. In step S200, if an operation command from the input device is detected, the operation command is responded to first, interrupting the current correction step. This design, by setting a mechanism that prioritizes user manual operation for interruption, ensures the user's absolute control over the electric height-adjustable desk under all circumstances, conforming to the basic principles of safety design. When the system mistakenly triggers correction due to posture recognition errors, the user can immediately terminate the correction action with a simple button operation, avoiding interference with the user's normal activities. When the physical intervention of the correction action itself (such as raising the desk) makes the user feel uncomfortable, the user can also interrupt it immediately, preventing safety hazards caused by forced system intervention. This mechanism also solves the legal issue of liability attribution—any automatic system action can be stopped by the user with a single button, significantly reducing the product's usage risk and the risk of user complaints.

[0017] In the above-mentioned posture correction method, the posture correction method further includes: Step S000: Determine if there is a person sitting at the table; If it is determined that there is a person in front of the table and they are already seated, then proceed to step S100. If it is determined that there is no human being in front of the table or that there is a human being but not seated, then the system will remain in a low-power standby state.

[0018] This design, by adding a pre-seating detection step before the posture judgment step, ensures that the system only activates posture monitoring and correction functions when the user is actually seated. This avoids the system mistakenly triggering posture correction actions when no one is using the device or when the user is only briefly near the table. This effectively reduces the overall power consumption of the system and extends the lifespan of the device. It also reduces the working frequency of the posture recognition unit. Finally, it prevents unnecessary lifting and lowering of the table when no one is using it, eliminating the safety hazards and interference to those around, and improving the system's intelligence and user-friendliness.

[0019] In the above-mentioned posture correction method, in step S000, the posture recognition unit collects continuous multi-frame coordinate data of key points of the human skeleton. To determine whether a bone point is stable, if the skeletal key points appear stably in multiple consecutive frames of coordinate data, then the bone point is considered stable. To determine if a sitting action exists, extract the key points of the head skeleton. If the Y-axis coordinate of the key points of the head skeleton shows a downward trend and the amount of decline exceeds the decline threshold, then a sitting action is determined to exist. To determine whether the seating position is stable, if the fluctuation range of the Y-axis coordinate of the key points of the head skeleton is less than the stability threshold within a preset time period, then the seating position is considered stable. If the conditions of bone point stability, sitting action, and stable sitting are met, it is determined that there is a human being in front of the table and has been seated.

[0020] Existing posture recognition units rely on face capture and calibration for posture identification. However, when users are wearing masks, hats, or are facing away from their faces, or when they are looking down, their faces cannot be effectively detected, resulting in low detection accuracy. Furthermore, capturing facial images throughout the process poses a privacy risk. This new technology, on the other hand, uses continuous multi-frame coordinate data of key skeletal points for judgment. It does not rely on facial features; it accurately determines the body's posture based solely on changes in the spatial position of skeletal points such as the head and shoulders. It can still recognize postures even when wearing masks or in obscured environments, thus avoiding facial data leakage. Moreover, it determines the sitting posture using a triple-condition combination of "skeletal point stability + sitting action + sitting stability," which offers significantly higher accuracy and robustness compared to single-condition judgments. Specifically, the "bony point stability" condition eliminates false alarms caused by users briefly passing through the camera's field of view or being mistaken for human bodies by background distractions; the "sitting action present" condition identifies the specific action event of "sitting down" by tracking the descending trend of the Y-axis coordinate of key head points, effectively distinguishing between seated usage and standing / approaching states; and the "stable seating" condition ensures that the user is fully seated rather than in a transitional state of adjusting their posture, avoiding initiating posture judgment when the user's body is in an unstable state. These three progressively layered and filtered conditions minimize false triggering rates, ensuring that the system only initiates subsequent posture monitoring processes when the user is indeed in a stable seated usage state, thus improving the system's judgment accuracy and user experience.

[0021] In the above posture correction method, step S100, which involves acquiring user posture data through the posture recognition unit and determining whether the user's posture conforms to a preset standard posture, includes: The posture recognition unit collects coordinate data of multiple key skeletal points of the user's head, shoulders, and upper limbs. Based on the coordinate data of the multiple skeletal key points, a geometric model of the user's current sitting posture is established; The geometric model of the user's current sitting posture is compared with the preset standard sitting posture geometric model, which is constructed based on the same skeletal key points; Based on the comparison results, it is determined whether the user's sitting posture conforms to the preset standard sitting posture. This design, which uses a geometric model built based on skeletal key points for posture recognition, has advantages over traditional recognition methods based on image pixels or depth map features, including lower computational load, higher real-time performance, and clear posture semantics. Specifically, the skeletal key point data is two-dimensional coordinate information, with low data dimensionality and clear semantics. This allows the construction and comparison calculation of the geometric model to be completed quickly on low-power embedded chips, without relying on cloud computing or high-performance GPUs, thus reducing hardware costs. In addition, the geometric model comparison method naturally possesses scale invariance and rotation invariance (through coordinate normalization), maintaining consistent judgment accuracy in usage scenarios with different heights, body types, and camera installation angles. It eliminates the need for complex parameter calibration for each user, improving the product's versatility and ease of use.

