A method for evaluating a level of a ski turn based on monocular video

Abstract

The application discloses a method for evaluating a skiing rotation level based on monocular video and belongs to the field of skiing rotation sports. The method is realized as follows: three-dimensional coordinates of each joint of a human body are captured, and kinematics inverse solution is performed to obtain posture angles of each segment of the human body; each segment centroid coordinate is obtained by using the posture angles of each segment of the human body and human anatomy parameters, the human body centroid coordinate is obtained by segment mass weighting calculation, and kinematics parameters are derived. Key evaluation indexes are constructed by analyzing kinematics parameters collected by monocular video, and the key evaluation indexes are obtained in combination with structural characteristics of three-dimensional coordinates of each joint collected by monocular video, skiing rotation movement characteristics and skiing rotation dynamics model characteristics. According to the human body kinematics parameters obtained by monocular video collection and analysis, the key evaluation indexes are used to evaluate the skiing rotation level, and the evaluation precision and efficiency of the skiing rotation level are improved. The application can be further applied to analysis, guidance evaluation and improvement of the skiing rotation movement of skiers.

Description

TECHNICAL FIELD

[0001] The application relates to a method for evaluating a skiing turn level based on monocular video and belongs to the field of skiing turn sports. BACKGROUND

[0002] As a main item in alpine skiing competition, many researchers hope to analyze the level of the turn action of athletes to analyze the sports performance of athletes and prevent injuries. In early research, researchers captured kinematic data of athletes during the turn process by placing a multi-view PTZ (Pan-Tilt-Zoom) camera to perform stereoscopic imaging. Although such a device can accurately collect human posture data, there is great complexity in the installation of the device and the processing of the collected data. In addition, the capture range of such a device is also limited, allowing only a part of the sliding sequence to be captured. With the continuous development of MEMS (Micro-Electro-Mechanical System) and GNSS (Global Navigation Satellite System) technologies and the miniaturization of products, recent sports analysis of alpine skiing prefers to use these technologies. Although the application of these technologies significantly reduces the difficulty of deploying the motion capture device, it will cause certain interference to the athlete's movement process, and due to the drift problem of the inertial sensor itself, it cannot work for a long time. In recent years, with the rapid development of GPU (Graphics Processing Unit) and deep learning technologies, researchers have gradually focused on extracting human posture through monocular camera video, which will greatly reduce the difficulty of device deployment and will not affect the athlete's movement. In the process of kinematic analysis of human kinematic data captured by a monocular camera, it is necessary to scientifically define the index and method for evaluating the level of skiing turns. SUMMARY

[0003] The main purpose of the application is to provide a method for evaluating the level of skiing turns based on monocular video, which uses a monocular camera to capture the three-dimensional coordinates of each joint of the human body and performs kinematic inverse solution to obtain the posture angle of each segment of the human body; using the posture angle of each segment of the human body and the anatomical parameters of the human body to obtain the centroid coordinates of each segment, and further calculating the centroid coordinates of the human body by segment mass weighting, and obtaining the kinematic parameters according to the centroid coordinates of the human body. The key evaluation index for evaluating the level of skiing turns based on monocular video is constructed by analyzing the kinematic parameters collected by the monocular video, and the key evaluation index is obtained by combining the structural characteristics of the three-dimensional coordinates of each joint collected by the monocular video, the movement characteristics of the skiing turn and the characteristics of the skiing turn dynamics model. According to the kinematic parameters of the human body obtained by analyzing the monocular video, the key evaluation index is used to evaluate the level of skiing turns, thereby improving the evaluation accuracy and efficiency of the level of skiing turns. The application can be further applied to the analysis, guidance and evaluation of skiing turn sports, and the movement level of skiers is improved.

[0004] The purpose of the application is achieved by the following technical solutions.

[0005] The application discloses a method for evaluating a skiing turn level based on monocular video, and comprises the following steps.

[0006] Step one: defining a human joint center and a segment coordinate system, and defining a human segment mass distribution based on the human joint center and the segment coordinate system and in combination with a human body segment inertia parameter, wherein the human segment mass distribution is used to calculate a human center of mass CoM coordinate in a subsequent step.

[0007] The human joint center and the segment coordinate system are defined. The joint center is divided into six points, including a neck joint center CJC, a shoulder joint center SJC, an elbow joint center EJC, a waist joint center LJC, a hip joint center HJC and a knee joint center KJC. In particular, LSJC and RSJC represent a left shoulder joint center and a right shoulder joint center respectively, and LEJC, REJC, LKJC, RKJC and LKJC represent a left elbow joint center, a right elbow joint center, a left hip joint center, a right hip joint center and a left knee joint center respectively. The body segments are divided by the defined joint center, and the human segment mass distribution is defined based on the body segment inertia parameter BSIPs and the defined segment coordinate system.

