A student posture correction system and method based on double bed

Abstract

The present invention relates to the field of intelligent posture processing technology, specifically a double-bed-based student posture correction system and method. The system comprises an image acquisition unit disposed within a classroom and configured to capture student posture image information, and further comprises an information matching unit, a posture detection unit, a control unit, and a correction unit, all connected to the image acquisition unit by signal signals. The correction unit is disposed within the double bed and corrects the student's posture upon receiving control information. Feature fusion using a recognition model within the posture detection unit accurately identifies the student's posture, and corresponding control information is generated based on the posture to control the correction process of the correction unit, thereby preventing certain posture disorders.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent posture processing, and in particular to a student posture correction system and method based on a double bed. Background Art

[0002] It's well known that people frequently change body positions during sleep, meaning that sleep is a dynamic process. Most people seek a comfortable sleeping position before falling asleep. According to statistics, after the sleep movement system activates, 65% of people tend to sleep on their side, 30% on their back, and 5% on their stomach.

[0003] Poor sleeping posture is one of the main causes of cervical spondylosis. During sleep, muscles and ligaments are in a non-protective state of relaxation. An inappropriate height can lead to cervical spine damage due to the lack of protective function of the cervical muscles. For teachers and students, prolonged incorrect sitting posture in the classroom can also cause significant damage to the cervical spine.

[0004] Therefore, there is an urgent need for a system that monitors and corrects the posture of students in school. Summary of the Invention

[0005] The present invention aims to provide a student posture correction system and method based on a double bed, which is used to generate a correction model according to the students' sitting habits in the classroom and perform posture correction through a correction unit on the double bed in the dormitory.

[0006] The present invention is achieved through the following technical solutions:

[0007] A student posture correction system based on a double bed comprises an image acquisition unit arranged in a classroom and used to collect body image information of students in the classroom, and further comprises: an information matching unit connected to the image acquisition unit by signal, wherein the information matching unit has the student's personal information recorded therein and is used to match the personal information with the body image information; a posture detection unit equipped with a recognition model, wherein the recognition model uses a VGGNet network to perform feature fusion to identify the student's body posture; a control unit, wherein the control unit generates control information based on the body posture; and a correction unit, wherein the correction unit is arranged in the double bed and corrects the student's posture after receiving the control information.

[0008] Based on the above, an image acquisition unit installed in the classroom detects and analyzes students' sitting postures during class. The primary information collected is the students' body posture images. Furthermore, the information matching unit stores the students' personal information, including gender, age, height, weight, and health information. It also stores images of standard sitting and standing postures, and binds the body posture images collected by the image acquisition unit to the personal information. Secondly, the body posture detection unit performs feature recognition and fusion detection on the body images containing personal information to identify body postures. It should be noted that due to the complexity of human motion states, the statistical recognition process categorizes postures into the following categories based on similarity: upright sitting (i.e., standard sitting posture), left and right head tilt, left and right body tilt, squinting, uneven shoulders, forward neck tilt, and chin-supporting postures. Finally, after receiving the body posture, the control unit generates control information based on the student's corresponding body posture and the corresponding matching time series. For example, if a student has high and low shoulders for a long time during the day's sitting posture, the corresponding correction unit will be controlled to make the shoulders tend to the standard posture. In short, the control unit controls the correction unit by generating control information so that the student's body posture tends to the standard posture during sleep at night, or even the opposite of the non-standard posture during the day. Feature fusion through the recognition model in the body posture detection unit can accurately identify the student's body posture, and generate corresponding control information based on the body posture to control the correction process of the correction unit, which can play a corresponding preventive role for some body posture diseases. Since smart teaching has been vigorously developed in existing classroom teaching methods, cameras are generally installed in classrooms, that is, the image acquisition unit can be a traditional camera, or a high-definition dome camera or a gun camera, and is connected to the information matching unit and the body posture detection unit in a wireless or limited manner.