[0022] In the above-mentioned posture correction methods, the key skeletal points of the head include the vertex of the head and the chin point; Key skeletal points in the shoulder include the left shoulder point and the right shoulder point; Key skeletal points of the upper limbs include the left elbow point, left wrist point, right elbow point, and right wrist point; A geometric model of the user's upper body is formed by connecting the top of the head to the chin, the chin to the left shoulder, the chin to the right shoulder, the left shoulder to the left elbow, the left elbow to the left wrist, the right shoulder to the right elbow, and the right elbow to the right wrist. This design, by selecting eight key skeletal points—the top of the head, the chin, both shoulders, both elbows, and both wrists—and establishing specific connections, constructs a geometric model that fully represents the user's upper body sitting posture with the most concise set of skeletal points. Compared to schemes using more key skeletal points (such as those for the lower body, including the hips, knees, and feet), this invention significantly reduces computational power requirements and data collection volume while ensuring core posture recognition functionality, making it more suitable for embedded real-time applications.

[0023] In the above-mentioned posture correction method, the standard sitting posture geometric model includes the following judgment conditions: head is centered and upright, left and right shoulders are horizontal, chin is in normal position, hands are placed naturally, and elbows are close to the torso; When the geometric model of the user's current sitting posture simultaneously meets all the judgment conditions of the standard sitting posture geometric model, the user's sitting posture is determined to conform to the preset standard sitting posture. If the geometric model of the user's current sitting posture does not meet any of the judgment conditions of the standard sitting posture geometric model, the user's sitting posture is determined to be inconsistent with the preset standard sitting posture.

[0024] By breaking down the standard sitting posture into five independent judgment conditions and setting tolerance thresholds for each condition, a multi-dimensional and refined representation and judgment of the human sitting posture is achieved. If any condition exceeds the tolerance threshold, it is judged that the user's sitting posture does not conform to the preset standard sitting posture, thereby triggering step S200 to remind the user to perform a corrective action. This ensures full coverage monitoring of various poor sitting postures. At the same time, the setting of tolerance thresholds allows the system to tolerate small and brief fluctuations in human posture, avoiding the unpleasant experience caused by frequent corrections due to excessive sensitivity. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of a user sitting in a standard sitting posture in a preferred embodiment of the present invention; Figure 2 This is a schematic diagram of a state in a preferred embodiment of the present invention when the user's sitting posture does not conform to the standard sitting posture; Figure 3 This is a schematic diagram of the preferred embodiment of the present invention when the correction action is the lifting of the tabletop; Figure 4 This is a control flowchart of the sitting posture correction method in a preferred embodiment of the present invention; In the picture, 100 is the tabletop; 200 is the extendable table leg; and 300 is the monocular camera. Detailed Implementation

[0026] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that the terms "upper," "lower," "left," "right," "longitudinal," "lateral," "inner," "outer," "vertical," "horizontal," "top," and "bottom," etc., which indicate orientation or positional relationships, are based solely on the orientation or positional relationships shown in the accompanying drawings and are used only for the convenience of describing the present invention and simplifying the description. They do not indicate or imply that the device / component referred to must have a specific orientation or be constructed and operated in a specific orientation, and therefore should not be construed as limiting the present invention. Example

[0027] Combination Figures 1 to 4 The present invention provides a posture correction method based on an electric height-adjustable desk. The electric height-adjustable desk includes a tabletop 100, retractable table legs 200, and a posture recognition unit. The number of retractable table legs 200 is at least one. The retractable table legs 200 are driven by a motor to extend and retract, thereby raising and lowering the tabletop 100. The posture recognition unit is used to collect the user's posture data.

[0028] The posture correction method in this embodiment includes: Step S100: Obtain user sitting posture data through the sitting posture recognition unit and determine whether the user sitting posture conforms to the preset standard sitting posture. If the user's sitting posture does not conform to the standard sitting posture, proceed to step S200; Step S200: Perform the correction steps, remind the user according to the preset correction actions, and determine whether the correction is successful; If the correction is successful, the execution count of step S200 is reset to zero, the electric lifting table is restored to the state before the correction step was executed, and the process returns to step S100. If the correction fails, proceed to step S300; Step S300: Determine whether the number of executions in step S200 has reached a preset threshold; If the number of times step S200 is executed does not reach the preset threshold N, return to step S200; If the number of times step S200 has been executed has reached the preset threshold N, step S200 is locked and the process waits for manual unlocking. The intensity of the corrective action increases with the number of times step S200 is executed, and the corrective action includes at least one of light prompts, voice prompts, and table lifting.

[0029] Unlike existing technologies that use a fixed-intensity, single-time prompt, this embodiment sets the intensity of the corrective action to increase with the number of times step S200 is executed. This gradually increases the intensity of the corrective reminder, thus gradually overcoming the perceptual blockage caused by the user's focus on work or study. The corrective reminder changes from being "easily ignored" to "not easily ignored," effectively guiding the user to instinctively adjust back to a standard sitting posture and improving the success rate of correction. Furthermore, the graded, progressively increasing reminder mechanism avoids excessive interference from a single high-intensity stimulus while ensuring that even when the user's self-awareness is weak, the gradually increasing reminder intensity ultimately achieves the corrective goal, effectively improving the success rate. Finally, after successful correction, the electric height-adjustable desk is restored to its state before the correction step, ensuring that the user's original desktop height usage habits are not changed. After multiple failed correction attempts in step S200, step S200 is locked and awaits manual release. This design avoids the interference of infinite loop correction on the user experience and prevents excessive wear on the motor and other actuators due to frequent ineffective actions, improving the safety and reliability of the system operation.