[0008] Preferably, in step one, the human segment mass distribution and the center of mass spatial position suitable for a male are defined based on the body segment inertia parameter BSIPs, as shown in Table 1.

[0009] Table 1: Human (male) segment mass distribution and center of mass spatial position

[0010]

[0011] Step two: acquiring motion video data of an athlete in a skiing turn training through a monocular camera, constructing a semi-supervised time domain convolution network, and inputting the motion video data into the semi-supervised time domain convolution network to estimate human joint three-dimensional coordinates, wherein the human joint three-dimensional coordinates are coordinates relative to a hip joint (origin), that is, the human joint spatial coordinates in a turn process are acquired.

[0012] The motion data of athletes in ski turn training is collected by monocular camera, and a semi-supervised time domain convolution network is constructed to estimate the three-dimensional coordinates of each joint of the human body. The node labeling follows the labeling specification of the Human3.6M human posture dataset, and the spatial coordinate system is defined by the intersection of the sagittal plane, coronal plane and transverse plane. The intersection of the sagittal plane and the coronal plane is defined as the Y axis, and the origin is positive downward. The intersection of the coronal plane and the transverse plane is defined as the X axis, and the origin is positive left. The intersection of the sagittal plane and the transverse plane is defined as the Z axis, and the origin is positive backward. The three-dimensional coordinates of the joint nodes obtained by the convolution network are the coordinates relative to the origin hip joint. The spatial coordinates of the joints of the human body during the turn are collected.

[0013] Step three: using the spatial coordinates of the joints of the human body collected in step two during the turn, the kinematic inverse solution is performed to obtain the attitude angle of each segment of the human body, i.e. using the position vector from the centroid of each segment to the origin of each segment to obtain the attitude angle matrix of each segment, using the joint center of the human body defined in step one and the segment coordinate system to expand the formula for obtaining the attitude angle matrix of each segment into two categories, respectively, the segments with CJC, LJC, HJC, KJC as the origin and the segments with EJC, SJC as the origin.

[0014] Using the spatial coordinates of the joints of the human body collected in step two during the turn, the kinematic inverse solution is performed to obtain the attitude angle of each segment of the human body, i.e. using the position vector from the centroid of each segment to the origin of each segment to obtain the attitude angle matrix of each segment, using the joint center of the human body defined in step one and the segment coordinate system to expand the formula for obtaining the attitude angle matrix of each segment into two categories, respectively, the segments with CJC, LJC, HJC, KJC as the origin and the segments with EJC, SJC as the origin.

[0015] The position of the human body centroid during the athlete's sliding process depends on the attitude of each segment of the human body. In order to obtain the segment attitude matrix R S , the attitude inverse solution of the attitude matrix is needed using three-dimensional spatial coordinates, R S is defined as follows:

[0016]

[0017] Where α, β, θ represent the angles of rotation around the X, Y, Z axes respectively, R S is solved by formula (1):

[0018]

[0019] Where is the unit direction vector of the segment, is the unit direction vector of the Y axis of the anatomical coordinate system of each segment, is the vector of the origin of the anatomical coordinate system of each segment relative to the origin of the global coordinate system. is the vector of the end point of each segment relative to the origin of the global coordinate system, ||d ot || is the absolute value of the scaled length of each segment in the video.

[0020] According to the coordinate system of different segments, formula (1) is expanded into formula (2) and formula (3), p x , p y , p z is X component, Y component, Z component of the position vector.

[0021]

[0022]

[0023] Step four: according to the segment attitude matrix obtained in step three, the mass distribution and the spatial position of the center of mass of each segment are calculated according to the human anatomy parameters, and the center of mass coordinates of each segment of the human body are calculated, and the center of mass coordinates of each segment of the human body are further calculated by segment mass weighting.

[0024] According to the segment attitude matrix shown in formula (2) and formula (3), the spatial position of the center of mass of each segment is calculated as follows:

[0025]

[0026]

[0027]

[0028]

[0029]

[0030]

[0031]

[0032]

[0033]

[0034]

[0035] where d a→brepresents the distance of a→b, where a can be CJC, LSJC, LEJC, RSJC, REJC, LHJC, LKJC, RHJC, RKJC, and b can be head centroid, torso centroid, left upper arm centroid, left lower arm centroid, right upper arm centroid, right lower arm centroid, left upper leg centroid, left lower leg centroid, right upper leg centroid, and right lower leg centroid. The centroid position of each segment is obtained by represents the centroid position of s segment, where s segment can be head, torso, left upper arm, left lower arm, right upper arm, right lower arm, left upper leg, left lower leg, right upper leg centroid, and right lower leg, p NJC represents the NJC joint center, and the value of N is the same as a. It can be further obtained by segment mass m s The human body centroid position is obtained by weighted calculation The calculation is as follows:

[0036]

[0037] Step five: according to the human body centroid coordinates obtained in step four, a human body centroid displacement curve is drawn, and the ski turning kinematics parameters are obtained by using the human body centroid displacement curve.