[0009] Furthermore, the posture detection unit includes signal connections: a node detection module, which realizes the detection of key nodes of the student's skeleton through a sitting posture estimation model; a feature extraction module, which calculates the student's sitting posture characteristics through key node information; a posture recognition module, wherein the recognition model is installed in the posture recognition module, and after the sitting posture characteristics are input into the recognition model, the recognition model identifies and outputs the student's posture after feature fusion.

[0010] Furthermore, the VGGNet network includes 6 groups of convolution modules, and adjacent convolution layers with the same feature map size in the VGGNet network structure constitute a residual module, which contains a total of 6 residual modules. A BN layer is added before each activation function in the residual module, and the activation function is LeakyRelu.

[0011] Furthermore, the sitting posture estimation model detects the key nodes of the skeleton and calculates the feature vectors according to the key nodes, and forms a skeleton image according to the key nodes. The feature extraction module calculates the student's sitting posture features according to the feature vectors in the skeleton image.

[0012] Furthermore, the control unit is also signal-connected to a reminder unit, in which a standard posture is provided. When the posture received by the reminder unit deviates from the standard posture by a threshold and lasts for a set time, the student's sitting posture is reminded.

[0013] A method for correcting student posture based on a double bed comprises the following steps: Step 1, information acquisition, in which the personal information of students in the classroom is input into the information matching unit, and the posture image information acquired by the image acquisition unit is bound to the personal information to obtain posture image information with the personal information; Step 2, posture detection, in which after Step 1 is completed, the posture detection unit performs feature extraction and fusion on the posture image information to obtain the posture of each student in the classroom; Step 3, sleep correction, in which the control unit generates control information based on the statistics of the posture, and controls the correction unit to correct the students' posture during sleep time.

[0014] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0015] 1. The present invention can accurately identify students' postures by integrating features of the recognition model in the posture detection unit, and generate corresponding control information based on the posture to control the correction process of the correction unit, thereby playing a corresponding preventive role for some posture diseases.

[0016] 2. In the sitting posture estimation model of the present invention, each identified skeleton is normalized by the maximum bounding box side length, and the skeleton center is aligned to [0.5, 0.5]. This maintains consistency in the skeleton structure features and is independent of the student's individual height or the distance and deviation direction of the image acquisition unit, making it highly adaptable.

[0017] 3. The present invention not only reduces the computing resources required for the model and achieves a balance between the best training-based accuracy and model size, but also effectively improves the inference speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0019] Figure 1 Schematic diagram of the structure of the system of the present invention;

[0020] Figure 2Schematic diagram of the process of the present invention;

[0021] Figure 3 It is a structural schematic diagram of the bed body of the present invention;

[0022] Figure 4 It is a structural schematic diagram of the plate of the present invention;

[0023] Figure 5 This is a schematic diagram of the structure of the present invention in which the movable main plate is folded upward and the movable side plates are folded toward the center;

[0024] Figure 6 Schematic diagram of the folding structure of the movable bottom plate;

[0025] Figure 7 Schematic diagram of the structure of the panel when viewed from above;

[0026] Figure 8 It is a schematic diagram of the local structure of the panel;

[0027] Figure 9 It is a structural diagram of the opening and closing parts on the movable side panel;

[0028] Figure 10 This is a structural diagram of the correction part.

[0029] Markings and corresponding parts names in the accompanying drawings:

[0030] 1-bed, 2-plate, 3-power part, 4-correction part, 5-through hole, 6-opening and closing part,

[0031] 21-frame, 22-movable back panel, 23-movable bottom panel, 24-base,

[0032] 221-movable side plate, 222-movable main plate, 223-slide, 224-first connecting rod, 225-first slider, 226-first worm, 227-first motor, 228-second motor, 229-second worm, 230-second slider, 220-second connecting rod,

[0033] 231-first bottom plate, 232-second bottom plate, 233-controllable gas spring, 234-fixed plate,

[0034] 41-correction seat, 42-correction motor, 43-cylinder, 44-correction head, 45-ball, 46-correction rod,

[0035] 61-rotating disk, 62-opening and closing plate, 63-opening and closing slot, 64-limiting pin, 65-limiting slot. DETAILED DESCRIPTION

[0036] To make the objectives, technical solutions, and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the examples and accompanying drawings. The exemplary embodiments of the present invention and their descriptions are only used to explain the present invention and are not intended to limit the present invention. It should be noted that the present invention is already in the actual development and use stage.

