Personalized exercise prescription recommendation system based on big data analysis

Abstract

The invention discloses a personalized exercise prescription recommendation system based on big data analysis. And the motion mode suitable for the user is found by adopting the information of the user. And under the condition of gravity, the pressure is downward, so that the human body is supported by two feet. And the pressure has a relationship with the posture of the human body, so that a plurality of training segmentation images can be obtained through historical user motion footprint images in the training process. And establishing a relationship between the segmented image and the pressure of the posture of the user on the footprint. Therefore, the plurality of segmented images not only can represent the motion posture of the user, but also can represent the force exertion condition of each part when the user makes the posture. And judging whether the action of the user is standard. Therefore, the exercise prescriptions that the user can make the exercise postures more smoothly and exert accurate force on each part are recommended.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a personalized exercise prescription recommendation system based on big data analysis. Background Art

[0002] At present, recommendations are mainly made based on user information. In addition, different disease information, scene information, and energy consumption information are added for personalized recommendations. The user's movement posture is not taken into account to judge the accuracy of the movement. Because during the user's exercise, non-standard postures cannot achieve the purpose of exercise, and some movements are difficult to correct due to the user's own reasons, so it is necessary to provide users with postures that are easier to do. Summary of the invention

[0003] The purpose of the present invention is to provide a personalized exercise prescription recommendation system based on big data analysis to solve the above-mentioned problems existing in the prior art, including a monitoring module and a processor.

[0004] The monitoring module is used to obtain user information; the user information includes the user's age, gender and symptoms; The processor is used to process the following method: Inputting the user information into a motion mode discrimination network to obtain a plurality of motion modes; the motion modes represent the types of motion suitable for the user; According to the motion mode, a user motion footprint image and a corresponding user motion posture image are obtained; the user motion footprint image represents a two-dimensional grayscale image including footprints left by the user during motion; the user motion posture image represents an image including the corresponding user motion posture when recording the user motion footprint image; Based on the user motion posture image and the user motion footprint image, a footprint support vector is obtained; the footprint support vector represents the pressure of the user's posture contained in the two-dimensional user motion posture image on the footprint; Through the segmentation discriminant network, based on the user's motion posture image, the pressure corresponding to each part of the user's motion footprint image caused by the posture is found, and multiple cutting angles are obtained; Based on the cutting angle and the user motion posture image, the user motion posture image is segmented at different angles to obtain a plurality of segmented images; the segmented images are images including the user motion footprint images; Through the posture discrimination network, based on the segmented image, it is judged whether the movement posture is correct, and a recommended movement prescription is obtained.

[0005] Optionally, the segmentation discriminant network is used to find the pressure corresponding to each part of the user's motion footprint image caused by the posture based on the user's motion posture image, and obtain multiple cutting angles, including: Inputting the user motion posture image into an edge detection network to detect the user posture to obtain a first posture image; the first posture image represents the user's body posture; Inputting the first posture image into a body part discrimination network to obtain a user body part position image; the user body part position image is an image marking the positions of multiple body parts of the user; The user's part position image is input into the segmentation discriminant network to obtain multiple cutting angles.

[0006] Optionally, the training method of the segmentation discriminant network includes: Acquire a plurality of training images; the training images are historical user body position images; Obtaining a footprint support vector corresponding to the training image as annotation data; Inputting the training image into a segmentation discriminant network to obtain a training cutting angle; Based on the training cutting angle and the annotation data, the training image is cut to obtain a plurality of training segmented images and corresponding second marking points; the training segmented image is an image cut according to the training cutting angle; the second marking point represents the midpoint of an edge of the first segmented image on the annotated basic straight line; Based on the training segmented image and the second marking point, the degree of force is determined with the midline as a reference, and a plurality of training segmented grayscale total values ​​and corresponding training segmented distances are obtained; the training segmented distance represents a distance from the ground in the training segmented image; the training segmented grayscale total value represents the pressure of each posture in the training segmented image at the training segmented distance; Based on multiple training segmentation grayscale total values ​​and corresponding training segmentation distances, a training pressure value is obtained; multiple training segmentation grayscale total values ​​correspond to one training pressure value; multiple training segmentation images correspond to multiple training pressure values; the training pressure value represents the pressure exerted by the entire training segmentation image on the ground through the footprints; Dividing the multiple training pressure values ​​by the sum of the multiple training pressure values, respectively, to obtain a training pressure ratio; the training pressure ratio represents the ratio of the pressures exerted on the ground by the footprints of the multiple training segmented images; According to the training pressure ratio and the footprint support vector, the loss is calculated and the segmentation discriminant network is trained.

