WO2021179230A1

WO2021179230A1 - Scoliosis detection model generating method and computer device

Info

Publication number: WO2021179230A1
Application number: PCT/CN2020/078899
Authority: WO
Inventors: 安丰伟; 刘展志
Original assignee: 南方科技大学
Priority date: 2020-03-12
Filing date: 2020-03-12
Publication date: 2021-09-16

Abstract

A scoliosis detection model generating method and a computer device. The scoliosis detection model generating method comprises: acquiring training data, and inputting the training data into an initial neural network, so as to obtain a prediction result corresponding to the training data, the training data comprising advancing parameters of a training sample at multiple moments in an advancing process and a real result corresponding to the training sample; according to the real result and the prediction result, adjusting parameters of the initial neural network, and continuing to execute the step of inputting the training data into the initial neural network so as to obtain a prediction result, until a preset training condition is satisfied, so as to obtain a trained scoliosis detection model. The training data comprises advancing parameters of a training sample at multiple moments in an advancing process, and is dynamic data. The dynamic data in the advancing process is used to train the initial neural network, so that a more accurate detection result can be obtained by means of a trained scoliosis detection model in actual use.

Description

Method and computer equipment for generating scoliosis detection model

Technical field

This application relates to the field of detection technology, and in particular to a method and computer equipment for generating a scoliosis detection model.

Background technique

Adolescent scoliosis is a common and frequently-occurring disease that endangers adolescents and children in my country, and the incidence is still increasing year by year. In 2015, a survey for junior high school students showed that 9.9% of students suffer from scoliosis. It is 3% of the world average. Therefore, the detection of scoliosis is particularly important.

The severity of scoliosis is mostly assessed by measuring the contralateral bending angle, and the most common method of angle measurement is the Cobb angle measurement method. At present, the image of the test person's spine bone is mainly obtained through X-ray irradiation, and the angle of the test person's Cobb angle is obtained through image recognition and calculation. This is the feature for classification through deep learning. The disadvantage is that all obtained static information is information analysis. The amount is small, and the accuracy needs to be improved.

Therefore, the existing technology needs to be improved.

Summary of the invention

The technical problem to be solved by the present invention is to provide a method for generating a scoliosis detection model and computer equipment, so that the scoliosis detection model obtained by training can obtain more accurate detection results.

In the first aspect, an embodiment of the present invention provides a method for generating a scoliosis detection model, and the method includes:

Obtain training data, and input the training data into the initial neural network to obtain the prediction result corresponding to the training data, wherein the training data includes the traveling parameters of a training sample at multiple times during the traveling process and the training sample Corresponding real results;

According to the real result and the prediction result, adjust the parameters of the initial neural network, and continue to execute the step of inputting the training data into the initial neural network to obtain the prediction result until the preset training conditions are met to obtain the Trained scoliosis detection model.

As a further improved technical solution, the traveling parameters include: the standing duration of the training sample during the traveling process, the arm length and leg length at each moment, and every two adjacent training samples during the traveling process The joint motion angle and travel speed at the moment; the acquiring training data includes:

Collect the spatial coordinates of the joint points of a training sample at each moment in the travel process;

According to the spatial coordinates of the joint points of the training sample at each moment in the marching process, the standing duration of the training sample in the marching process, the arm length and leg length at each moment are obtained, and the training sample is marching The joint motion angle and travel speed at every two adjacent moments in the process.

As a further improved technical solution, the collection of the spatial coordinates of the joint points at each moment in the traveling process includes:

At each moment during the travel of the training sample, the training sample is photographed by a three-dimensional camera to obtain a depth image of the training sample at each moment;

According to the depth information at each time, the space coordinates of the joint points at each time are obtained.

As a further improved technical solution, according to the spatial coordinates of the joint points of the training sample at each time during the traveling process, the standing duration of the training sample during the traveling process, and the arm length and leg length at each time are obtained, And the joint motion angles and travel speeds of the training sample at every two adjacent moments in the travel process include:

Obtaining the standing duration of the training sample during the traveling process according to the spatial coordinates of the joint points at each moment in the traveling process of the training sample;

For a moment in the traveling process, obtain the arm length and leg length of the training sample at that moment according to the spatial coordinates of the joint points of the training sample at that moment;

According to the spatial coordinates of the joint points of the training sample at this moment and the spatial coordinates of the joint points at the moment before this moment, the joint motion angles and travel speeds of two adjacent moments are obtained.