[0030] Step S100 involves acquiring user posture data through a posture recognition unit and determining whether the user's posture conforms to a preset standard posture, including: The posture recognition unit collects coordinate data of multiple skeletal key points of the user's head, shoulders and upper limbs; Based on the coordinate data of multiple skeletal key points, a geometric model of the user's current sitting posture is established; The geometric model of the user's current sitting posture is compared with the preset standard sitting posture geometric model, which is built based on the same skeletal key points. Based on the comparison results, it is determined whether the user's sitting posture conforms to the preset standard sitting posture. This design, which uses a geometric model built based on skeletal key points for posture recognition, has advantages over traditional recognition methods based on image pixels or depth map features, including lower computational load, higher real-time performance, and clear posture semantics. Specifically, the skeletal key point data is two-dimensional coordinate information, with low data dimensionality and clear semantics. This allows the construction of the geometric model and comparison calculations to be completed quickly on low-power embedded chips without transmitting any image data, eliminating the need for cloud computing or high-performance GPUs, reducing hardware costs, and ensuring user privacy and security. In addition, the geometric model comparison method naturally possesses scale invariance and rotation invariance (through coordinate normalization), maintaining consistent judgment accuracy in usage scenarios with different heights, body types, and camera installation angles. It eliminates the need for complex parameter calibration for each user, improving the product's versatility and ease of use.

[0031] Specifically, the posture recognition unit in this embodiment includes a monocular camera 300 and an edge AI processing module. The monocular camera 300 is mounted on the table and is responsible for capturing monocular images of the user's upper body. The edge AI processing module extracts two-dimensional pixel coordinate data of multiple skeletal key points of the human head, shoulders and upper limbs from the images, and establishes a geometric model of the user's current posture based on the coordinate data of multiple skeletal key points.

[0032] The key skeletal points for the head include the vertex and chin; for the shoulders, they include the left and right shoulder points; and for the upper limbs, they include the left elbow, left wrist, right elbow, and right wrist points. Lines connecting the vertex to the chin, chin to the left shoulder, chin to the right shoulder, left shoulder to the left elbow, left elbow to the left wrist, right shoulder to the right elbow, and right elbow to the right wrist form the geometric model of the user's upper body. This design, by selecting eight key skeletal points (vertex, chin, both shoulders, both elbows, and both wrists) and establishing specific connections, constructs a geometric model that fully represents the user's seated upper body posture using the most concise set of skeletal points. Compared to solutions that use more skeletal key points (such as lower body key points including hips, knees, and feet), this invention significantly reduces computing power requirements and data collection volume while ensuring the core posture recognition function, making it more suitable for embedded real-time application scenarios.

[0033] In this embodiment, eight skeletal key points are connected by seven lines to form an upper body skeleton topology. Conversely, if there are only eight discrete bone points, although they contain coordinate position information, there is no clear geometric topological relationship. Since the model emphasizes structure and topological relationships in the algorithm and visual context, that is, which point is connected to which point and what shape is formed, the judgment of whether the model is standard as a whole is based on the position of each point.

[0034] The standard sitting posture geometric model in this embodiment includes the following 5 judgment conditions: head centered and upright, shoulders horizontal, chin in normal position, hands naturally placed, and elbows close to the torso; When the geometric model of the user's current sitting posture simultaneously meets all the judgment conditions of the standard sitting posture geometric model, the user's sitting posture is determined to conform to the preset standard sitting posture. If the geometric model of the user's current sitting posture does not meet any of the judgment conditions of the standard sitting posture geometric model, the user's sitting posture is determined to be inconsistent with the preset standard sitting posture.

[0035] The following are the criteria for determining the five conditions: 1. Head centered and upright: The vertex H of the head is located above the midpoint Ms of the line connecting the two shoulder points, and satisfies |Hx-Msx|≤ δh, where Hx is the x-coordinate of the vertex of the head in the image plane, Msx is the x-coordinate of the midpoint Ms of the two shoulders, and δh is the horizontal offset tolerance threshold under the standard sitting posture, usually taken as 15 pixels. If the above conditions are met, it can be determined that the head is centered and upright.

[0036] 2. Horizontal shoulders: The left shoulder point LS and the right shoulder point RS have the same height, i.e., |LSy-RSy|≤δs, and the shoulder line slope ks=|RSy-LSy| / |RSx-LSx|,ks≤ε1; where LSy and RSy are the ordinates of the left and right shoulder points, respectively, LSx and RSx are the abscissas of the left and right shoulder points, respectively, δs is the tolerance threshold for vertical shoulder offset under standard sitting posture, usually taken as 10 pixels, and ε1 is the tolerance threshold for shoulder line slope under standard sitting posture, usually taken as 0.3. If the above conditions are met, the horizontal shoulders can be determined.

[0037] 3. Normal chin position: The difference between the ordinate of the chin point C and the ordinate of the midpoint of the shoulders Ms is within a reasonable range, i.e., |Cy-Msy|≤δc, where Cy is the ordinate of the chin point, Msy is the ordinate of the midpoint of the shoulders, and δc is the vertical offset tolerance threshold between the chin point and the midpoint of the shoulders in a standard sitting posture, usually taken as 20 pixels. If the above conditions are met, the chin position can be judged to be normal.

[0038] 4. Hands are placed naturally: The vertical coordinate of the wrist is not lower than the vertical coordinate of the elbow on the same side, that is, LWy≥LEy and RWy≥REy, where LWy and LEy are the vertical coordinates of the left wrist and left elbow, respectively, and RWy and REy are the vertical coordinates of the right wrist and right elbow, respectively. If the above conditions are met, it can be determined that the hands are placed naturally.