[0007] Optionally, obtaining a footprint support vector based on the user motion posture image and the user motion footprint image includes: Taking the straight line where the user motion posture image and the user motion footprint image intersect as a basic straight line; Segmenting the user's motion footprint image in a direction perpendicular to the base straight line to obtain a plurality of first segmented images; Adding the gray values ​​in the first segmented images to obtain a footprint pressure value; obtaining a plurality of footprint pressure values ​​corresponding to the plurality of first segmented images; According to the positions of the plurality of first segmented images in the user's motion footprint image, the plurality of footprint pressure values ​​are arranged in sequence to obtain a footprint support vector.

[0008] Optionally, the step of cutting the training image based on the training cutting angle and the annotation data to obtain a plurality of training cut images and corresponding second marking points includes: The first segmented image corresponding to the user running footprint image corresponding to the annotated data is used as the annotated first segmented image; Using the basic straight line corresponding to the annotated first segmented image as the annotated basic straight line; Marking the first segmented image at positions corresponding to the training image on the marking base straight line to obtain a plurality of first marking points; The midpoint of two connected first marking points is used as the second marking point; Find the angle difference from the marked base straight line to train the cutting angle, and use the straight line through the second marked point as the cutting line; The training image is segmented using segmentation lines to obtain a training segmented image.

[0009] Optionally, based on the training segmentation image and the second marking point, the degree of force is determined based on the midline as a reference to obtain a plurality of training segmentation grayscale total values ​​and corresponding training segmentation distances, including: According to the training segmentation image and the second marking point, finding the midline of the training segmentation image as the training segmentation midline; Based on the training segmentation image and the training segmentation center line, the pressure of the user's posture is projected onto the training segmentation center line to obtain a plurality of training segmentation grayscale total values; one value on the training segmentation center line corresponds to one training segmentation grayscale total value; Obtain the position of the value on the training segmentation midline on the training image as the training segmentation position; According to the training segmentation position and the position of the second marking point, the distance to the second marking point is calculated by the Pythagorean theorem to obtain the training segmentation distance; The training segmentation grayscale total value is associated with the corresponding training segmentation distance.

[0010] Optionally, based on the training segmentation image and the training segmentation centerline, projecting the pressure of the user's posture onto the training segmentation centerline to obtain a plurality of training segmentation grayscale total values ​​includes: A point on the training segmentation midline is obtained as a segmentation point; multiple points on the training segmentation midline correspond to multiple segmentation points; A straight line passing through the segmentation point and perpendicular to the training segmentation midline is obtained as the segmentation bisector; multiple segmentation bisectors are obtained corresponding to multiple segmentation points; In the training segmentation image, the grayscale values ​​on the segmentation bisector are added to obtain a training segmentation total grayscale value; and a plurality of segmentation bisectors correspond to obtaining a plurality of training segmentation total grayscale values.

[0011] Optionally, obtaining the training pressure value based on a plurality of training segmentation grayscale total values ​​and corresponding training segmentation distances includes: Add multiple training segmentation distances to get the total predicted distance value; The training segmentation distance is divided by the total value of the predicted distance to obtain a predicted distance pressure value; multiple training segmentation distances correspond to multiple predicted distance pressure values; The training segmentation grayscale total value is multiplied by the corresponding predicted distance pressure value to obtain a training segmentation pressure value; multiple training segmentation grayscale total values ​​correspond to multiple training segmentation pressure values; Adding multiple training segmentation pressure values ​​together to obtain a training pressure value; the total grayscale values ​​of multiple training segmentations correspond to one training pressure value; Multiple training segmentation images correspond to multiple training pressure values.

[0012] Optionally, finding the midline of the training segmentation image as the training segmentation midline according to the training segmentation image and the second marking point includes: If the training segmented image is a quadrilateral, four quadrilateral vertices are obtained; Connect two non-adjacent four-sided vertices to obtain two four-sided line segments; The point where two four-sided line segments intersect is taken as the first focus; The second marking point corresponding to the training segmentation image is connected to the first key point to obtain the training segmentation centerline.