As a further improved technical solution, according to the spatial coordinates of the joint points of the training sample at each moment in the traveling process, the standing duration of the training sample during the traveling process is obtained, including:

Acquiring each duration when the space coordinate of the joint point is lower than a preset value;

The respective durations are added to obtain the standing duration.

As a further improved technical solution, the parameters of the initial neural network are adjusted according to the real result and the prediction result, and the step of inputting training data into the initial neural network to obtain the prediction result is continued until the prediction result is satisfied. Set training conditions to obtain the trained scoliosis detection model, including:

Calculating a loss value according to the real result and the predicted result;

Adjust the parameters of the initial neural network according to the loss value, and continue to perform the step of inputting training data into the initial neural network to obtain a prediction result until a preset training condition is met to obtain a trained scoliosis detection model.

In the second aspect, an embodiment of the present invention provides a method for detecting scoliosis, including:

Obtain test data corresponding to the object to be tested, where the test data includes the continuous standing time of the object to be tested in the process of traveling, the arm length and leg length at each moment, the joint motion angles at two adjacent moments, and Travel speed

The test data is input into a trained scoliosis detection model to obtain a detection result, wherein the detection result indicates the type of scoliosis and the degree of scoliosis of the object to be tested, and the trained spine side The curve detection model is a scoliosis detection model obtained by the above-mentioned method for generating a scoliosis detection model.

As a further improved technical solution, the acquiring test data corresponding to the object to be tested includes:

Collect the space coordinates of the joint points at each moment in the travel process;

Obtain the standing duration according to the space coordinates of the joint points at each moment in the travel process;

Obtain the arm length and leg length of the object to be measured at each time according to the spatial coordinates of the joint points at each time during the traveling process of the object to be measured;

According to the space coordinates of the joint points of the object to be measured at two adjacent moments in the traveling process, the joint motion angle and the traveling speed of the object to be measured at the two adjacent moments are obtained.

In a third aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, the memory stores a computer program, and the processor implements the following steps when the computer program is executed:

According to the real result and the prediction result, adjust the parameters of the initial neural network, and continue to execute the step of inputting the training data into the initial neural network to obtain the prediction result until the preset training conditions are met to obtain the Trained scoliosis detection model;

Or, obtain test data corresponding to the object to be tested, where the test data includes the continuous standing time of the object to be tested in the process of traveling, the arm length and leg length at each moment, and the joint movement at two adjacent moments. Angle and speed of travel;

In a fourth aspect, an embodiment of the present invention also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

Compared with the prior art, the embodiments of the present invention have the following advantages:

Obtain training data, and input the training data into the initial neural network to obtain the prediction result corresponding to the training data, wherein the training data includes the traveling parameters of a training sample at multiple times during the traveling process and the training sample Corresponding real results; according to the real results and the predicted results, adjust the parameters of the initial neural network, and continue to perform the step of inputting the training data into the initial neural network to obtain the predicted results, until the preset training is satisfied Condition to get a trained scoliosis detection model. The training data includes the travel parameters of a training sample at multiple times during the travel process, which is dynamic data. The initial neural network is trained through the dynamic data during the travel process, so that the trained scoliosis detection model is actually used. Can get more accurate test results.

Description of the drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only These are some embodiments described in the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work.

FIG. 1 is a schematic flowchart of a method for generating a scoliosis detection model in an embodiment of the present invention;

2 is a schematic diagram of calculating the angle of the shoulder joint at the first moment in an embodiment of the present invention;

FIG. 3 is a schematic flowchart of a method for detecting scoliosis in an embodiment of the present invention;

Figure 4 is a diagram of the internal structure of a computer device in an embodiment of the present invention.

Detailed ways

In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be described clearly and completely in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only These are a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.

The inventor found through research that the severity of the existing scoliosis is mostly evaluated by the measurement of the contralateral bending angle, and the Cobb angle measurement method is most commonly used for angle measurement. At present, the image of the test person’s spine bone is mainly obtained through X-ray irradiation or ultrasound, and the angle of the test person’s Cobb angle is obtained through image recognition and calculation. This is the feature for classification through deep learning. The disadvantage is that the cost of obtaining information is high, and All are static information, the amount of information analysis is small, and the accuracy needs to be improved.