[0038] The human body centroid coordinates obtained in step four are separated into three variables according to the X axis, Y axis, and Z axis, and the human body centroid X axis, Y axis, and Z axis coordinates in continuous time are drawn to obtain a human body centroid displacement curve. The curve changes periodically according to the turning process. In each period, the maximum value, minimum value, and difference between the maximum value and minimum value of the X axis, Y axis, and Z axis centroid coordinates are selected as the ski turning kinematics parameters.

[0039] Step six: key evaluation indexes for evaluating the ski turning level based on monocular video are constructed by analyzing the kinematics parameters obtained by video acquisition, and the key evaluation indexes are obtained by combining the monocular video acquisition joint three-dimensional coordinate structure characteristics, ski turning motion characteristics, and ski turning dynamics model characteristics.

[0040] In order to further analyze the changes in the three axes, the arithmetic center CoF of the two-foot space position is defined to represent the snowboard motion, and the motion curve of the X axis thereof is taken as the human body motion reference line. By introducing the CoF curve, the turning time sequence can be better analyzed. By combining the monocular video acquisition joint three-dimensional coordinate structure characteristics, ski turning motion characteristics, and ski turning dynamics model characteristics, key evaluation indexes for evaluating the ski turning level based on monocular video are constructed by analyzing the kinematics parameters obtained by monocular video acquisition. The key evaluation indexes include turn switch TS, forward inclination average FM, vertical drop VD, and symmetry ratio SR, VR, and FR.

[0041] The key evaluation index turn switch TS is calculated as follows: according to the characteristics of the skiing turning movement and the three-dimensional coordinate structure of each joint collected by the monocular video, the human body is projected onto the XOZ plane, and the start and end of the turning are defined, that is, the intersection of the X-axis curve of the CoF and the X-axis curve of the CoM is the turning switch TS. The turning switch is divided into left turning and right turning, which respectively represent the next turning direction as left turning and right turning.

[0042] The key evaluation index forward mean FM is calculated as follows: in the process of skiing turning, the forward direction of the human body is analyzed, and the kinematics of the center of mass is applied to the monocular video evaluation of the skiing turning level system, that is, the human body is projected onto the XOZ plane to define the forward inclination to represent the average forward value of left turning and right turning in a turning period. The calculation value of the forward inclination is the average value of the CoM in the forward direction, that is, the Z-axis direction in each turning period (divided by the turning switch)

[0043]

[0044] Where f1 and f2 represent the forward inclination peak value of the center of mass during left and right turning, respectively.

[0045] The key evaluation index vertical drop VD is calculated as follows: the vertical drop of the human body center of mass has a great influence on skiing performance during skiing turning, and the difference Δy between the maximum and minimum values of the center of mass in the Y-axis direction in a turning period is defined as the vertical drop.

[0046] The key evaluation index symmetry ratio SR, VR and FR is calculated as follows: the complete turning process is divided into left and right turning by the turning switch. Due to different sliding strategies and muscle strength of athletes, there are different symmetry ratios of left and right turning in the turning process of different athletes. The motion performance parameters related to symmetry ratio have evaluation and guiding significance for the training of athletes. The parameters are turning ratio SR, that is, the length ratio of left and right turning according to TS; vertical ratio VR, that is, the peak value ratio of left and right turning in a single turning period; and forward inclination ratio FR, that is, the peak value ratio of left and right turning in a single turning period.

[0047] Step seven: according to the human kinematics parameters obtained by monocular video collection and analysis, the key evaluation index is used to evaluate the skiing turning level, so as to improve the evaluation accuracy and efficiency of the skiing turning level.

[0048] The human body center of mass space position calculated through steps one to four and the skiing rotation kinematics parameters obtained in step five are used to evaluate the skiing rotation level by using the key evaluation index of the skiing rotation level based on the monocular video constructed in step six, so that the skiing rotation level evaluation precision and efficiency are improved: the higher the sports level of the athlete, the better the active force control, when the human body center of mass dynamics state of the X axis and the Y axis remains unchanged, the smaller the FM, the greater the active force effect of the human body on the Z axis; when the human body center of mass dynamics state of the X axis and the Z axis remains unchanged, the greater the VD, the greater the active force effect of the human body on the Y axis. The higher the sports level of the athlete, the better the active force control. The smaller the FR, the better the active force effect of the human body on the Z axis; the greater the VR, the better the active force effect of the human body on the Y axis; in the skiing strength training, SR, VR and FR should be as close to 1 as possible, at this time, the left and right rotation training effects in the rotation process are symmetrical, and the strength training is balanced.