[0013] Optionally, finding the midline of the training segmentation image as the training segmentation midline according to the training segmentation image and the second marking point includes: If the training segmented image is a triangle, three triangle vertices are obtained; The angle containing the second marked point is taken as the second marked angle; The side corresponding to the second marked angle is used as the second marked side; Take the midpoint of the second marked side as the first midpoint; Connect the first midpoint with the second marked point to obtain the training segmentation midline.

[0014] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects: The embodiment of the present invention also provides a personalized exercise prescription recommendation system based on big data analysis: The user's information is used to find a movement method that suits the user. The personalization of the exercise prescription recommendation can be improved. The user's posture and the pressure of the human body on the ground represented by the user's footprints are associated. Because under gravity, the pressure is downward, and the human body is supported by both feet. The pressure is related to the posture of the human body, so during the training process, multiple training segmented images can be obtained through historical user movement footprint images. The training segmented midline is found in the training segmented image, and the training segmented midline indicates that the training segmented image can be divided into two parts of equal area, and the position information corresponding to the posture is retained. The value on the training segmented midline reflects the pressure received by a point. The training segmented midline and the training segmented image are visualized through the segmentation discriminant network. Thereby, the segmented image and the pressure of the user's posture on the footprint are related. Therefore, multiple segmented images can not only represent the posture of the user's movement, but also represent the force of each part when the user makes the posture. Thereby judging whether the user's action is standard. The technical effect of recommending a sports prescription that allows the user to make a more smooth exercise posture and accurately apply force to each part is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a method flow chart corresponding to a processor in a personalized exercise prescription recommendation system based on big data analysis provided by an embodiment of the present invention.

[0016] Figure 2 It is a schematic diagram of the installation positions of a footprint detection device and a posture camera device in a personalized exercise prescription recommendation system based on big data analysis provided by an embodiment of the present invention.

[0017] Figure 3 It is a schematic diagram of a method for obtaining footprint support vectors corresponding to a processor in a personalized exercise prescription recommendation system based on big data analysis provided by an embodiment of the present invention.

[0018] Figure 4 It is a schematic diagram of a training segmentation image acquisition method corresponding to a processor in a personalized exercise prescription recommendation system based on big data analysis provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The present invention will be described in detail below in conjunction with the accompanying drawings. Example

[0020] like Figure 1 As shown, an embodiment of the present invention provides a personalized exercise prescription recommendation system based on big data analysis, and the system includes a monitoring module and a processor: The monitoring module is used to obtain a user's motion footprint image and a user's motion posture image. The user's motion footprint image represents a two-dimensional footprint image left by the user when exercising.

[0021] The footprint image is a grayscale image, and grayscale values ​​are used to represent the force applied to the foot.

[0022] The monitoring module is used to obtain user information; the user information includes the user's age, gender and symptoms; The processor is used to process the following method: S101: Inputting the user information into a motion mode discrimination network to obtain a plurality of motion modes; the motion modes represent the types of motion suitable for the user.

[0023] Wherein, the movement mode discrimination network is a fully connected neural network (Fully Connected Neural Network).

[0024] The sports category represents a set of pre-set sports movements.

[0025] S102: According to the movement mode, obtain a user movement footprint image and a corresponding user movement posture image; the user movement footprint image represents a two-dimensional grayscale image including footprints left by the user when exercising; the user movement posture image represents an image including the corresponding user movement posture when recording the user movement footprint image.

[0026] The user's motion footprint image can be obtained by a footprint detection device placed under the user's feet, and the user's motion footprint image can be obtained by a posture camera device that shoots perpendicular to the horizontal plane. In this embodiment, the footprint detection device can use a gravity sensor, or it can be stained with paint on the sole of the foot, and then scanned after printing the footprint. By finding the proportion of all values ​​and multiplying it by 256, it is converted into a value of 0-255. Finally, the gravity or printed footprint is converted into a grayscale image that can reflect the pressure of the footprint.

[0027] Among them, in the user movement posture image, raising a hand is a user's posture.

[0028] The installation positions of the footprint detection device and the posture camera device are as follows: Figure 2 In this embodiment, the footprint detection device is placed on the ground, and the posture camera device is placed on a wall perpendicular to the ground.

[0029] S103: Obtaining a footprint support vector based on the user motion posture image and the user motion footprint image; the footprint support vector represents the pressure of the user's posture contained in the two-dimensional user motion posture image on the footprint.