In order to solve the above problems, in the embodiment of the present invention, a method for generating a scoliosis detection model is provided, training data is obtained, and the training data is input to an initial neural network to obtain a prediction result corresponding to the training data. The training data includes the traveling parameters of a training sample at multiple moments in the traveling process and the real result corresponding to the training sample; according to the real result and the prediction result, the parameters of the initial neural network are adjusted and the execution continues The step of inputting the training data into the initial neural network to obtain the prediction result until the preset training condition is satisfied to obtain the trained scoliosis detection model. The training data includes the travel parameters of a training sample at multiple times during the travel process, which is dynamic data. The initial neural network is trained through the dynamic data during the travel process, so that the trained scoliosis detection model is actually used. Can get more accurate test results.

In the following, various non-limiting embodiments of the present invention will be described in detail with reference to the accompanying drawings.

Please refer to FIG. 1, which shows a method for generating a scoliosis detection model in an embodiment of the present invention, and the method includes:

S1. Obtain training data, and input the training data into the initial neural network to obtain the prediction result corresponding to the training data, where the training data includes the travel parameters of a training sample at multiple times during the travel process and the training sample Corresponding real results.

In the embodiment of the present invention, the training sample is a human, there are multiple training samples, the training data corresponds to one training sample, and the traveling process is the process of the training sample walking within a preset time. The training data is obtained from the state at each moment during the journey. The traveling parameters are dynamic parameters of a person during the traveling process, and the traveling parameters include: the standing duration of the training sample during the traveling process, the arm length and leg length at each moment, and the training sample during the traveling process The joint motion angle and travel speed at every two adjacent moments in the.

Specifically, in step S1, the acquiring training data includes:

S11. Collect the space coordinates of the joint points of a training sample at each moment in the traveling process.

In the embodiment of the present invention, when training data is collected, for a training sample, the training sample is allowed to start walking. You can choose to walk on a device with a conveyor belt. The traveling process lasts for a preset time, and the preset time is Custom settings, for example, can be set to one minute, or half a minute. Collect the space coordinates of the joint points at each moment in the travel process during the travel process.

Specifically, collecting the space coordinates of the joint points at each moment in the travel process in step S11 includes:

S111. At each time during the travel of the training sample, the training sample is photographed by a three-dimensional camera to obtain a depth image of the training sample at each time.

In the embodiment of the present invention, the depth image of the training sample during the travel is captured by a three-dimensional stereo camera, and the interval time between the two moments can be customized. The shorter the interval time, the more abundant training data is obtained. The interval time can be set to 1/12 second, that is, one frame is an interval time, and one depth image is taken for each frame.

S112. Obtain the space coordinates of the joint points at each time according to the depth image at each time.

In the embodiment of the present invention, the space coordinates of the joint points at each moment are extracted by the depth sensor. For a depth image at a moment, extract the depth information (spatial position) corresponding to each joint point in the depth image according to the pixel point corresponding to each joint point in the depth image, and express the depth information in the form of spatial coordinate points (X, Y, Z), That is, the space coordinates of the joint points are obtained.

S12. According to the spatial coordinates of the joint points of the training sample at each time during the traveling process, obtain the standing duration of the training sample during the traveling process, the arm length and leg length at each time, and the training sample The joint motion angle and travel speed at every two adjacent moments in the travel process.

Specifically, step S12 includes:

S121: Obtain the standing duration of the training sample during the traveling process according to the spatial coordinates of the joint points at each moment in the traveling process of the training sample.

In the embodiment of the present invention, the spatial coordinates of the joint points are expressed in the form of (X, Y, Z). In actual use, a point in the traveling process is set as the origin. For example, when the designated training sample has a conveyor belt When walking on the device, use a vertex of the conveyor belt on the device as the origin. The Z value of the space coordinate of the joint point (representing the height of the joint point from the origin) can be used to obtain the standing duration of the training sample during travel.

Specifically, step S121 includes:

S1211. Obtain each duration when the space coordinate of the joint point is lower than a preset value.