[0037] Example 1:

[0038] As attached Figure 1 As shown, a student posture correction system based on a double bed includes an image acquisition unit arranged in a classroom and used to collect body image information of students in the classroom, and also includes: an information matching unit connected to the image acquisition unit by signal, the information matching unit having the student's personal information recorded therein and used to match the personal information with the body image information; a posture detection unit, the posture detection unit is equipped with a recognition model, and the recognition model uses the VGGNet network to perform feature fusion to identify the student's body posture; a control unit, the control unit generates control information through the body posture; and a correction unit, the correction unit is arranged in the double bed, and after receiving the control information, corrects the student's posture.

[0039] It's important to note that according to a World Health Organization survey of people of different ages in multiple countries and regions, approximately two million people worldwide develop occupational-related bone diseases each year due to long-term improper sitting posture. Unhealthy sitting is a significant factor affecting the physical and mental health of people of all ages. This is especially true for students. Long-term poor sitting posture can affect vision, potentially leading to myopia, and unevenly distribute force across the waist, leading to scoliosis and impacting the respiratory and digestive systems. Furthermore, long-term poor sitting posture in childhood can affect back muscle development, leading to back muscle damage and spasms, which can eventually cause dizziness and headaches. Long-term poor sitting posture can lead to degenerative cervical disc disease and cervical spine problems in adulthood. Students are prone to adopting unhealthy sitting postures in daily life. This is because, while comfortable, it can inadvertently affect muscle development, resulting in uneven force distribution and poor bone growth. Over time, this can lead to changes in posture.

[0040] Based on the above, it is essential to perform posture correction through statistical analysis of students' sitting and sleeping postures during school. To this end, the applicant has proposed a student posture correction system based on a double bed. First, an image acquisition unit installed in the classroom detects and analyzes students' sitting postures during class. The primary information collected is the student's posture image information. Furthermore, the information matching unit records the student's personal information, including gender, age, height, weight, and health information. It also records image information of standard sitting and standing postures, and binds the posture image information collected by the image acquisition unit to the personal information. Second, the posture detection unit performs feature recognition and fusion detection on the posture images containing personal information to identify postures. It should be noted that due to the complexity of human motion states, the statistical recognition process categorizes postures into the following categories based on similarity: upright sitting (i.e., standard sitting posture), left and right head tilt, left and right body tilt, squinting, uneven shoulders, forward neck tilt, and chin support. Finally, after receiving the body posture, the control unit generates control information based on the student's corresponding body posture and the corresponding matching time series. For example, if a student has a high and low shoulder posture for a long time during the day's sitting posture, the corresponding correction unit will be controlled to make the shoulders tend to the standard posture. In short, the control unit controls the correction unit by generating control information so that the student's body posture tends to the standard posture during sleep at night, or even the opposite of the non-standard posture during the day. Feature fusion through the recognition model in the body posture detection unit can accurately identify the student's body posture, and generate corresponding control information based on the body posture to control the correction process of the correction unit, which can play a corresponding preventive role for some body posture diseases. Since smart teaching has been vigorously developed in existing classroom teaching methods, cameras are generally installed in classrooms, that is, the image acquisition unit can be a traditional camera, or a high-definition dome camera or a gun camera, and is connected to the information matching unit and the body posture detection unit in a wireless or limited manner. For the control unit, a PLC programmable controller or a fuzzy controller is more preferred in this embodiment.

[0041] It should be noted that the posture detection unit includes signal connections: a node detection module, which detects key skeleton nodes of students through a sitting posture estimation model; a feature extraction module, which calculates the student's sitting posture features through key node information; and a posture recognition module. The recognition model is installed in the posture recognition module. After the sitting posture features are input into the recognition model, the recognition model identifies and outputs the student's posture after feature fusion. After the sitting posture estimation model detects the key skeleton nodes, it calculates the feature vectors based on the key nodes and forms a skeleton image based on the key nodes. The feature extraction module calculates the student's sitting posture features based on the feature vectors in the skeleton image.