[0049] Further comprising step eight: the evaluation results obtained in step seven are applied to the analysis and guidance evaluation of the skiing rotation movement, and the sports level of the skier is improved.

[0050] Beneficial effects:

[0051] 1. The method for evaluating the skiing rotation level based on the monocular video disclosed in the application uses a monocular camera to capture the three-dimensional coordinates of each joint of the human body and performs kinematics inverse solution to obtain the posture angle of each segment of the human body; the center of mass coordinates of each segment are obtained by using the posture angle of each segment of the human body and the human body anatomy parameters, and the center of mass coordinates of the human body are further obtained by segment mass weighting calculation; and the kinematics parameters are obtained according to the center of mass coordinates of the human body. The key evaluation index for evaluating the skiing rotation level based on the monocular video is constructed by analyzing the kinematics parameters collected by the monocular video, and the key evaluation index is obtained by combining the structural characteristics of the three-dimensional coordinates of each joint collected by the monocular video, the skiing rotation movement characteristics and the skiing rotation dynamics model characteristics. The key evaluation index is used to evaluate the skiing rotation level according to the human body kinematics parameters obtained by analyzing the monocular video, so as to improve the skiing rotation level evaluation precision and efficiency.

[0052] 2. The method for evaluating the skiing rotation level based on the monocular video disclosed in the application draws a human body center of mass displacement curve by using the human body center of mass coordinates in continuous time, and obtains skiing rotation kinematics parameters by using the human body center of mass displacement curve. The key evaluation index for evaluating the skiing rotation level based on the monocular video is constructed by combining the skiing rotation kinematics parameters, the structural characteristics of the three-dimensional coordinates of each joint collected by the monocular video, the skiing rotation movement characteristics and the skiing rotation dynamics model characteristics. The key index can measure the human body active force control ability of the skiing athlete in different axial directions, and is used to evaluate the skiing rotation level; the key index can measure the rotation training effect balance and force balance of the skiing athlete, and is used to improve the rotation training effect.

[0053] 3. The method for evaluating the level of ski turning based on monocular video can be further applied to the analysis, guidance evaluation and improvement of ski turning movement. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 A flowchart of the method for evaluating the level of ski turning based on monocular video.

[0055] Figure 2 A joint center and a segment coordinate system of a human body;

[0056] Figure 3 An indoor ski simulation movement collection system based on a monocular camera;

[0057] Figure 4 A spatial coordinate system of a human body;

[0058] Figure 5 A spatial position of a center of mass of a human body;

[0059] Figure 6 A CoM curve with a CoF as a reference curve;

[0060] Figure 7 TS, FM and VD in the representation of the kinematic curve of the center of mass when doing turning movement;

[0061] Figure 8 SR, VR and FR in the representation of the kinematic curve of the center of mass when doing turning movement;

[0062] Figure 9 A turning movement physical model of an athlete on an indoor ski simulation platform. DETAILED DESCRIPTION

[0063] In order to better illustrate the purposes and advantages of the present application, the content of the application is further described below in combination with the drawings and examples.

[0064] In order to verify the feasibility of the method, the embodiment selects to collect movement data in the ski turning process on an indoor ski simulation platform (Germany SkyTech Sport ski & fit).

[0065] The method for evaluating the level of ski turning based on monocular video disclosed in the embodiment specifically implements the following steps:

[0066] Step 1: defining a joint center and a segment coordinate system of a human body, defining the mass distribution of each segment of the human body on the basis of the joint center and the segment coordinate system of the human body in combination with the inertial parameters of the body segment, and using the mass distribution of each segment of the human body for subsequent steps to obtain the center of mass CoM coordinate of the human body.

[0067] Define the joint center and segment coordinate system of human joints. The joint center is divided into six points: cervical joint center (CJC), shoulder joint center (SJC), elbow joint center (EJC), waist joint center (LJC), hip joint center (HJC) and knee joint center (KJC). In particular, LSJC and RSJC are used to represent the left shoulder joint center and the right shoulder joint center respectively. Similarly, LEJC, REJC, LHJC, RHCJ, LKJC and RKJC represent the left elbow joint center, the right elbow joint center, the left hip joint center, the right hip joint center, the left knee joint center and the right knee joint center respectively. The coordinate systems of each joint point are shown in the attached figure. Figure 2 As shown in Figure 2, the body segments are segmented by the defined joint center, and the mass distribution of each segment is defined based on the body segment inertial parameters (BSIPs) and the defined segment coordinate system.