[0039] 5. Elbows close to the torso: The horizontal distance between the elbow and the shoulder on the same side is within the normal range, i.e., |LEx-LSx|≤δe and |REx-RSx|≤δe, where LEx and LSx are the x-coordinates of the left elbow and left shoulder, respectively, REx and RSx are the x-coordinates of the right elbow and right shoulder, respectively, and δe is the tolerance threshold for the horizontal offset between the elbow and the shoulder on the same side under the standard sitting posture, which is usually taken as 20 pixels.

[0040] By breaking down the standard sitting posture into five independent judgment conditions and setting tolerance thresholds for each condition, a multi-dimensional and refined representation and judgment of the human sitting posture is achieved. If any condition exceeds the tolerance threshold, it is judged that the user's sitting posture does not conform to the preset standard sitting posture, thereby triggering step S200 to remind the user to perform a corrective action. This ensures full coverage monitoring of various poor sitting postures. At the same time, the setting of tolerance thresholds allows the system to tolerate small and brief fluctuations in human posture, avoiding the unpleasant experience caused by frequent corrections due to excessive sensitivity.

[0041] It should be noted that when the posture recognition unit determines that the user's posture does not conform to the standard posture, it immediately starts the abnormal posture timer tl. If the duration of the abnormal posture tl is greater than or equal to Tl (default 30-60 seconds), it is considered a valid abnormal posture event, meaning the user's posture does not conform to the standard posture. If the posture returns to the standard posture during the timer period, the timer tl is reset to 0, and the user's posture is determined to conform to the standard posture. The purpose of this continuous confirmation mechanism is to prevent users' brief normal behaviors such as looking down to read a book, picking up a pen, or drinking water from being mistakenly identified as abnormal postures requiring correction.

[0042] Additionally, users can manually adjust the desk to the most suitable height for normal sitting posture reading / writing / office work using the manual controller, based on their own height and habits. The system automatically records the current motor position encoder reading as the original comfortable height H0. All subsequent correction actions are based on H0 as the reference surface. H0 can be recalibrated by the user at any time (after manual adjustment, press and hold a button to confirm and overwrite the old value). In this embodiment, in step S200, if the correction is successful, the execution count of step S200 is reset to zero, restoring the electric height-adjustable desk to its state before the correction step. This means that after the desk is raised to remind the user that the correction is successful, the desk needs to be lowered back to the original comfortable height H0 to ensure that the user's original desktop height usage habits are not changed.

[0043] Furthermore, in step S200, determining whether the correction was successful includes: If the user's sitting posture returns to the preset standard sitting posture during or after the correction steps are performed, the correction is considered successful. If the user's posture does not return to the preset standard posture after the correction steps are completed, the electric height-adjustable desk is restored to its state before the correction steps were executed, and a correction non-response event is recorded. The correction steps are repeated, and when the cumulative number of correction non-response events reaches a preset number K, the correction is deemed a failure. This design monitors the posture recovery status in real time during the correction steps, allowing users to actively adjust their posture and terminate the correction before the correction action is completed, reflecting the system's encouragement of user cooperation and improved response efficiency. Simultaneously, the posture status is reviewed and judged after the correction action is completed, ensuring the authenticity and reliability of the correction effect. Failure is only judged when the cumulative number of correction non-response events reaches a preset number K, rather than judging a single failure, avoiding the misjudgment of correction failure due to brief physical activities of the user (such as stretching, turning their head to look at their phone, picking up items, looking down at a book, drinking water, picking up a pen) and the resulting system lock, thus improving the robustness of the judgment. In addition, the tabletop is restored to its original state after each correction failure, ensuring the traceability and repeatability of the system state. This ensures that subsequent correction actions always start from the same baseline state, guaranteeing the consistency of the intensity increment logic.

[0044] It should be noted that when performing the K correction steps in step S200, the correction action types and correction action intensities are consistent across the K correction steps.

[0045] The following explanation uses the table lifting motion as an example to illustrate the correction steps. The specific correction process is as follows: 1. Record the current position: Read the current desktop height H and determine its relationship with H0. If H = H0 (normal situation), then the target height for correction H1 = H0 + ΔH, where ΔH is 3 cm to 8 cm. If H ≠ H0 (e.g., the previous correction has not been fully reset), first return the desktop to H0 before performing the correction.

[0046] 2. Slow Rise: The motor-driven tabletop rises from H0 at a speed of Vup (1.0~2.5 cm / s), moving away from directly above the user's body. This speed is designed to be gradual, gentle, and without abrupt stops—allowing the user to continue writing or reading without noticeable interruption, but they will gradually perceive the tabletop height deviating from a comfortable position, creating a natural sense of discomfort to serve as a reminder to the user.

[0047] 3. Real-time Posture Monitoring (During Ascent): During the table's ascent, the posture recognition unit continuously operates, checking frame by frame whether the user's current posture geometric model simultaneously meets all the criteria for the standard posture geometric model. If all criteria for the standard posture geometric model are met consecutively for a time T2 (default 2-5 seconds), the correction is considered successful. Once the correction is successful, the system immediately stops the table's ascent, records the current stopping height Hs (H0 < Hs < H0 + ΔH), and then controls the motor to drive the table to descend smoothly at a preset speed Vdown, returning to the original comfortable height H0. The system also resets the error timer and the execution count of step S200. If the geometric model of the user's current sitting posture does not meet any of the judgment conditions of the standard sitting posture geometric model, and the duration of the abnormality exceeds Tl, it is determined that the user's sitting posture does not conform to the preset standard sitting posture. If the table has continued to reach the preset maximum lifting height H1 and still does not conform to the standard sitting posture, the motor drives the table to descend smoothly at the preset speed Vdown, returning to the original comfortable height H0, and records a correction non-response event. The correction steps are repeated. When the correction non-response event accumulates to a preset number K, the correction is determined to have failed.