[0042] It should also be noted that the OpenPose model is preferably used for the sitting posture estimation model. On this basis, in order to improve the accuracy of sitting posture recognition, the feature extraction module fuses the bone image feature vector on the basis of the bone joint feature vector. On the one hand, 12 bone joint points are used to calculate the feature vector representing the human body's sitting posture, including: 10 angle features, 12 normalized coordinates of bone joint points, 66 normalized distance features, and 22 vector features; on the other hand, the output of the network is fused with the bone joint feature vector through the feature extraction module to form a new sitting posture feature vector. The fused feature vector largely represents the student's sitting posture characteristics and is used as the input of the recognition model. Specifically, when performing sitting posture feature extraction, information on 12 bone joint points of the upper body is used, including left and right eyes, left and right ears, nose, mouth, neck, left and right shoulders, left and right elbows, and left and right wrists. For the skeletal joint information, its coordinates are normalized to the range of [0, 1]. That is, the skeleton is normalized using the maximum bounding box side length of the skeleton in the training set, which is also the bounding box-based normalization method. In this way, each skeleton is normalized by the constant of the maximum bounding box side length, and the skeleton center is aligned to [0.5, 0.5]. At this time, the skeleton structure features remain consistent and do not depend on the height of the individual student or the distance and deviation direction of the image acquisition unit, and have strong adaptability.

[0043] Furthermore, based on the original OpenPose model, structural replacement and pruning optimizations were performed, significantly reducing the number of model parameters and computational complexity, effectively improving inference speed. Specifically, the original VGG network was replaced with the Mobilenet network, while Conv4_3 and Conv4_4 were removed. The output feature map size of the Block_12_add layer was modified, and finally concatenated with the output feature map of the Block_5_add layer as the input feature map in the initial stage. These improvements not only reduced the computational resources required by the model, but also achieved an optimal balance between training-based accuracy and model size, effectively improving inference speed.

[0044] It should be noted that the VGGNet network includes 6 groups of convolution modules, and the adjacent convolution layers with the same feature map size in the VGGNet network structure constitute a residual module, which contains a total of 6 residual modules. A BN layer is added before each activation function in the residual module, and the activation function is LeakyRelu.

[0045] It should also be noted that the traditional VGGNet network has a total of 5 groups of convolution modules, which are not suitable for multi-scale body image information feature map detection. To address this situation, an additional group of convolution modules is added after the 5th pooling layer to extract feature maps of multiple scales for body image information feature extraction. In addition, since posture feature recognition is a single-category detection with a small number of categories, the network structure in the prior art has the problem of channel redundancy. In this application, the number of channels of the 4 groups of convolution modules in the shallow part of the VGGNet network is reduced to half of that of the original network, which perfectly solves the problem of channel redundancy. The addition of the BN layer can improve the network convergence speed and training stability. At the same time, using LeakyRelu as the activation function can reduce the appearance of silent neurons, ensuring that the gradient can effectively pass through the entire network architecture.

[0046] The feature extraction network needs to be combined with the detection network to achieve target detection output. Pyramid pooling is used in the detection network. Pyramid pooling performs three maximum pooling operations with a step size of 1 on the feature map. The pooled feature map is fused with the original feature map. The fused feature map can not only improve the receptive field, but also extract feature information at different scales.

[0047] Furthermore, two rounds of feature fusion are performed to achieve fine-grained detection. In the first round, feature maps of three different scales are fused from deep layers to shallow layers through upsampling. In the second round, the three feature maps of different scales output from the first round are fused again from shallow layers to deep layers through downsampling, ultimately generating feature maps of three different scales. This network structure enables iterative feature extraction, enabling more efficient feature utilization. The resulting feature maps of different scales are used to detect objects of different sizes.

[0048] Finally, three detection boxes are set for each feature point in the feature map. Each detection box contains four coordinate offsets, one target prediction confidence, and one category prediction probability. Therefore, using multi-scale feature maps for detection and fusing low-resolution and high-resolution feature maps gives the detection feature maps both shallow fine-grained features and deep semantic features, which can achieve better detection performance. To reduce the number of network parameters and make the network more lightweight, the remaining conventional convolutional structures in the recognition model are replaced with a deep separable convolutional structure. Based on the above, the VGGNet network can accurately identify the sitting postures of different students.