[0030] S104: using a segmentation discriminant network, based on the user's motion posture image, finding the pressure corresponding to each part of the user's motion footprint image caused by the posture, and obtaining a plurality of cutting angles; S105: Based on the cutting angle and the user motion posture image, the user motion posture image is segmented at different angles to obtain a plurality of segmented images; the segmented images are images including the user motion footprint images.

[0031] S106: Using a posture discrimination network, based on the segmented image, determine whether the exercise posture is correct, and obtain a recommended exercise prescription.

[0032] The posture discrimination network is a LetNet network. The posture discrimination network is used to determine whether the user's exercise posture is a correct exercise posture. If the accuracy of each segmented image is low, an exercise prescription with a higher accuracy is recommended to the user.

[0033] Optionally, the segmentation discriminant network is used to find the pressure corresponding to each part of the user's motion footprint image caused by the posture based on the user's motion posture image, and obtain multiple cutting angles, including: The user motion posture image is input into an edge detection network to detect the user posture to obtain a first posture image; the first posture image represents the user's body posture.

[0034] Among them, the canny operator is used as the edge detection network to extract the edges of the user's posture.

[0035] The first posture image is input into a part discrimination network to obtain a user part position image; the user part position image is an image marking the positions of multiple parts of the user's body.

[0036] The part discrimination network is a yolov5 model. The part discrimination network is a network that uses multiple marked positions of user parts such as the user's arms, legs, head, and torso.

[0037] The user's part position image is input into the segmentation discriminant network to obtain multiple cutting angles.

[0038] Optionally, the training method of the segmentation discriminant network includes: Acquire multiple training images; the training images are historical user body position images.

[0039] The footprint support vector corresponding to the training image is obtained as the annotation data.

[0040] Among them, multiple training images correspond to multiple footprint support vectors and multiple annotation data.

[0041] The training image is input into a segmentation discriminant network to obtain a training cutting angle.

[0042] Among them, the segmentation discriminant network can visualize the values ​​on the subsequent training segmentation centerline.

[0043] Based on the training cutting angle and the annotation data, the training image is cut to obtain a plurality of training segmented images and corresponding second marking points; the training segmented image is an image cut according to the training cutting angle; the second marking point represents the midpoint of the edge of the first segmented image on the annotated basic straight line; The training segmented image is segmented according to the force conditions of the footprints.

[0044] Based on the training segmented image and the second marking point, the degree of force is determined with the midline as a reference, and a plurality of training segmented grayscale total values ​​and corresponding training segmented distances are obtained; the training segmented distance represents a distance from the ground in the training segmented image; the training segmented grayscale total value represents the pressure of each posture in the training segmented image at the training segmented distance.

[0045] Based on multiple training segmentation grayscale total values ​​and corresponding training segmentation distances, a training pressure value is obtained; multiple training segmentation grayscale total values ​​correspond to one training pressure value; multiple training segmentation images correspond to multiple training pressure values; the training pressure value represents the pressure exerted by the entire training segmentation image on the ground through footprints.

[0046] The multiple training pressure values ​​are divided by the sum of the multiple training pressure values ​​to obtain multiple training pressure ratios; the training pressure ratios represent the ratios of the pressures generated by the multiple training segmented images on the ground through the footprints.

[0047] Among them, one training pressure value corresponds to one training pressure ratio value.

[0048] Among them, if the training stress values ​​are 23, 34, 53, and 21 respectively. 23+34+53+21=131, then the training stress ratios are 23 / 131, 34 / 131, 53 / 131, and 21 / 131 respectively.

[0049] According to the training pressure ratio and the footprint support vector, the loss is calculated and the segmentation discriminant network is trained.

[0050] The values ​​in the footprint support vector are added to obtain the total value of the footprint support vector. The value corresponding to an element in the footprint support vector is divided by the total value of the footprint support vector to obtain the footprint support vector ratio.

[0051] The loss is calculated by comparing the footprint support vector ratio with the corresponding training pressure ratio. The correlation between the footprint support vector ratio and the corresponding training pressure ratio is constructed by the position order of multiple training segmented images and the element subscripts of multiple footprint support vectors. For example, the subscripts of multiple training segmented images are marked as 0, 1, 2, and 3 from left to right in the training image. Then the training segmented image with subscript 0 corresponds to the element of the footprint support vector with subscript 0.

[0052] Among them, the cross entropy loss function is used to calculate the loss.