[0048] In this embodiment, the intensity of the corrective action increases with the number of times step S200 is executed. This means that the lifting stroke and / or lifting speed of the table increases with the number of times step S200 is executed. Specifically, when the user is alerted only by the corrective action of lifting the table, if the first correction fails and step S200 is executed for the second time, the table lifting speed Vup in the correction step will be adjusted to 1.5–3.0 cm / s, and / or the target lifting height ΔH will be increased by 10%–20% to enhance the significance of the discomfort. If the correction still fails after the second execution of step S200, the table lifting speed Vup in the correction step will be further increased to 2.5–4.0 cm / s in the third execution of step S200, and the target lifting height ΔH will be increased by 15%–30% on the original basis. This continues until the number of times step S200 is executed reaches a preset threshold, at which point step S200 is locked, awaiting manual unlocking.

[0049] It should be noted that the desktop elevation in this embodiment is only a corrective measure, not an end in itself. After correction is complete or the time limit is exceeded, the desktop returns to its original comfortable height H0. This differs from the traditional forced correction scheme of "abnormal posture → raising the desktop → remaining at a high position." This embodiment utilizes only the "discomfort at a height" as a subconscious corrective driving force, rather than forcibly changing the desktop height to compel the user to change their posture. It employs a closed-loop correction strategy of "slow elevation – discomfort perception – self-correction – automatic reset." The corrective action of the electric height-adjustable desk does not aim to remain at the corrected height, but rather creates a slight discomfort that does not affect writing / reading by slowly and slightly raising the desktop height. Utilizing the human body's natural sensitivity to height changes, it induces the user to instinctively adjust back to a standard sitting posture. Once the posture returns to standard, the desktop immediately and automatically returns to the user-set original comfortable height H0. If the posture has not returned to normal by the end of the correction period, the desktop also returns to its original height to avoid continuous interference.

[0050] In addition, the electric height-adjustable desk in this embodiment also includes a light prompt module and a voice prompt module. The light prompt module provides light prompts, and the voice prompt module provides voice prompts. When only the light prompt is used to remind the user, the intensity of the correction action increases with the number of times step S200 is executed. This means that the output intensity of the light increases with the number of times step S200 is executed. That is, the light brightness is soft during the first prompt. If the user still does not adjust, the light output intensity in the correction step will be increased to medium brightness during the second execution of step S200 to enhance the visual stimulation to the user. If the correction is still determined to have failed after the second execution of step S200, the light output intensity in the correction step will be further increased to high brightness during the third execution of step S200 to strengthen the visual stimulation. This continues until the number of times step S200 is executed reaches a preset threshold, at which point step S200 is locked, awaiting manual unlocking.

[0051] In addition, when the user is reminded only by the corrective action of voice prompt, the intensity of the corrective action increases with the number of times step S200 is executed. This means that the output intensity of the voice increases with the number of times step S200 is executed. That is, the voice volume is low and the tone is flat when the prompt is given for the first time. If the user does not respond, the volume of the voice output intensity in the corrective step is increased by 10% to 20% when step S200 is executed for the second time, so as to enhance the auditory stimulation to the user. If the correction is still determined to be unsuccessful after the second execution of step S200, the volume in the corrective step is increased by 30% to 40% when step S200 is executed for the third time, so as to strengthen the auditory stimulation. This continues until the number of times step S200 is executed reaches the preset threshold, at which point step S200 is locked and the user waits for manual unlocking.

[0052] All three reminder methods are performed under the same type of corrective action. By continuously increasing the output intensity of a single physical quantity, stimulation can be applied in a gradual manner that is perceptible to the user, so as to gradually improve the user's perception of the corrective action reminder, thereby effectively guiding the user to adjust to the standard sitting posture. Specifically, regarding the increased lifting range of the tabletop, gradually raising the tabletop forces the user's elbows and forearms to be raised progressively, causing discomfort in the arms and guiding the user to adjust to a standard sitting posture. Regarding the increased lifting speed, increasing the lifting speed allows the tabletop to complete the lifting action in a shorter time, producing a more noticeable displacement sensation, thus enhancing the user's tactile feedback and making it easier for them to perceive changes in the tabletop, guiding them to adjust to a standard sitting posture. Regarding lighting cues, gradually increasing the light output intensity changes the light from soft to bright, creating an increasingly strong visual signal in the user's peripheral vision, gradually breaking through the visual barrier of a focused state, making the user feel glared, and guiding them to adjust to a standard sitting posture. Regarding voice cues, gradually increasing the voice output intensity changes the cues from soft to piercing, gradually penetrating the auditory barrier of a focused state, and guiding the user to adjust to a standard sitting posture.