[0049] In this embodiment, it is preferred that the control unit is further signal-connected to a reminder unit, which is provided with a standard posture. When the posture received by the reminder unit deviates from the standard posture by a threshold and persists for a set period of time, it will remind the student of their sitting posture. Based on existing intelligent multimedia teaching, the reminder unit can be placed in teaching equipment such as a tablet. When a student's non-standard posture persists for a certain period of time, a text reminder will be sent to the student via the teaching terminal, and a standard posture will be provided to the student.

[0050] Example 2:

[0051] This embodiment only describes the parts that differ from the first embodiment, specifically:

[0052] As attached Figure 2 As shown, a method for correcting student posture based on a double bed includes the following steps: Step 1, information acquisition, in which the personal information of students in the classroom is input into the information matching unit, and the posture image information acquired by the image acquisition unit is bound to the personal information to obtain the posture image information with the personal information; Step 2, posture detection, in which after Step 1 is completed, the posture detection unit performs feature extraction and fusion on the posture image information to obtain the posture of each student in the classroom; Step 3, sleep correction, in which the control unit generates control information through statistics of the posture, and controls the correction unit to correct the students' posture during sleep time.

[0053] It should be noted that in step 1, the information matching unit records the student's personal information, including gender, age, height, weight, and health information. Image information of standard sitting and standing postures is also recorded. The image acquisition unit collects the posture image information and binds it to the personal information. The time period for recording personal information can be one semester or one academic year. In this embodiment, the preferred update period is the academic year. Step 2 specifically includes the following processes: First, the node detection module detects the key nodes of the student's skeleton using a sitting posture estimation model; second, the feature extraction module calculates the student's sitting posture features using the key node information; finally, the recognition model is installed in the posture recognition module. After the sitting posture features are input into the recognition model, the recognition model identifies and outputs the student's posture after feature fusion. The identified student postures are classified into: sitting upright (i.e., standard sitting posture), head tilted left and right, body tilted left and right, squinting, uneven shoulders, forward neck tilt, and chin-supporting posture.

[0054] Example 3:

[0055] This embodiment only describes the parts that differ from embodiment 2, specifically:

[0056] For head tilt posture,

[0057] The determination of head tilt mainly uses the coordinates of the key points of both eyes. The head condition includes left and right tilt, and the horizontality of the line connecting the key points of both eyes is determined by whether the key points of the human head are in a horizontal position. To increase the robustness and practicality of sitting posture determination, considering the influence of the front-to-back distance between the pupil and the image acquisition unit and the horizontal distance of the pupil in space, a horizontal threshold α is added to the model. The specific judgment formula is:

[0058] Formula for judging left tilt of head:

[0059] When the student is on the right side of the picture: ;

[0060] When the student is on the left side of the picture: ;

[0061] Formula for judging head right tilt:

[0062] When the student is on the right side of the picture:

[0063] When the student is on the left side of the picture:

[0064] in,

[0065] Indicates the y coordinate of the left eye;

[0066] Indicates the y coordinate of the right eye;

[0067] represents the threshold of both eyes on the vertical coordinate. In this embodiment, it is a constant of 2. Due to the error in key point detection, that is, the error in the vertical coordinates of the left and right eyes, both eyes are also judged to be in the horizontal position if it is within this threshold.

[0068] L represents the distance between the center of both eyes and the horizontal center of the image;

[0069] A represents the horizontal length of the divided area;

[0070] α represents the horizontal threshold parameter.

[0071] The difference in the vertical coordinates of the left and right eyes is used to determine head tilt. When the person is on the left side of the image and the head is horizontal, the left eye will be slightly higher than the right eye. When the person is on the right side, the right eye is slightly higher than the left eye. The farther away from the center, the more significant the difference in the key point coordinates. Therefore, the difference used in different areas varies, enabling multi-angle head tilt detection. Due to the error in key point detection, a vertical coordinate threshold is added.