[0068] In step 1, based on the body segment inertia parameters (BSIPs), the mass distribution and center of mass spatial position of each human body segment applicable to males are defined as shown in Table 1.

[0069] Table 1 Mass distribution and center of mass spatial position of each segment of the human body (male)

[0070]

[0071] Step 2: Use a monocular camera to collect motion video data of athletes in ski slalom training, construct a semi-supervised time-domain convolutional network, and input the motion video data into the semi-supervised time-domain convolutional network to estimate the three-dimensional coordinates of each human joint. The three-dimensional coordinates of each human joint are the coordinates relative to the hip joint (origin), that is, the spatial coordinates of the human joints during the rotation process are collected.

[0072] The overall system is constructed as shown in the attached Figure 3 As shown in the figure, a monocular camera is used to collect the slalom training data of athletes on the indoor simulated skiing platform Germany SkyTechSport ski&fit, and a semi-supervised time-domain convolutional network is constructed to estimate the three-dimensional coordinates of each human joint. The node annotation follows the annotation specifications of the human posture dataset Human3.6M, refer to the attached Figure 4 The spatial coordinate system is defined by the intersection of the sagittal, frontal, and transverse planes in the human T-pose. The intersection of the sagittal and frontal planes is defined as the Y-axis, with the origin pointing downward. The intersection of the frontal and transverse planes is defined as the X-axis, with the origin pointing leftward. The intersection of the sagittal and transverse planes is defined as the Z-axis, with the origin pointing backward. The 3D coordinates of the joint points obtained by the convolutional network are the coordinates of the hip joint relative to the origin. The spatial coordinates of the human joints are collected during the rotation process.

[0073] Step 3: Use the spatial coordinates of the human joints collected in the rotation process in step 2 to perform kinematic inverse solution to obtain the posture angles of each human segment, that is, use the position vector from the center of mass of each segment to the origin of each segment to obtain the posture angle matrix of each segment. Use the human joint center and segment coordinate system defined in step 1 to expand the formula for obtaining the posture angle matrix of each segment into two categories, namely, using formula (2) for the segments with CJC, LJC, HJC, KJC as the origin and using formula (3) for the segments with EJC and SJC as the origin.

[0074] The position of the center of mass of the athlete during sliding depends on the posture of each segment of the human body. In order to obtain the segment posture matrix R S , it is necessary to use the three-dimensional space coordinates to perform the inverse solution of the attitude matrix, R S The definition is as follows:

[0075] Among them, α, β, and θ represent the angles of rotation around the X axis, Y axis, and Z axis respectively, and R S Solve by formula (1):

[0076]

[0077] in is the unit direction vector of the segment, For attachment Figure 2 The unit direction vector of the Y axis of the anatomical coordinate system of each segment, The origin of the anatomical coordinate system of each segment is relative to the attached Figure 4 The vector of the origin of the global coordinate system in , The end points of each segment are relative to the attached Figure 4 The vector of the origin of the global coordinate system in ||d ot || is the absolute value of the scaled length of each segment in the video.

[0078] According to the coordinate system of different segments, formula (1) is expanded into formula (2) and formula (3), p x 、p y 、p z for The X, Y, and Z components of the position vector.

[0079]

[0080]

[0081] Step 4: Based on the posture matrix of each segment obtained in step 3, the mass distribution of each segment and the spatial position of the center of mass of each segment of the human body are calculated using the human anatomical parameters. The center of mass coordinates of the human body are further calculated by segment mass weighting based on the center of mass coordinates of each segment of the human body.

[0082] According to the segment posture matrix shown in formula (2) and formula (3), the spatial position of the segment centroid is calculated as follows:

[0083]

[0084]

[0085]

[0086]

[0087]

[0088]

[0089]

[0090]

[0091]

[0092]

[0093] wherein d a→b represents the distance from a to b, wherein a can be CJC, LSJC, LEJC, RSJC, REJC, LHJC, LKJC, RHJC, RKJC, and b can be the head centroid, the torso centroid, the left upper arm centroid, the left lower arm centroid, the right upper arm centroid, the right lower arm centroid, the left upper leg centroid, the left lower leg centroid, the right upper leg centroid, and the right lower leg centroid. By the segment centroid position, represents the centroid position of the s segment, wherein the s segment can be the head, the torso, the left upper arm, the left lower arm, the right upper arm, the right lower arm, the left upper leg, the left lower leg, the right upper leg centroid, and the right lower leg, and p NJC represents the NJC joint center, and the value of NJC is the same as a. Further, the segment mass m s The human body centroid position is calculated by weighting calculation is calculated as follows:

[0094]

[0095] Figure 5 is the human body centroid spatial position in the application example.