In the embodiment of the present invention, the preset value is a value of Z. For example, the preset value can be set to 5cm or 10cm. During travel, when it is detected that the Z value of the space coordinate of the joint point is lower than the preset value Start timing at, until it is detected that the Z value of the space coordinate of the joint point is higher than the preset value, and a duration is obtained. During the traveling process, the duration of the space coordinates of all related nodes below the preset value is recorded.

S1212. Add the respective durations to obtain the standing duration.

In the embodiment of the present invention, the durations for which the spatial coordinates of the joint points are lower than the preset value are added to obtain the standing duration of the training sample during the traveling process.

In another implementation manner, the standing duration of the training sample during travel can also be obtained by measuring the plantar pressure. Specifically, the plantar pressure is detected by a pressure sensor to obtain the plantar pressure of the training sample during the traveling process, and the plantar pressure can be detected by a wearable pressure sensor.

If the plantar pressure is greater than the threshold, it means that you are standing. When the plantar pressure is less than the threshold, it means that your feet leave the ground; when the plantar pressure is detected to be greater than the threshold, start timing until the plantar pressure is less than the threshold. A duration is obtained. During the process, record the duration of all plantar pressures greater than the threshold. The duration of each plantar pressure greater than the threshold is added to obtain the standing duration of the training sample during the marching process.

In another implementation, the sole of the training sample can be divided into several regions of interest (such as toes, soles, etc.), and the instantaneous average pressure value of each region can be calculated and recorded. The pressure is distinguished by the color depth, and the plantar pressure distribution is reflected on the display screen in real time. The displayed chromatogram image can be used to confirm whether the data collection process is smooth, so that when invalid data is collected, the posture can be adjusted or recollected in time.

S122. For a moment in the traveling process, obtain the arm length and leg length of the training sample at that moment according to the spatial coordinates of the joint points of the training sample at that moment.

In the embodiment of the present invention, for a moment in the traveling process, the length of the forearm can be calculated by using the distance formula between the two points according to the space coordinates of the wrist joint and the elbow joint; the length of the forearm can be calculated according to the space coordinates of the elbow joint and the shoulder joint Boom length, calculate the boom length based on the forearm length and the boom length. Similarly, the calf length is calculated according to the spatial coordinates of the ankle joint and knee joint; and the thigh length is calculated according to the spatial coordinates of the knee joint and the iliac joint, and the leg length is obtained according to the calf length and the thigh length.

S123: Obtain joint motion angles and travel speeds at two adjacent moments according to the spatial coordinates of the joint points of the training sample at the moment and the spatial coordinates of the joint points at the moment before the moment.

In the embodiment of the present invention, the joint motion angle includes the shoulder joint motion angle. See FIG. 2. Specifically, for a moment in the traveling process, this moment is regarded as the first moment, and the moment before this moment is regarded as the second moment. Time; Obtain the spatial coordinates of the elbow joint, shoulder joint and cervical spine in the depth image corresponding to the first moment, calculate the straight-line distance AC from the elbow joint to the shoulder joint according to the spatial coordinates of the elbow joint and shoulder joint, and calculate the linear distance AC from the elbow joint to the shoulder joint according to the spatial coordinates of the shoulder joint and cervical spine Calculate the straight-line distance BC from the shoulder joint to the cervical spine, and calculate the straight-line distance AB from the elbow joint to the cervical spine according to the spatial coordinates of the elbow joint and the cervical spine; take AC, BC and AB into the law of cosines to calculate the cosine value of the shoulder joint; The trigonometric function obtains the angle a of the shoulder joint at the first moment. For the second moment, the same method is used to calculate the angle of the shoulder joint corresponding to the second moment. According to the angle of the shoulder joint at the first moment, the angle of the shoulder joint at the second moment, and the time interval between the first moment and the second moment, Obtain the joint motion angles of the training sample at two adjacent moments.

In the embodiment of the present invention, for the first moment, the spatial coordinates of any joint at the first moment are acquired, and the spatial coordinates of any joint are acquired to calculate the distance between the same joint at two adjacent moments to obtain training The displacement of the sample in a time interval, according to the time interval between the first moment and the second moment and the displacement, obtain the travel speed of the training sample at two adjacent moments. Obviously, since the starting time in the traveling process does not have the previous time, the first time cannot be the starting time in the traveling process.