[0053] Optionally, obtaining a footprint support vector based on the user motion posture image and the user motion footprint image includes: The straight line where the user motion posture image and the user motion footprint image intersect is used as the basic straight line.

[0054] The basic straight line is as follows: Figure 3 .

[0055] The user movement footprint image is segmented in a direction perpendicular to the basic straight line to obtain a plurality of first segmented images.

[0056] The first segmented image and the user motion footprint image are in the same horizontal plane.

[0057] Among them, the multiple first segmented images have only one segmentation direction which is a direction perpendicular to the base straight line.

[0058] In this embodiment, the direction of the user's motion footprint image parallel to the basic straight line is used as the horizontal coordinate, the direction perpendicular to the basic straight line is used as the vertical coordinate, and the lower left corner of the user's motion footprint image is used as the origin. The width of the first segmented image is equal to the width of the user's motion footprint image. The lengths of the multiple first segmented images are equal, and the sum of the lengths of the multiple first segmented images is equal to the length of the user's motion footprint image.

[0059] The segmentation direction of the first segmented image is as follows: Figure 3 shown.

[0060] Adding the gray values ​​in the first segmented images to obtain a footprint pressure value; obtaining a plurality of footprint pressure values ​​corresponding to the plurality of first segmented images; According to the positions of the plurality of first segmented images in the user's motion footprint image, the plurality of footprint pressure values ​​are arranged in sequence to obtain a footprint support vector.

[0061] Among them, if the user's motion footprint image is 64*64, the lower corner of the user's motion footprint image is taken as the origin, the length is the horizontal coordinate, and the width is the vertical coordinate. Divide into 4 first segmented images, each of which is 16 long and 64 wide. Put the footprint pressure value corresponding to the first segmented image corresponding to the horizontal coordinate 0-15 at the position with subscript 0 in the footprint support vector. Put the footprint pressure value corresponding to the first segmented image corresponding to the horizontal coordinate 16-31 at the position with subscript 1 in the footprint support vector. Put the footprint pressure value corresponding to the first segmented image corresponding to the horizontal coordinate 32-47 at the position with subscript 2 in the footprint support vector. Put the footprint pressure value corresponding to the first segmented image corresponding to the horizontal coordinate 48-64 at the position with subscript 3 in the footprint support vector.

[0062] Optionally, the step of cutting the training image based on the training cutting angle and the annotation data to obtain a plurality of training cut images and corresponding second marking points includes: The first segmented image corresponding to the user's running footprint image corresponding to the annotated data is used as the annotated first segmented image; and a plurality of annotated first segmented images are obtained corresponding to the plurality of first segmented images; Using the basic straight line corresponding to the annotated first segmented image as the annotated basic straight line; The first segmented image is marked by marking positions on the marking base straight line corresponding to positions on the training image to obtain a plurality of first marking points.

[0063] The midpoint of two connected first marking points is used as the second marking point.

[0064] Wherein, the first marking point and the second marking point are as follows: Figure 4 shown.

[0065] The angle difference from the marked base straight line is found to be the training cutting angle, and the straight line passing through the second marked point is used as the cutting line.

[0066] In this embodiment, the angle from right to left of the labeled basic straight line is 0 degrees, and the angle from left to right is 180 degrees. If the following index is 0, the training cutting angle corresponding to the first training cutting angle of the first labeled first segmented image is 45 degrees.

[0067] The dividing line is Figure 4 shown.

[0068] The training image is segmented using segmentation lines to obtain a training segmented image.

[0069] Optionally, based on the training segmentation image and the second marking point, the degree of force is determined based on the midline as a reference to obtain a plurality of training segmentation grayscale total values ​​and corresponding training segmentation distances, including: According to the training segmentation image and the second marking point, a midline of the training segmentation image is found as the training segmentation midline.

[0070] Wherein, a plurality of vertices in the training segmented image are obtained. If the training segmented image is a triangle, then there are three vertices. Through the above method, because under gravity, the pressure is downward, but the human body is supported by the footprints. The pressure is related to the posture of the human body, so the training segmented image can be allocated according to the pressure supported by the footprints. The training segmented midline indicates that the training segmented image can be divided into two parts of equal area, the area is divided equally, and the position information corresponding to the posture can be retained.