[0096] Step five: according to the human body centroid coordinates obtained in step four, a human body centroid displacement curve is drawn, and the ski turning kinematics parameters are obtained by using the human body centroid displacement curve.

[0097] The human body center of mass coordinates obtained in step four are separated into three variables according to the X axis, the Y axis and the Z axis, and the human body center of mass X axis, Y axis and Z axis coordinates in continuous time are drawn to obtain the human body center of mass displacement curve. The curve periodically changes according to the rotation process. In each period, the maximum value, the minimum value and the difference between the maximum value and the minimum value of the X axis, Y axis and Z axis center of mass coordinates are selected as the kinematic parameters of the skiing rotation.

[0098] Step six: key evaluation indexes for evaluating the skiing rotation level based on monocular video are constructed by analyzing the kinematic parameters obtained through video acquisition. The key evaluation indexes are obtained by combining the monocular video acquisition joint three-dimensional coordinate structure characteristics, the skiing rotation movement characteristics and the skiing rotation dynamics model characteristics.

[0099] In order to further analyze the changes in the three axes, the arithmetic center CoF of the biped space position is defined to represent the snowboard movement, and the X axis movement curve thereof is taken as the human body movement reference line. By introducing the CoF curve, the rotation timing can be better analyzed. Figure 6 The CoM curve after the introduction of CoF as a reference in the application example. By combining the monocular video acquisition joint three-dimensional coordinate structure characteristics, the skiing rotation movement characteristics and the skiing rotation dynamics model characteristics, key evaluation indexes for evaluating the skiing rotation level based on monocular video are constructed by analyzing the kinematic parameters obtained through video acquisition. The key evaluation indexes include turn switch TS, forward inclination average FM, vertical drop VD and symmetry ratio SR, VR and FR.

[0100] The key evaluation index turn switch TS is calculated as follows: according to the skiing rotation movement characteristics and the monocular video acquisition joint three-dimensional coordinate structure characteristics, the human body is projected onto the XOZ plane and the start and end of the turn are defined, i.e. the intersection point of the X axis curve of CoF and the X axis curve of CoM is the turn switch TS. The turn switch is divided into left turn switch and right turn switch, which respectively represent the following rotation direction as left rotation and right rotation. In the application example, Figure 7 (a) is the turn switch in the rotation action, and the two intersection points are left rotation switch and right rotation switch, respectively.

[0101] The key evaluation index forward inclination average FM is calculated as follows: in the skiing rotation process, the human body forward direction center of mass kinematics is analyzed and applied to the monocular video evaluation skiing rotation level system, i.e. the human body is projected onto the XOZ plane to define the forward inclination to represent the average forward value of left rotation and right rotation in a rotation period. The calculated value of the forward inclination is the average value of the CoM in the forward direction, i.e. the Z axis direction, in each rotation period (divided by the turn switch)

[0102]

[0103] Where f1 and f2 represent the forward lean peak value of the center of mass when turning left and right respectively. In the application example, the forward lean peak value of the center of mass when turning left and right is shown in FIG. 2. Figure 7 (b) is the average forward shift value of left turn and right turn in a turn cycle.

[0104] The key evaluation index vertical drop VD is calculated as follows: The change of the vertical drop of the center of mass in the process of skiing turn has a great influence on skiing performance, and the maximum value and the minimum value difference Ay of the center of mass in the Y axis direction in a turn cycle is defined as the vertical drop. In the application example, the maximum value and the minimum value difference Ay of the center of mass in the Y axis direction in a turn cycle is shown in FIG. 3. Figure 7 (c) is the maximum value and the minimum value difference of the center of mass in the Y axis direction in a turn cycle.

[0105] The key evaluation index symmetry ratio SR, VR and FR is calculated as follows: The complete turn process is divided into left turn and right turn by the turn switch. Due to the different habits of sliding strategy and the unbalanced muscle strength of athletes, there are different situations of symmetry ratio of left turn and right turn in the turn process of different athletes. The performance parameters related to symmetry ratio have evaluation and guidance significance for the training of athletes. The parameters are turn ratio SR, which is the length ratio of left turn and right turn according to TS; vertical ratio VR, which is the peak value ratio of left turn and right turn in a single turn cycle; and forward lean ratio FR, which is the peak value ratio of left turn and right turn in a single turn cycle. In the application example, the turn ratio SR, the vertical ratio VR and the forward lean ratio FR in the kinematics curve of the center of mass when doing turn movement are shown in FIG. 4. Figure 8

[0106] Step seven: according to the human kinematics parameters obtained by monocular video acquisition analysis, the key evaluation index for evaluating the level of skiing turn is used to evaluate the level of skiing turn, so as to improve the evaluation accuracy and efficiency of the level of skiing turn.