In the embodiment of the present invention, there are multiple sets of training data involved in the training, and the multiple sets of training data correspond to multiple different real results. The real results are the real results of the training sample scoliosis, and the real results represent the training samples. The type of scoliosis and the degree of scoliosis can be marked with real marks for different real results. For example, the real result of a training sample is: no scoliosis, its real mark is T-0; a training sample The true result of is: mild chest curvature, its true identification is T-1; the true result of a training sample is: moderate chest curvature, its true identification is T-2. The corresponding relationship between the specific real mark and the real result can be set according to actual needs.

In the embodiment of the present invention, the initial neural network predicts the result of the scoliosis of the training sample according to the training data, which is the prediction result, which is equivalent to the answer that the initial neural network solves according to the training data, and the real result is the standard Answer. The prediction result is the result output by the initial neural network according to the training data, and the output result may be a prediction label. For example, the training data DATA-1 is input to the initial neural network, and the initial neural network outputs M-2, which identifies the initial neural network prediction The prediction result corresponding to DATA-1 is the result corresponding to M-2, that is, moderate chest curvature.

S2. Adjust the parameters of the initial neural network according to the real result and the prediction result, and continue to perform the step of inputting the training data into the initial neural network to obtain the prediction result until the preset training conditions are met, Obtain the trained scoliosis detection model.

In the embodiment of the present invention, the loss value is calculated according to the real result and the predicted result, and then the parameters of the initial neural network are adjusted according to the loss value. Specifically, step S2 includes:

S21: Calculate a loss value according to the real result and the predicted result.

In the embodiment of the present invention, an existing loss function is used to calculate the loss value, for example, a cross-entropy loss function is used to calculate the loss value.

S22. Adjust the parameters of the initial neural network according to the loss value, and continue to execute the step of inputting the training data into the initial neural network to obtain a prediction result until a preset training condition is met to obtain the trained spine side Bend detection model.

In the embodiment of the present invention, it is assumed that the parameters of the initial neural network are that the loss value is propagated back to modify the parameters of the initial neural network to obtain the modified parameters.

In the embodiment of the present invention, after the parameters are modified, the step of inputting training data into the initial neural network is continued until a preset training condition is met, where the preset training condition includes that the loss value meets the preset requirement or the number of training times reaches the preset number of times . The preset requirement may be determined according to the trained scoliosis detection model, which will not be described in detail here. The preset number of times may be the maximum number of training times of the initial neural network, for example, 50,000 times. Therefore, after the loss value is calculated, it is determined whether the loss value meets the preset requirements, if the loss value meets the preset requirements, the training is ended, and if the loss value does not meet the preset requirements, the initial neural network is determined Whether the number of training times reaches the number of training times, if it does not reach the preset number of times, the parameters of the initial neural network are adjusted according to the loss value, and if the number of times reaches the preset number, the training ends, so that the loss function value and the number of training times are used to determine Judging whether the initial neural network training is over can prevent the initial neural network from entering an endless loop because the loss function value cannot meet the preset requirements.

Further, because the parameters of the initial neural network are modified when the training condition of the initial neural network does not meet the preset conditions, after the parameters of the initial neural network are corrected according to the loss value, the initial neural network needs to be continuously trained.

Wherein, the execution continues to input the training data into the initial neural network, and the input training data may be training data that has not been input to the initial neural network. For example, all training data in the training data has a unique identification. The identification of the training data input to the initial neural network for the first training is different from the identification of the training data input to the initial neural network for the second training. The identification of the training data of the neural network is 1, the identification of the training data input to the initial neural network during the second training is 2, and the identification of the training data input to the initial neural network during the Nth training is N, when all training data is input After the initial neural network, the training data that has participated in the training can be repeatedly input, so that the training data is input to the initial neural network in cycles. In this embodiment, the specific implementation manner of "continue to execute and input the training data into the initial neural network" is not limited.

In the prior art, for the tester, only one static photo of the tester is taken, and the scoliosis result of the tester is obtained by analyzing the static photo; only one static photo of the tester requires too little data to be analyzed. Lead to inaccurate test results. In the embodiment of the present invention, the data of the training sample during the travel process is used as the training data (dynamic data), the training data is more comprehensive, and the scoliosis model obtained by training can obtain more accurate detection results.