[0053] Alternatively, in other embodiments of the present invention, the intensity of the corrective action is the number of combinations of corrective actions. As the number of executions of step S200 increases, more types of corrective actions are superimposed. This design, by superimposing multiple types of corrective actions to achieve incremental intensity, utilizes the synergistic effect of multimodal stimulation, which can significantly enhance the user's perception probability. Specifically, visual stimulation (light), auditory stimulation (voice), and somatosensory stimulation (table lifting) act on different human sensory channels. When a single-channel stimulus is ignored by the user, multi-channel superimposed stimulation can overcome the user's perceptual barrier. For example, initially, only light cues are used, relying on the user's visual attention; if the user ignores the light due to looking at the screen, then in the second execution of step S200, voice cues can be superimposed, providing simultaneous visual and auditory cues; if the user ignores the voice due to wearing headphones or a noisy environment, then in the third execution of step S200, table lifting is further superimposed, directly intervening through the visual, auditory, and tactile channels. The synergistic effect of multimodal superposition is far greater than the simple summation of single-channel intensity. It can achieve a higher overall perception rate with a lower single-channel stimulus intensity, avoiding the interference and discomfort caused to users or surrounding people by over-enhancing a single stimulus (such as extremely bright light or extremely loud voice).

[0054] Alternatively, in other embodiments of the present invention, the intensity of the corrective action is the perceived intensity of the corrective action itself. The corrective actions include light cues, voice cues, and table lifting, with the intensity of the light cues < the intensity of the voice cues < the intensity of the table lifting. This design, by hierarchically sorting different corrective actions according to the user's perceived intensity and enabling them sequentially, achieves a progressive corrective logic "from weak to strong, from passive to active." Specifically, the light cues, as the weakest perceived corrective action, are passive visual stimuli with the lowest degree of interruption to the user's current operation, making them suitable as initial reminders; the voice cues, as a medium perceived corrective action, are auditory stimuli, and their directness and enforceability of information transmission are superior to light, making them suitable as secondary reminders; the table lifting, as the strongest perceived corrective action, is an active intervention at the physical level, passively guiding the user to adjust their sitting posture by changing the spatial position of the user's limb support surface, making it suitable as a final means. This sequence aligns with the human perception priority of multi-sensory stimuli, and can achieve the corrective goal while minimizing disruption to the user's work, thereby enhancing the user experience, based on the principle of "minimum necessary intervention".

[0055] Furthermore, in step S200 of this embodiment, when the correction is successful, the first cooling stage is executed before returning to step S100. This design avoids frequent execution of correction steps (especially frequent driving of the tabletop height adjustment) from interfering with the user. The duration of the first cooling stage can be 60 to 120 seconds.

[0056] In step S300, if the number of times step S200 has been executed has not reached a preset threshold, the system executes a second cooling-off phase and then returns to step S200. This design avoids frequent execution of correction steps (especially frequent table height adjustments) from interfering with the user. Simultaneously, the second cooling-off phase provides the user with a window of time to adjust their posture independently. The user may become aware of the posture problem and actively adjust it during this cooling-off period, after the correction action ends and before the next action begins, thus achieving the correction goal with lower system intervention costs and enhancing the user's sense of control over the system. The duration of the second cooling-off phase can be 120 to 180 seconds.

[0057] Preferably, the cooling duration of the second cooling stage increases with the number of times step S200 is executed. This design, where the cooling duration of the second cooling stage gradually increases with the number of times step S200 is executed, stems from the following logic: when a user fails to respond to the correction action multiple times consecutively, it indicates that the user may be in a state of deep concentration (such as writing an important document or participating in a video conference) or has briefly left their seat but was misjudged by the system as still seated. In this situation, the probability of the user responding to the correction action is low. Frequent and intensive correction interventions are not only unlikely to be effective but will also severely disrupt the user experience and even cause user resentment. By increasing the cooling duration, the system automatically reduces the intervention frequency after multiple failures, extending the waiting time and providing the user with a more sufficient response window, while reducing the number of ineffective interventions. This protects the user experience and reduces system power consumption and the frequency of actions of the execution components.

[0058] Secondly, the electric height-adjustable desk in this embodiment also includes an input device that can be manually operated by the user. In step S200, if an operation command from the input device is detected, the operation command is responded to first, interrupting the current correction step. This design, by setting a mechanism that prioritizes user manual operation for interruption, ensures the user's absolute control over the electric height-adjustable desk under any circumstances, conforming to the basic principles of safety design. When the system mistakenly triggers correction due to posture recognition errors, the user can immediately terminate the correction action with a simple button operation, avoiding interference with the user's normal activities. When the physical intervention of the correction action itself (such as raising the tabletop) makes the user feel uncomfortable, the user can also interrupt it immediately, preventing safety hazards caused by forced system intervention. This mechanism also solves the legal issue of liability attribution—any automatic system action can be stopped by the user with a single button, significantly reducing the product's usage risk and the risk of user complaints.

[0059] For example, when reminding users by raising the tabletop, if the system detects that the user manually presses any button on the height adjustment controller during the adjustment process, the adjustment action will stop immediately. The system determines that the user has realized the change in tabletop height, and the adjustment is considered complete. Then, the system can enter the adjustment cooling period.

[0060] Finally, the posture correction method in this embodiment also includes: Step S000: Determine if there is a person sitting at the table; If it is determined that there is a person in front of the table and they are already seated, then proceed to step S100. If it is determined that there is no human being in front of the table or that there is a human being but not seated, then the system will remain in a low-power standby state.

[0061] This design, by adding a pre-seating detection step before the posture judgment step, ensures that the system only activates posture monitoring and correction functions when the user is actually seated. This avoids the system mistakenly triggering posture correction actions when no one is using the device or when the user is only briefly near the table. This effectively reduces the overall power consumption of the system and extends the lifespan of the device. It also reduces the working frequency of the posture recognition unit. Finally, it prevents unnecessary lifting and lowering of the table when no one is using it, eliminating the safety hazards and interference to those around, and improving the system's intelligence and user-friendliness.