[0072] The principles of posture detection for body tilting left and right, squinting, uneven shoulders, forward neck, and chin support are similar to those described above and will not be described in detail in this embodiment.

[0073] Example 4:

[0074] This embodiment only describes the parts that differ from the first embodiment, specifically:

[0075] As attached Figure 3 As shown, the correction unit includes a double bed body 1, a panel 2, and a power member 3 disposed below the panel 2. The panel 2 includes a frame 21, a movable backboard 22, and a movable bottom board 23. The movable backboard 22 is hinged to the movable bottom board 23, and together they constitute a bed board disposed on the frame 21. The movable backboard 22 can be moved upward by the power member 3, the bottom of the movable bottom board 23 can be folded downward by the power member 3, and the middle of the movable bottom board 23 can be folded upward by the power member 3. It should be noted that the length and width structure of the frame 21 are consistent with those of a traditional bed board. In this embodiment, the height of the frame 21 is preferably 20 cm to 25 cm. The power member 3 is disposed in the middle, and the space at the left and right ends can be utilized as storage cabinets. Another preferred implementation in this embodiment is: for the movable backboard 22, the overall length is preferably two-fifths of the entire bed board length, and the overall length of the movable baseboard 23 is preferably two-fifths of the entire bed board length, and the remaining one-fifth of the length is used as a fixed part. The connection method of the movable backboard 22 and the movable baseboard 23 using this method is changed accordingly. Corresponding control information is generated by statistical analysis of the student's posture during the day, and the nighttime sleeping posture is sensed by the pressure sensors on the movable backboard 22 and the movable baseboard 23. The control information is then used to control the correction and then move to the correction part for skeletal auxiliary correction. In addition, the movable structure of the movable backboard 22 and the movable baseboard 23 can be changed, which is more ergonomic and can help improve the student's sleep state at night.

[0076] It should be noted that the movable back panel 22 includes a movable main panel 222 and two movable side panels 221. The two movable side panels 221 are hingedly arranged on the left and right sides of the movable main panel 222. The movable bottom panel 23 includes a first bottom panel 231 and a second bottom panel 232. The first bottom panel 231 is hingedly connected to the second bottom panel 232, and the first bottom panel 231 is also hingedly connected to the movable main panel 222. The movable main panel 222 can be flipped around the hinged portion by the power member 3. The movable side panels 221 can be flipped toward the center compared to the movable main panel 222 to adapt to the posture of side sleeping correction. When the pressure sensor detects that the student's sleeping posture is side sleeping, it will control the corresponding movable side panel 221 to flip, so that the correction member 4 can massage the student's back or neck.

[0077] As attached Figure 7 As shown, a slot 223 is provided below the movable mainboard 222. A first connecting rod 224 is slidably provided within the slot 223. A first slider 225 is rotatably provided at the other end of the first connecting rod 224. The power member 3 includes a first motor 227. The output end of the first motor 227 is connected to a first worm 226. The first worm 226 movably penetrates the first slider 225 via a thread. When the first motor 227 is started, the first slider 225 slides on the first worm 226 to achieve the raising and lowering of the movable mainboard 222. It should be noted that the lower end surface of the skeleton 21 is a plate structure. The first motor 227 is provided on this plate structure and serves as the power source for the flipping of the movable backboard 22. The first motor 227 can cause the first slider 225 to move relative to the first worm 226. One end of the first connecting rod 224 is slidably provided with the slot 223, and the other end is hinged to the first slider 225. Based on the above structure, the movement of the first slider will cause the movable mainboard 222 to flip.

[0078] As attached Figure 8 As shown, two bases 24 are symmetrically provided on the sides of the slide 223. The two bases 24 are fixedly connected to the lower end surface of the movable main board 222. A second motor 228 is provided on the base 24. The output end of the second motor 228 is connected to a second worm 229. A second slider 230 is slidably provided on the second worm 229. A second connecting rod 220 is hingedly provided on the side of the base 24. One end of the second connecting rod 220 is hinged to the second slider 230, and the other end of the second connecting rod 220 is slidably connected to the movable side panel 221. It should be noted that based on the above structure, since the base 24 is fixedly connected to the movable side panel 221 and the second motor 228 is provided on the base 24, when the second motor 228 is started, the movement process is similar to that of the movable main board 222, except that the movable side panel 221 can be folded upward along the center, thereby being suitable for students' side sleeping posture.