Referring to FIG. 3, an embodiment of the present invention also provides a method for detecting scoliosis, including:

K1. Obtain test data corresponding to the object to be tested, where the test data includes the continuous standing time of the object to be tested in the process of traveling, the arm length and leg length at each moment, and the joint movement at two adjacent moments. Angle and speed of travel.

In the embodiment of the present invention, the object to be tested is a person. To test a person's scoliosis, first collect test data of the object to be tested. Specifically, step K1 includes:

K11. Collect the space coordinates of the joint points of the object to be measured at each moment in the traveling process.

In the embodiment of the present invention, when the test data is collected, the object to be tested may start walking, and may choose to walk on a device with a conveyor belt. The traveling process lasts for a preset time, and the preset time may be set to one minute. Start to collect the space coordinates of the joint points at each moment during the journey.

In the embodiment of the present invention, the depth image of the object to be measured is captured by a three-dimensional stereo camera during the traveling process, and the interval time between the two moments can be customized. The shorter the interval time, the more abundant training data is obtained. The interval time can be set to 1/12 second, that is, one frame is an interval time, and one depth image is taken for each frame. The space coordinates of the joint points at each moment are extracted through the depth image. For the depth image at a time, the depth information (spatial position) corresponding to each joint point in the depth image is extracted according to the pixel point corresponding to each joint point in the depth image, and the depth information is The form of spatial coordinate points (X, Y, Z) is expressed, that is, the spatial coordinates of each joint point of the object to be measured are obtained.

K12. Obtain the standing duration of the collected object according to the space coordinates of the joint points at each moment in the traveling process.

In the embodiment of the present invention, the spatial coordinates of the joint points are expressed in the form of (X, Y, Z). In actual use, a point in the traveling process is set as the origin. For example, when the object to be measured is specified When walking on a conveyor belt device, use a vertex of the conveyor belt on the device as the origin. The Z value of the space coordinate of the joint point can be used to obtain the standing duration of the object to be measured during the traveling process.

In the embodiment of the present invention, the preset value is a value of Z. For example, the preset value can be set to 5 cm or any value between 5 cm and 10 cm. During travel, when the spatial coordinates of the joint points are detected Start timing when the Z value of is lower than the preset value, until it is detected that the Z value of the space coordinate of the joint point is higher than the preset value, and a duration is obtained. During the travel, the spatial coordinates of all the related nodes are recorded below the preset value. The duration of each value. The duration of the space coordinates of the joint points lower than the preset value is added to obtain the standing duration of the object to be tested during the traveling process.

In another implementation manner, the standing duration of the object to be measured during travel can also be obtained by measuring the plantar pressure. Specifically, the plantar pressure is detected by a pressure sensor to obtain the plantar pressure of the sample to be tested during traveling, and the plantar pressure can be detected by a wearable pressure sensor.

If the plantar pressure is greater than the threshold, it means that you are standing. When the plantar pressure is less than the threshold, it means that your feet leave the ground; when the plantar pressure is detected to be greater than the threshold, start timing until the plantar pressure is less than the threshold. A duration is obtained. During the process, record the duration of all plantar pressures greater than the threshold. The duration of each plantar pressure greater than the threshold is added to obtain the standing duration of the object to be tested during travel.

K13. Obtain the arm length and leg length of the object to be measured at each time according to the space coordinates of the joint points at each time during the traveling process of the object to be measured.

In the embodiment of the present invention, for each moment, according to the space coordinates of the wrist and elbow joints, the length of the forearm can be calculated by using the distance formula between the two points; the length of the forearm can be calculated according to the space coordinates of the elbow and shoulder joints, Calculate the arm length based on the forearm length and the boom length. Similarly, the calf length is calculated according to the spatial coordinates of the ankle joint and knee joint; and the thigh length is calculated according to the spatial coordinates of the knee joint and the iliac joint, and the leg length is obtained according to the calf length and the thigh length.

K14. According to the space coordinates of the joint points of the object to be measured at two adjacent moments in the traveling process, the joint motion angle and the traveling speed of the object to be measured at two adjacent moments are obtained.