[0062] In step S000, the posture recognition unit collects continuous multi-frame coordinate data of key points of the human skeleton. To determine whether a bone point is stable, if the skeletal key points appear stably in multiple consecutive frames of coordinate data, then the bone point is considered stable. To determine if a sitting action exists, extract the key points of the head skeleton. If the Y-axis coordinate of the key points of the head skeleton shows a downward trend and the amount of decline exceeds the decline threshold, then a sitting action is determined to exist. To determine whether the seating position is stable, if the fluctuation range of the Y-axis coordinate of the key points of the head skeleton is less than the stability threshold within a preset time period, then the seating position is considered stable. If the conditions of bone point stability, sitting action, and stable sitting are met, it is determined that there is a human being in front of the table and has been seated.

[0063] Existing posture recognition units achieve posture recognition through face capture and calibration. However, when users are wearing masks, hats, or looking at their profile, or with their heads down, faces cannot be effectively detected, resulting in low detection accuracy. Furthermore, capturing facial images throughout the process poses a privacy risk. In contrast, this embodiment judges posture by collecting continuous multi-frame coordinate data of key points in the human skeleton. It does not rely on facial features; it can accurately determine the human body's state based solely on changes in the spatial position of bone points such as the head and shoulders. It can still recognize postures even when wearing masks or having obscured areas, thus avoiding facial data leakage. In addition, the combination of three conditions—"bone point stability + sitting action + sitting stability"—to determine the sitting state has significantly higher accuracy and robustness compared to single-condition judgment. Specifically, the "bony point stability" condition eliminates false alarms caused by users briefly passing through the camera's field of view or being mistaken for human bodies by background distractions; the "sitting action present" condition identifies the specific action event of "sitting down" by tracking the downward trend of the Y-axis coordinate of key head points, effectively distinguishing between seated usage and non-sitting behaviors such as standing, bending over to pick up an object, or brief pauses; the "stable seating" condition ensures that the user is fully seated rather than in a transitional state of adjusting their posture, avoiding initiating posture judgment when the user's body is in an unstable state. These three conditions are progressively enhanced and filtered step by step, minimizing the false trigger rate and ensuring that the system only initiates the subsequent posture monitoring process when the user is indeed in a stable seated usage state, thus improving the system's judgment accuracy and user experience.

[0064] The following will explain whether the skeletal points are stable, whether there is a sitting motion, and how to determine whether the sitting motion is stable: Determination of bone point stability: N frames are continuously collected (N≥15, frame rate≥15fps, corresponding to an observation window of ≥1 second). For the bone point set Pt= {p1,p2, ...,pk} detected in each frame t, the following is performed: (1) Calculate the detection confidence Ci of each bone point in N frames. If the average confidence Chead of the head bone points is ≥ Tconf (Tconf is 0.6 ~ 0.8) and the average confidence Cshoulder of the shoulder bone points is ≥ Tconf, then the bone points are determined to be stable. If the confidence of the head bone points is lower than the threshold for M consecutive frames (M≥5), then the bone points are determined to be lost and unstable, and the stability counter is reset.

[0065] Determining whether a sitting action exists: Take the sequence of head bone point y-coordinates HY= {Hy1,Hy2, ..., HyN} within N frames of the sliding window. If the head bone key points show a downward trend in the Y-axis coordinate (i.e., Hy1<Hy2<……<HyN) and the amount of decline exceeds the decline threshold (i.e., HyN-Hy1 is greater than Dmin), where Dmin is the minimum effective number of declining pixels used to distinguish minute body swaying, then a sitting action is determined to exist. Is the sitting position stable? After the descent action is completed, if the fluctuation amplitude Δy of the skeletal key point y of the head is less than ε (ε is the stability threshold) within S consecutive frames (S≥10), it indicates that the target has been in a stable sitting position, rather than just bending over to pick up an object. This indicates that the sitting position is stable.

[0066] It should be noted that when the system collects multiple frames of coordinate data of key points of the human skeleton based on the posture recognition unit, and determines that there is no human body in front of the table or that there is a human body but it is not seated, the system enters a low-power standby state—that is, it does not enter the posture recognition process; it does not calculate the geometric parameters of the key points of the skeleton; it does not trigger abnormal posture reminders (sound / light / table lifting); and it does not link the lifting table motor control.

[0067] In addition to the preferred embodiments described above, the present invention may have other embodiments. Those skilled in the art can make various changes and modifications based on the present invention, and all such changes and modifications should fall within the scope defined in the claims of the present invention, as long as they do not depart from the spirit of the present invention.

Claims

1. A method for correcting sitting posture based on an electrically adjustable desk, characterized in that, The electric height-adjustable desk includes: Tabletop; At least one retractable table leg driven by a motor is used to raise and lower the tabletop; The posture recognition unit is used to collect the user's posture data; The posture correction method includes: Step S100: Obtain user sitting posture data through the sitting posture recognition unit and determine whether the user sitting posture conforms to the preset standard sitting posture. If the user's sitting posture does not conform to the standard sitting posture, proceed to step S200; Step S200: Perform the correction steps, remind the user according to the preset correction actions, and determine whether the correction is successful; If the correction is successful, the execution count of step S200 is reset to zero, the electric lifting table is restored to the state before the correction step was executed, and the process returns to step S100. If the correction fails, proceed to step S300; Step S300: Determine whether the number of executions in step S200 has reached a preset threshold. If the number of times step S200 is executed does not reach the preset threshold, return to step S200; If the number of times step S200 has been executed has reached a preset threshold, step S200 is locked and the process waits for manual unlocking. The intensity of the corrective action increases with the number of times step S200 is executed, and the corrective action includes at least one of light prompts, voice prompts, and table lifting.