[0071] Based on the training segmentation image and the training segmentation center line, the pressure of the user's posture is projected onto the training segmentation center line to obtain a plurality of training segmentation grayscale total values; a value on the training segmentation center line corresponds to a training segmentation grayscale total value.

[0072] Obtain the position of the value on the training segmentation midline on the training image as the training segmentation position; According to the training segmentation position and the position of the second marking point, the distance to the second marking point is calculated by the Pythagorean theorem to obtain the training segmentation distance.

[0073] The lower left corner of the training image is used as the origin, the length is the horizontal coordinate, and the width is the vertical coordinate to establish the coordinate axis. The absolute value of the horizontal coordinate of the second marking point is subtracted from the horizontal coordinate of the training segmentation position as the first Pythagorean value, the horizontal coordinate of the training segmentation position is used as the second Pythagorean value, the square of the first Pythagorean value is added to the square root of the second Pythagorean value, and the training segmentation distance is obtained.

[0074] The training segmentation grayscale total value is associated with the corresponding training segmentation distance.

[0075] Optionally, based on the training segmentation image and the training segmentation centerline, projecting the pressure of the user's posture onto the training segmentation centerline to obtain a plurality of training segmentation grayscale total values ​​includes: A point on the training segmentation midline is obtained as a segmentation point; multiple points on the training segmentation midline correspond to multiple segmentation points; A straight line passing through the segmentation point and perpendicular to the training segmentation midline is obtained as the segmentation bisector; multiple segmentation bisectors are obtained corresponding to multiple segmentation points.

[0076] Through the above method, the value on the training segmentation centerline represents the pressure exerted on the posture as a point.

[0077] In the training segmentation image, the grayscale values ​​on the segmentation bisector are added to obtain a training segmentation total grayscale value; and a plurality of segmentation bisectors correspond to obtaining a plurality of training segmentation total grayscale values.

[0078] Optionally, obtaining the training pressure value based on a plurality of training segmentation grayscale total values ​​and corresponding training segmentation distances includes: Add multiple training segmentation distances to get the total predicted distance value; The training segmentation distance is divided by the total value of the predicted distance to obtain a predicted distance pressure value; multiple training segmentation distances correspond to multiple predicted distance pressure values; The training segmentation grayscale total value is multiplied by the corresponding predicted distance pressure value to obtain a training segmentation pressure value; multiple training segmentation grayscale total values ​​correspond to multiple training segmentation pressure values; Adding multiple training segmentation pressure values ​​together to obtain a training pressure value; the total grayscale values ​​of multiple training segmentations correspond to one training pressure value; Multiple training segmentation images correspond to multiple training pressure values.

[0079] Optionally, finding the midline of the training segmentation image as the training segmentation midline according to the training segmentation image and the second marking point includes: If the training segmented image is a quadrilateral, four quadrilateral vertices are obtained; Connect two non-adjacent four-sided vertices to obtain two four-sided line segments; The point where two four-sided line segments intersect is taken as the first focus; The second marking point corresponding to the training segmentation image is connected to the first key point to obtain the training segmentation centerline.

[0080] Optionally, the personalized exercise prescription recommendation system based on big data analysis according to claim 6 is characterized in that finding the midline of the training segmented image as the training segmented midline according to the training segmented image and the second marking point comprises: If the training segmented image is a triangle, three triangle vertices are obtained; The angle containing the second marked point is taken as the second marked angle; The side corresponding to the second marked angle is taken as the second marked side.

[0081] Among them, the two edges that construct the split angle are regarded as connected edges, and the edges other than the adjacent edges are regarded as the edges corresponding to the split angle, that is, the split edges.

[0082] Take the midpoint of the second marked side as the first midpoint; Connect the first midpoint with the second marked point to obtain the training segmentation midline.

[0083] The algorithm and display provided herein are not inherently related to any particular computer, virtual system or other device. Various general purpose systems can also be used together with the teachings based on this. According to the above description, it is obvious that the structure required for constructing such systems. In addition, the present invention is not directed to any specific programming language either. It should be understood that various programming languages ​​can be utilized to realize the content of the present invention described herein, and the description of the above specific languages ​​is for disclosing the best mode of the present invention.

[0084] Similarly, it should be understood that in order to streamline the present disclosure and aid in understanding one or more of the various inventive aspects, in the above description of exemplary embodiments of the present invention, various features of the present invention are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed method should not be interpreted as reflecting the intention that the claimed invention requires more features than those explicitly recited in each claim. More specifically, as reflected in the claims below, inventive aspects lie in less than all the features of the individual embodiments disclosed above. Therefore, the claims that follow the specific embodiment are hereby expressly incorporated into the specific embodiment, with each claim itself serving as a separate embodiment of the present invention.