In the embodiment of the present invention, the joint motion angle includes the shoulder joint motion angle. Specifically, for the first moment of two adjacent moments, the space of the elbow joint, shoulder joint, and cervical spine in the depth image corresponding to the first moment is acquired Coordinates, calculate the straight-line distance from the elbow joint to the shoulder joint according to the spatial coordinates of the elbow and shoulder joints, calculate the straight-line distance from the shoulder joint to the cervical spine according to the spatial coordinates of the shoulder joint and the cervical spine, and calculate the elbow joint according to the spatial coordinates of the elbow joint and the cervical spine The straight-line distance to the cervical spine; the straight-line distance from the elbow joint to the shoulder joint, the straight-line distance from the shoulder joint to the cervical spine and the straight-line distance from the elbow joint to the cervical spine are brought into the theorem of cosines to calculate the cosine value of the shoulder joint; The angle of the shoulder joint at a moment.

Similarly, for the second moment of the two adjacent moments, the angle of the shoulder joint corresponding to the second moment is calculated, based on the angle of the shoulder joint at the first moment, the angle of the shoulder joint at the second moment, and the first moment and the second moment. The time interval between two moments is used to obtain the joint motion angles of the object to be measured at two adjacent moments.

In the embodiment of the present invention, for the first moment of two adjacent moments, the space coordinates of any joint at the first moment are obtained, and at the second moment of the two adjacent moments, the space of any joint is obtained The coordinates are used to calculate the distance between the same joint at two adjacent moments to obtain the displacement of the training sample in a time interval. According to the time interval between the first moment and the second moment and the displacement, the object to be tested is obtained in two phases. The speed of travel at the next moment.

In the embodiment of the present invention, the dynamic test data of the object to be tested during the traveling process is obtained according to the above steps.

K2. Input the test data into a trained scoliosis detection model to obtain a detection result, where the detection result indicates the type and degree of scoliosis of the object to be tested, and the trained The scoliosis detection model is a trained scoliosis detection model obtained by the above-mentioned method for generating a scoliosis detection model.

In the embodiment of the present invention, the test data is input into the trained scoliosis detection model, the detection result can be obtained, and the scoliosis type and the degree of the scoliosis of the object to be tested can be obtained conveniently during medical diagnosis, for example, through The trained scoliosis detection model obtains the corresponding detection result of the test data: moderate chest curvature.

In the above-mentioned method for generating a scoliosis detection model, the present invention collects dynamic information, and the amount of information is much larger than traditional static information, and can collect multi-dimensional information that cannot be collected by static information, for example: the object to be tested is in the process of traveling The plantar pressure information, the arm length and leg length at each moment, the joint motion angle and travel speed of two adjacent moments provide conditions for improving the accuracy of judgment. The scoliosis model obtained from this training is used in actual detection In scoliosis, more accurate detection results can be obtained, and the performance is better.

In one embodiment, the present invention provides a computer device, which may be a terminal, and the internal structure is shown in FIG. 4. The computer equipment includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide calculation and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for generating a scoliosis detection model is realized. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, or it can be a button, trackball or touchpad set on the housing of the computer equipment , It can also be an external keyboard, touchpad, or mouse.

Those skilled in the art can understand that the block diagram shown in FIG. 4 is only a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer equipment to which the solution of the present application is applied. The specific computer equipment may include a diagram More or fewer components are shown in, or some components are combined, or have different component arrangements.

The embodiment of the present invention provides a computer device, including a memory and a processor, the memory stores a computer program, and is characterized in that the processor implements the following steps when the computer program is executed:

The embodiment of the present invention also provides a computer-readable storage medium on which a computer program is stored, and is characterized in that, when the computer program is executed by a processor, the following steps are implemented:

The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be It is considered as the range described in this specification.

The above-mentioned embodiments only express several implementation manners of the present application, and their descriptions are relatively specific and detailed, but they should not be understood as limiting the scope of invention patents. It should be noted that for those of ordinary skill in the art, without departing from the concept of this application, several modifications and improvements can be made, and these all fall within the protection scope of this application. Therefore, the scope of protection of the patent of this application shall be subject to the appended claims.