[0079] It should be noted that a controllable gas spring 233 is hingedly installed below each of the first and second base plates 231, 232. A fixed plate 234 is fixedly connected to the lower end surface of the first base plate 231. The free end of the controllable gas spring 233 hinged to the second base plate 232 is hinged to the fixed plate 234, while the free end of the controllable gas spring 233 hinged to the first base plate 231 is hinged to the frame 21. Based on this structure, when the controllable gas spring 233 is extended and retracted, it will correspond to the movement of the first or second base plate 231, 232, respectively, to achieve a more ergonomic posture.

[0080] It should be noted that the interior of the movable side panel 221 and the movable main board 222 is hollow, and a correction member 4 is provided inside. The correction member 4 can slide inside the movable side panel 221 and the movable main board 222. A plurality of through holes 5 are provided on the upper end surfaces of the movable side panel 221 and the movable main board 222, and an opening and closing member 6 is provided at the through hole 5. The correction member 4 is connected to the control unit signal, and can slide to a specified position after receiving the control information, and the opening and closing of the opening and closing member 6 is controlled by the control unit. The correction member 4 then corrects the posture of the student above. It should also be noted that the movement of the correction member 4 inside the movable side panel 221 and the movable main board 222 is preferably carried out by means of a slide rail or the like, and can also be moved by means of a worm or the like. As for the control process, the control process of the correction unit by the control unit is a pre-edited continuous control process, that is, through the analysis of the student's individual posture change process during the day, a customized correction process is generated by the convolutional neural network prediction. The correction part 4 can move continuously in the movable side panel 221 and the movable main board 222, and perform massage correction on different points.

[0081] As attached Figure 9 As shown, a plurality of limiting grooves 65 are provided on the outer side of the lower end surface of each through hole 5. The opening and closing member 6 includes: a rotating disk 61 rotatably arranged below the through hole 5 and a plurality of opening and closing plates 62. The upper end surface of the rotating disk 61 is provided with an opening and closing groove 63. The plurality of opening and closing plates 62 are arranged in the opening and closing groove 63 and can be spliced together to close the through hole 5. A limiting pin 64 is provided above the opening and closing plate 62. The limiting pin 64 can slide in the limiting groove 65. When the rotating disk 61 rotates, the opening and closing plates 62 can be opened and closed by the action of the limiting groove 65 and the opening and closing groove 63. It should be noted that based on the above structure, when the rotating disk 61 rotates under the action of an external force, it will drive the opening and closing plates 62 to move in the limiting groove 65, and under the combined action of the opening and closing groove 63, it can close the through hole 5, thereby preventing the student's bedding from being moved inside in an uncorrected state.

[0082] As attached Figure 10As shown, the correction member 4 includes: a correction seat 41, a correction motor 42 and several correction heads 44, the correction motor 42 is arranged in the correction seat 41, the output end of the correction motor 42 is connected to the correction head 44 through a gear set, the upper end of the correction head 44 is provided with a plurality of balls 45 for rolling, the outer peripheral surface of the correction head 44 is provided with a rack, the outer peripheral surface of the rotating disk 61 is provided with a toothed belt, and the correction head 44 can be extended and retracted by the correction motor 42. When the correction member 4 moves to the bottom of the through hole 5, the correction motor 42 is started, and the correction motor 42 can drive the correction head 44 to move upward. When the rack is engaged with the toothed belt, it will drive the opening and closing plate 62 to open. At the same time, the correction head 44 continues to move upward. In the above process, the correction head 44 will massage and squeeze the student's back to reduce the impact of non-standard posture on the bones. It should be noted that all of the above-mentioned motors, controllable gas springs 233 and other structures are subject to integrated control by the control unit. In this embodiment, it is more preferred that the correction head 44 can also slide up and down in the correction seat 41. A cylinder 43 is also provided in the correction seat 41. A wedge block is provided at the end of the cylinder 43. The wedge block cooperates with the bottom of the correction head 44. When the cylinder 43 is extended or retracted, the wedge block is used to enable the correction head 44 to slide up and down in the correction seat 41, thereby passing through the through hole 5 opened by the opening and closing member 6 to correct the student's individual homework. In addition, the rising process of the correction head 44 and the rotation process of the rotating disk 61 are a linked process, that is, the rising process of the correction head 44 and the opening and closing process of the opening and closing plate 62 are synchronized. In addition, a correction rod 46 is also provided in the center of the correction head 44. A spring is provided at the bottom of the correction rod 46. During the rising process, the correction head 44 can also press the correction point back and forth through the correction rod 46 to relieve back fatigue.