[0107] The human body is simplified and modeled as shown in FIG. 1. Figure 9 The motion slide can generate a pulling force F T =k·α.

[0108]

[0109] Where the length of the rod is l s , θ and are the angle between the xoy plane and the projection of the x axis in the cylindrical coordinate system, and the resultant external force in the x, y and z directions of the human body is: ​The acceleration of the center of mass in three axial directions, respectively, is calculated according to the kinematic law. is the support force of the ground, is the active force of the human body.

[0110] The analytical expressions of θ and φ in the cylindrical coordinate system are obtained from the particle dynamics (formula) in the simulation of the skiing turning training process. The analytical expressions of θ and φ in the cylindrical coordinate system are obtained from the particle dynamics (formula) in the simulation of the skiing turning training process.

[0111]

[0112]

[0113]

[0114] The higher the sports level of the athletes, the better the active force control. According to formula 9, when the kinematic state of the center of mass of the human body in the X-axis and Y-axis remains unchanged, the smaller the FM, the greater the active force of the human body in the Z-axis; when the kinematic state of the center of mass of the human body in the X-axis and Z-axis remains unchanged, the greater the VD, the greater the active force of the human body in the Y-axis. The smaller the FR, the better the active force effect of the human body in the Z-axis; the greater the VR, the better the active force effect of the human body in the Y-axis. In the skiing strength training, SR, VR and FR should be as close to 1 as possible, at which time the training effect of the left and right turns in the turning process is symmetrical, and the strength training is balanced.

[0115] Step eight: apply the evaluation results obtained in step seven to the analysis, guidance and evaluation of the skiing turning movement, and improve the sports level of the skiers.

[0116] The above specific description further details the purpose, technical solution and beneficial effects of the application. It should be understood that the above description is only a specific embodiment of the application and is not used to limit the protection scope of the application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the application should be included in the protection scope of the application.

Claims

1. A method for evaluating ski slalom performance based on monocular video, characterized by: The following steps are included: Step 1: Define the human joint center and segment coordinate system. Based on the human joint center and segment coordinate system, combine the human body segment inertia parameters to define the mass distribution of each human body segment. The mass distribution of each human body segment is used to obtain the center of mass CoM coordinates of the human body in the subsequent steps. Step 2: Using a monocular camera to capture motion video data of athletes during ski slalom training, a semi-supervised temporal convolutional network is constructed. The motion video data is input into the semi-supervised temporal convolutional network to estimate the three-dimensional coordinates of each human joint. The three-dimensional coordinates of each human joint are relative to the hip joint, thus acquiring the spatial coordinates of the human joints during the slalom. Step 3: Using the human joint space coordinates collected in step 2 during the rotation process, perform kinematic inverse solution to obtain the posture angle of each human segment. That is, use the position vector from the center of mass of each segment to the origin of each segment to obtain the posture angle matrix of each segment. Use the human joint center and segment coordinate system defined in step 1 to expand the formula for obtaining the posture angle matrix of each segment into two categories, namely, the segments with CJC, LJC, HJC, KJC as the origin and the segments with EJC and SJC as the origin. Step 4: Based on the posture matrix of each segment obtained in step 3, the mass distribution of each segment and the spatial position of the center of mass of each segment of the human body are calculated using the human anatomical parameters, and the center of mass coordinates of the human body are further calculated by segment mass weighting based on the center of mass coordinates of each segment of the human body; Step 5: Based on the coordinates of the human body center of mass obtained in step 4, a human body center of mass displacement curve is plotted, and the kinematic parameters of the ski slalom are obtained using the human body center of mass displacement curve; Step 6: Analyze the kinematic parameters obtained from the video capture to construct key evaluation indicators for evaluating the ski slalom skill based on monocular video. The key evaluation indicators are obtained by combining the three-dimensional coordinate structure characteristics of each joint captured by the monocular video, the characteristics of the ski slalom motion, and the characteristics of the ski slalom dynamics model; Step 7: Based on the human kinematic parameters obtained from the monocular video acquisition and analysis, key evaluation indicators are used to evaluate the ski slalom level, thereby improving the accuracy and efficiency of the ski slalom level evaluation.

2. The method for evaluating ski slalom skill based on monocular video according to claim 1, characterized in that: The method further includes step eight, applying the evaluation results obtained in step seven to the analysis, guidance and evaluation of ski slalom sports to improve the sports level of skiers.