2. The method of claim 1, wherein the electric lifting table is a standing desk. The intensity of the corrective action is the output intensity of a single corrective action. Increasing the output intensity of a single corrective action includes at least one of the following: Increase the lifting stroke and / or lifting speed of the tabletop; Increase the output intensity of the light; Enhance the output strength of the speech. 3.The method of claim 1, wherein the method further comprises: determining a distance between the user and the table based on the height of the table and the height of the user; and adjusting the height of the table based on the distance. The intensity of the corrective action is the number of combinations of corrective actions. As the number of execution steps S200 increases, more types of corrective actions are superimposed.

4. The method of claim 1, wherein the electric lifting table is a standing desk. The intensity of the corrective action is the perceived intensity of the corrective action itself. The corrective action includes light cues, voice cues, and table lifting. The intensity of the light cues is less than the intensity of the voice cues, which is less than the intensity of the table lifting.

5. The posture correction method based on an electrically adjustable desk as described in claim 1, characterized in that, In step S200, determining whether the correction was successful includes: If the user's sitting posture returns to the preset standard sitting posture during or after the correction steps are performed, the correction is considered successful. If the user's sitting posture does not return to the preset standard sitting posture after the correction steps are completed, the electric height-adjustable desk will be restored to the state before the correction steps were executed, and a correction non-response event will be recorded. The correction steps will be repeated. When the cumulative number of correction non-response events reaches the preset number K, the correction is judged to have failed.

6. The posture correction method based on an electrically adjustable desk as described in claim 1, characterized in that, In step S200, when the correction is successful, the first cooling stage is performed before returning to step S100.

7. The posture correction method based on an electrically adjustable desk as described in claim 1, characterized in that, In step S300, if it is determined that the number of times step S200 has been executed has not reached the preset threshold, the second cooling stage is executed and then the process returns to step S200.

8. The posture correction method based on an electrically adjustable desk as described in claim 7, characterized in that, The cooling duration of the second cooling stage increases with the number of times step S200 is executed.

9. The posture correction method based on an electrically adjustable desk as described in claim 1, characterized in that, The electric height-adjustable desk also includes an input device that can be manually operated by the user; in step S200, if an operation command from the input device is detected, the operation command is responded to first, and the current correction step is interrupted.

10. The posture correction method based on an electrically adjustable desk as described in claim 1, characterized in that, The posture correction method also includes: Step S000: Determine if there is a person sitting at the table; If it is determined that there is a person in front of the table and they are already seated, then proceed to step S100. If it is determined that there is no human being in front of the table or that there is a human being but not seated, then the system will remain in a low-power standby state.

11. The posture correction method based on an electrically adjustable desk as described in claim 10, characterized in that, In step S000, the sitting posture recognition unit collects continuous multi-frame coordinate data of key points of the human skeleton. To determine whether a bone point is stable, if the skeletal key points appear stably in multiple consecutive frames of coordinate data, then the bone point is considered stable. To determine if a sitting action exists, extract the key points of the head skeleton. If the Y-axis coordinate of the key points of the head skeleton shows a downward trend and the amount of decline exceeds the decline threshold, then a sitting action is determined to exist. To determine whether the seating position is stable, if the fluctuation range of the Y-axis coordinate of the key points of the head skeleton is less than the stability threshold within a preset time period, then the seating position is considered stable. If the conditions of bone point stability, sitting action, and stable sitting are met, it is determined that there is a human being in front of the table and has been seated.

12. The posture correction method based on an electrically adjustable desk as described in claim 1, characterized in that, Step S100, which involves acquiring user posture data through the posture recognition unit and determining whether the user's posture conforms to a preset standard posture, includes: The posture recognition unit collects coordinate data of multiple key skeletal points of the user's head, shoulders, and upper limbs. Based on the coordinate data of the multiple skeletal key points, a geometric model of the user's current sitting posture is established; The geometric model of the user's current sitting posture is compared with the preset standard sitting posture geometric model, which is constructed based on the same skeletal key points; Based on the comparison results, it is determined whether the user's sitting posture conforms to the preset standard sitting posture.

13. The posture correction method based on an electrically adjustable desk as described in claim 12, characterized in that, Key skeletal points of the head include the vertex and the chin. Key skeletal points in the shoulder include the left shoulder point and the right shoulder point; Key skeletal points of the upper limbs include the left elbow point, left wrist point, right elbow point, and right wrist point; The geometric model of the user's upper body is formed by connecting the top of the head to the chin, the chin to the left shoulder, the chin to the right shoulder, the left shoulder to the left elbow, the left elbow to the left wrist, the right shoulder to the right elbow, and the right elbow to the right wrist.

14. The posture correction method based on an electrically adjustable desk as described in claim 13, characterized in that, The standard sitting posture geometric model includes the following criteria: head centered and upright, shoulders level, chin in normal position, hands naturally placed, and elbows close to the torso; When the geometric model of the user's current sitting posture simultaneously meets all the judgment conditions of the standard sitting posture geometric model, the user's sitting posture is determined to conform to the preset standard sitting posture. If the geometric model of the user's current sitting posture does not meet any of the judgment conditions of the standard sitting posture geometric model, the user's sitting posture is determined to be inconsistent with the preset standard sitting posture.

Images