Claims

1. A personalized exercise prescription recommendation system based on big data analysis, characterized in that: Including monitoring module and processor: The monitoring module is used to obtain user information; the user information includes the user's age, gender and symptoms; The processor is used to process the following method: Inputting the user information into a motion mode discrimination network to obtain a plurality of motion modes; the motion modes represent the types of motion suitable for the user; According to the motion mode, a user motion footprint image and a corresponding user motion posture image are obtained; the user motion footprint image represents a two-dimensional grayscale image including footprints left by the user during motion; the user motion posture image represents an image including the corresponding user motion posture when recording the user motion footprint image; Based on the user motion posture image and the user motion footprint image, a footprint support vector is obtained; the footprint support vector represents the pressure of the user's posture contained in the two-dimensional user motion posture image on the footprint; Through the segmentation discriminant network, based on the user's motion posture image, the pressure corresponding to each part of the user's motion footprint image caused by the posture is found, and multiple cutting angles are obtained; Based on the cutting angle and the user motion posture image, the user motion posture image is segmented at different angles to obtain a plurality of segmented images; the segmented images are images including the user motion footprint images; Through the posture discrimination network, based on the segmented image, it is judged whether the movement posture is correct, and a recommended movement prescription is obtained.

2. The personalized exercise prescription recommendation system based on big data analysis according to claim 1 is characterized in that: The segmentation discriminant network is used to find the pressure corresponding to each part of the user's motion footprint image caused by the posture based on the user's motion posture image, and obtain multiple cutting angles, including: Inputting the user motion posture image into an edge detection network to detect the user posture to obtain a first posture image; the first posture image represents the user's body posture; Inputting the first posture image into a body part discrimination network to obtain a user body part position image; the user body part position image is an image marking the positions of multiple body parts of the user; The user's part position image is input into the segmentation discriminant network to obtain multiple cutting angles.

3. The personalized exercise prescription recommendation system based on big data analysis according to claim 2 is characterized in that: The training method of the segmentation discriminant network comprises: Acquire a plurality of training images; the training images are historical user body position images; Obtaining a footprint support vector corresponding to the training image as annotation data; Inputting the training image into a segmentation discriminant network to obtain a training cutting angle; Based on the training cutting angle and the annotation data, the training image is cut to obtain a plurality of training segmented images and corresponding second marking points; the training segmented image is an image cut according to the training cutting angle; the second marking point represents the midpoint of an edge of the first segmented image on the annotated basic straight line; Based on the training segmented image and the second marking point, the degree of force is determined with the midline as a reference, and a plurality of training segmented grayscale total values ​​and corresponding training segmented distances are obtained; the training segmented distance represents a distance from the ground in the training segmented image; the training segmented grayscale total value represents the pressure of each posture in the training segmented image at the training segmented distance; Based on multiple training segmentation grayscale total values ​​and corresponding training segmentation distances, a training pressure value is obtained; multiple training segmentation grayscale total values ​​correspond to one training pressure value; multiple training segmentation images correspond to multiple training pressure values; the training pressure value represents the pressure exerted by the entire training segmentation image on the ground through the footprints; Dividing the multiple training pressure values ​​by the sum of the multiple training pressure values, respectively, to obtain a training pressure ratio; the training pressure ratio represents the ratio of the pressures exerted on the ground by the footprints of the multiple training segmented images; According to the training pressure ratio and the footprint support vector, the loss is calculated and the segmentation discriminant network is trained.

4. The personalized exercise prescription recommendation system based on big data analysis according to claim 1 is characterized in that: The obtaining of the footprint support vector based on the user motion posture image and the user motion footprint image includes: Taking the straight line where the user motion posture image and the user motion footprint image intersect as a basic straight line; Segmenting the user's motion footprint image in a direction perpendicular to the base straight line to obtain a plurality of first segmented images; Adding the grayscale values ​​in the first segmented images to obtain a footprint pressure value; obtaining a plurality of footprint pressure values ​​corresponding to the plurality of first segmented images; According to the positions of the plurality of first segmented images in the user's motion footprint image, the plurality of footprint pressure values ​​are arranged in sequence to obtain a footprint support vector.