Claims

A method for generating a scoliosis detection model, characterized in that the method includes:

Obtain training data, and input the training data into the initial neural network to obtain the prediction result corresponding to the training data, wherein the training data includes the traveling parameters of a training sample at multiple times during the traveling process and the training sample Corresponding real results;

According to the real result and the prediction result, adjust the parameters of the initial neural network, and continue to execute the step of inputting the training data into the initial neural network to obtain the prediction result until the preset training conditions are met to obtain the Trained scoliosis detection model.
The method according to claim 1, wherein the traveling parameters include: standing duration of the training sample during the traveling process, arm length and leg length at each moment, and the training sample during the traveling process The joint motion angle and travel speed at every two adjacent moments in each of the two adjacent moments; the acquiring training data includes:

Collect the spatial coordinates of the joint points of a training sample at each moment in the travel process;

According to the spatial coordinates of the joint points of the training sample at each moment in the marching process, the standing duration of the training sample in the marching process, the arm length and leg length at each moment are obtained, and the training sample is marching The joint motion angle and travel speed at every two adjacent moments in the process.
The method according to claim 2, wherein the collecting the spatial coordinates of the joint points at each moment in the traveling process comprises:

At each moment during the travel of the training sample, the training sample is photographed by a three-dimensional camera to obtain a depth image of the training sample at each moment;

According to the depth information at each time, the space coordinates of the joint points at each time are obtained.
The method according to claim 2, characterized in that, according to the spatial coordinates of the joint points of the training sample at each time during the traveling process, the standing duration of the training sample during the traveling process and the standing time of the training sample at each time are obtained. The arm length and leg length, as well as the joint motion angle and travel speed at every two adjacent moments during the travel of the training sample, include:

Obtaining the standing duration of the training sample during the traveling process according to the spatial coordinates of the joint points at each moment in the traveling process of the training sample;

For a moment in the traveling process, obtain the arm length and leg length of the training sample at that moment according to the spatial coordinates of the joint points of the training sample at that moment;

According to the spatial coordinates of the joint points of the training sample at this moment and the spatial coordinates of the key nodes at the moment before this moment, the joint motion angles and travel speeds of two adjacent moments are obtained.
The method according to claim 4, wherein the obtaining the standing duration of the training sample during the traveling process according to the spatial coordinates of the joint points at each moment during the traveling process of the training sample comprises:

Acquiring each duration when the space coordinate of the joint point is lower than a preset value;

The respective durations are added to obtain the standing duration.
The method according to claim 1, wherein the parameters of the initial neural network are adjusted according to the real result and the prediction result, and the execution of inputting training data into the initial neural network is continued to obtain the prediction result The steps until the preset training conditions are met to obtain the trained scoliosis detection model, including:

Calculating a loss value according to the real result and the predicted result;

Adjust the parameters of the initial neural network according to the loss value, and continue to perform the step of inputting training data into the initial neural network to obtain a prediction result until a preset training condition is met to obtain a trained scoliosis detection model.
A method for detecting scoliosis, characterized in that the method includes:

Obtain test data corresponding to the object to be tested, where the test data includes the continuous standing time of the object to be tested in the process of traveling, the arm length and leg length at each moment, the joint motion angles at two adjacent moments, and Travel speed

The test data is input into a trained scoliosis detection model to obtain a detection result, wherein the detection result indicates the scoliosis type and degree of the scoliosis of the object to be tested, and the trained spine side The bending detection model is the scoliosis detection model according to any one of claims 1 to 6.
The method according to claim 7, wherein said obtaining test data corresponding to the object to be tested comprises:

Collecting the space coordinates of the joint points of the object to be measured at each moment in the traveling process;

Obtain the standing duration according to the space coordinates of the joint points at each moment in the travel process;

Obtain the arm length and leg length of the object to be measured at each time according to the spatial coordinates of the joint points at each time during the traveling process of the object to be measured;

According to the space coordinates of the joint points of the object to be measured at two adjacent moments in the traveling process, the joint motion angle and the traveling speed of the object to be measured at the two adjacent moments are obtained.
A computer device, comprising a memory and a processor, the memory storing a computer program, wherein the processor implements the spinal side according to any one of claims 1 to 6 when the computer program is executed. A method for generating a curve detection model, or the steps of a method for detecting scoliosis according to any one of claims 7 to 8.
A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, realizes the generation of a scoliosis detection model according to any one of claims 1 to 6 Method, or a step of a method for detecting scoliosis according to any one of claims 7 to 8.