[0083] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A student posture correction system based on a double bed, comprising an image acquisition unit disposed in a classroom and configured to acquire posture image information of students in the classroom, characterized in that: Also included is: An information matching unit, wherein the student's personal information is recorded in the information matching unit and is used to match the personal information with the body image information; A posture detection unit, wherein the posture detection unit is equipped with a recognition model, and the recognition model uses a VGGNet network to perform feature fusion to recognize the student's posture; a control unit, the control unit generating control information through body posture; and a correction unit, which is disposed in the double bed and corrects the student's posture after receiving the control information; The correction unit includes a double bed body (1), a plate (2), and a power member (3) arranged below the plate (2), wherein the plate (2) includes a frame (21), a movable backboard (22), and a movable bottom board (23), wherein the movable backboard (24) is hinged to the movable bottom board (23), and together constitutes a bed board arranged on the frame (21), wherein the movable backboard (22) can be moved upward by the power member (3), the bottom of the movable bottom board (23) can be folded downward by the power member (3), and the middle of the movable bottom board (23) can be folded upward by the power member (3); corresponding control information is generated by statistical analysis of the student's posture during the day, and the sleeping posture at night is sensed by the pressure sensors on the movable backboard (22) and the movable bottom board (23), and then the correction is controlled by the control information and the skeleton is assisted in correction when it moves to the correction part.

2. The student posture correction system based on a double bed according to claim 1, characterized in that: The posture detection unit includes signal-connected: A node detection module, which detects key nodes of the student skeleton through a sitting posture estimation model; Feature extraction module, which calculates the student's sitting posture features through key node information, The posture recognition module includes the recognition model installed in the posture recognition module. After the sitting posture features are input into the recognition model, the recognition model recognizes and outputs the student's posture after feature fusion.

3. The student posture correction system based on a double bed according to claim 1, characterized in that: The VGGNet network includes 6 groups of convolution modules, and adjacent convolution layers with the same feature map size in the VGGNet network structure constitute a residual module, which contains a total of 6 residual modules. A BN layer is added before each activation function in the residual module, and the activation function is LeakyRelu.

4. The double bed-based student posture correction system according to claim 2, characterized in that: The sitting posture estimation model detects the key nodes of the skeleton and calculates the feature vectors according to the key nodes, and forms a skeleton image according to the key nodes. The feature extraction module calculates the student's sitting posture features according to the feature vectors in the skeleton image.

5. The student posture correction system based on a double bed according to claim 1, characterized in that: The control unit is also signal-connected to a reminder unit, in which a standard posture is provided. When the posture received by the reminder unit deviates from the standard posture by a threshold value and lasts for a set time, the student is reminded of the sitting posture.

6. A method for correcting student posture based on a double bed, characterized by: A student posture correction system based on a double bed according to any one of claims 1 to 5, comprising the following steps: Step 1: Information collection: inputting personal information of students in the classroom into the information matching unit, and binding the body image information collected by the image collection unit with the personal information to obtain body image information with personal information; Step 2: Posture detection. After step 1 is completed, the posture detection unit extracts and fuses the posture image information to obtain the posture of each student in the classroom. Step 3, sleep correction, the control unit generates control information through statistics of body posture, and controls the correction unit to correct the student's posture during sleep time.

Images