5. The personalized exercise prescription recommendation system based on big data analysis according to claim 3 is characterized in that: The step of cutting the training image based on the training cutting angle and the annotation data to obtain a plurality of training cut images and corresponding second marking points includes: The first segmented image corresponding to the user running footprint image corresponding to the annotated data is used as the annotated first segmented image; Using the basic straight line corresponding to the annotated first segmented image as the annotated basic straight line; The first segmented image is marked by marking positions on the marking base straight line corresponding to positions on the training image to obtain a plurality of first marking points; The midpoint of two connected first marking points is used as the second marking point; Find the angle difference from the marked base straight line to train the cutting angle, and use the straight line through the second marked point as the cutting line; The training image is segmented using segmentation lines to obtain a training segmented image.

6. The personalized exercise prescription recommendation system based on big data analysis according to claim 3 is characterized in that: The method of determining the degree of force based on the training segmentation image and the second marking point with the midline as a reference, and obtaining a plurality of training segmentation grayscale total values ​​and corresponding training segmentation distances, comprises: According to the training segmentation image and the second marking point, finding the midline of the training segmentation image as the training segmentation midline; Based on the training segmentation image and the training segmentation center line, the pressure of the user's posture is projected onto the training segmentation center line to obtain a plurality of training segmentation grayscale total values; one value on the training segmentation center line corresponds to one training segmentation grayscale total value; Obtain the position of the value on the training segmentation midline on the training image as the training segmentation position; According to the training segmentation position and the position of the second marking point, the distance to the second marking point is calculated by the Pythagorean theorem to obtain the training segmentation distance; The training segmentation grayscale total value is associated with the corresponding training segmentation distance.

7. The personalized exercise prescription recommendation system based on big data analysis according to claim 6 is characterized in that: The method of projecting the pressure of the user's posture onto the training segmentation midline based on the training segmentation image and the training segmentation midline to obtain a plurality of training segmentation grayscale total values ​​includes: A point on the training segmentation midline is obtained as a segmentation point; multiple points on the training segmentation midline correspond to multiple segmentation points; A straight line passing through the segmentation point and perpendicular to the training segmentation midline is obtained as the segmentation bisector; multiple segmentation bisectors are obtained corresponding to multiple segmentation points; In the training segmentation image, the grayscale values ​​on the segmentation bisector are added to obtain a training segmentation total grayscale value; and a plurality of segmentation bisectors correspond to obtaining a plurality of training segmentation total grayscale values.

8. The personalized exercise prescription recommendation system based on big data analysis according to claim 3 is characterized in that: The step of obtaining a training pressure value based on a plurality of training segmentation grayscale total values ​​and corresponding training segmentation distances includes: Add multiple training segmentation distances to get the total predicted distance value; The training segmentation distance is divided by the total value of the predicted distance to obtain a predicted distance pressure value; multiple training segmentation distances correspond to multiple predicted distance pressure values; The training segmentation grayscale total value is multiplied by the corresponding predicted distance pressure value to obtain a training segmentation pressure value; multiple training segmentation grayscale total values ​​correspond to multiple training segmentation pressure values; Adding multiple training segmentation pressure values ​​together to obtain a training pressure value; the total grayscale values ​​of multiple training segmentations correspond to one training pressure value; Multiple training segmentation images correspond to multiple training pressure values.

9. The personalized exercise prescription recommendation system based on big data analysis according to claim 6, characterized in that: The step of finding the midline of the training segmentation image as the training segmentation midline according to the training segmentation image and the second marking point includes: If the training segmented image is a quadrilateral, four quadrilateral vertices are obtained; Connect two non-adjacent four-sided vertices to obtain two four-sided line segments; The point where two four-sided line segments intersect is taken as the first focus; The second marking point corresponding to the training segmentation image is connected to the first key point to obtain the training segmentation centerline.

10. The personalized exercise prescription recommendation system based on big data analysis according to claim 6, characterized in that: The step of finding the midline of the training segmentation image as the training segmentation midline according to the training segmentation image and the second marking point includes: If the training segmented image is a triangle, three triangle vertices are obtained; The angle containing the second marked point is taken as the second marked angle; The side corresponding to the second marked angle is used as the second marked side; Take the midpoint of the second marked side as the first midpoint; Connect the first midpoint with the second marked point to obtain the training segmentation midline.

Images