3. The method for assessing ski slalom skill based on monocular video according to claim 1 or 2, characterized in that: The implementation method of step one is: Define the joint centers and segment coordinate system of the human body; the joint centers are divided into six points: cervical joint center CJC, shoulder joint center SJC, elbow joint center EJC, waist joint center LJC, hip joint center HJC and knee joint center KJC, LSJC and RSJC are used to represent the left shoulder joint center and the right shoulder joint center respectively, LEJC, REJC, LHJC, RHCJ, LKJC, RKJC are used to represent the left elbow joint center, right elbow joint center, left hip joint center, right hip joint center, left knee joint center and right knee joint center respectively; the body segments are divided by the defined joint centers, and the mass distribution of each segment of the human body is defined based on the body segment inertia parameters BSIPs and the defined segment coordinate system.

4. The method for evaluating ski slalom skill based on monocular video according to claim 3, characterized in that: The implementation method of step 2 is: A monocular camera was used to collect motion data of athletes during ski slalom training, and a semi-supervised time-domain convolutional network was constructed to estimate the three-dimensional coordinates of each human joint. The node annotation followed the annotation specifications of the human posture dataset Human3.6M, and the spatial coordinate system was defined by the intersection of the sagittal plane, frontal plane, and transverse plane. The intersection of the sagittal and frontal planes was defined as the Y-axis, with the origin pointing downward as the positive direction; the intersection of the frontal plane and the transverse plane was defined as the X-axis, with the origin pointing left as the positive direction; and the intersection of the sagittal and transverse planes was defined as the Z-axis, with the origin pointing backward as the positive direction. The three-dimensional coordinates of the joint points obtained by the convolutional network were the coordinates of the hip joint relative to the origin. The spatial coordinates of the human joints during the rotation process were collected.

5. The method for evaluating ski slalom skill based on monocular video according to claim 4, characterized in that: The implementation method of step three is: Using the human joint space coordinates collected in step 2 during the rotation process, perform kinematic inverse solution to obtain the posture angles of each human segment, that is, use the position vector from the center of mass of each segment to the origin of each segment to obtain the posture angle matrix of each segment. Using the human joint center and segment coordinate system defined in step 1, the formula for obtaining the posture angle matrix of each segment is divided into two categories, namely, using formula (2) for segments with CJC, LJC, HJC, and KJC as the origin and using formula (3) for segments with EJC and SJC as the origin. The position of the center of mass of the athlete during sliding depends on the posture of each segment of the human body. In order to obtain the segment posture matrix R S , it is necessary to use the three-dimensional space coordinates to perform the inverse solution of the attitude matrix, R S The definition is as follows: Among them, α, β, and θ represent the angles of rotation around the X axis, Y axis, and Z axis respectively, and R S Solve by formula (1): in is the unit direction vector of the segment, is the unit direction vector of the Y axis of the anatomical coordinate system of each segment, is the vector of the origin of the anatomical coordinate system of each segment relative to the origin of the global coordinate system, is the vector of each segment end point relative to the origin of the global coordinate system, ‖d ot ‖ is the absolute value of the scaled length of each segment in the video; According to the coordinate system of different segments, formula (1) is expanded into formula (2) and formula (3), p x 、p y 、p z for X component, Y component, Z component of the position vector; 。 6. The method for evaluating ski slalom skill based on monocular video according to claim 5, characterized in that: The implementation method of step 4 is: According to the posture matrix of each segment shown in formula (2) and formula (3), the spatial position of the center of mass of each segment is calculated as follows: where d a→b Represents the distance from a to b, where a is CJC, LSJC, LEJC, RSJC, REJC, LHJC, LKJC, RHJC, or RKJC, and b is the center of mass of the head, trunk, left upper arm, left forearm, right upper arm, right forearm, left thigh, left calf, right thigh, and right calf; obtained by the center of mass of each segment, Indicates the center of mass position of the s segment, which is the head, trunk, left upper arm, left forearm, right upper arm, right forearm, left thigh, left calf, right thigh center of mass and right calf, p NJC Represents the joint center of the joint, and the value of NJC is the same as a; to further pass the segment mass m s Weighted calculation to obtain the center of mass of the human body The calculation is as follows: 。 7. The method for evaluating ski slalom skill based on monocular video according to claim 6, characterized in that: The implementation method of step five is: The center of mass coordinates obtained in step 4 are separated into three variables according to the X-axis, Y-axis, and Z-axis. The X-axis, Y-axis, and Z-axis coordinates of the center of mass of the human body are plotted over a continuous time to obtain a center of mass displacement curve. The curve changes periodically according to the rotation process. Within each cycle, the maximum value, minimum value, and difference between the maximum and minimum values ​​of the center of mass coordinates of the X-axis, Y-axis, and Z-axis are selected as the kinematic parameters of the ski slalom.

8. The method for evaluating ski slalom skill based on monocular video according to claim 7, characterized in that: The implementation method of step six is: To further analyze changes in the three axes, the arithmetic center of gravity (CoF) of the bipedal spatial position is defined to characterize snowboard motion, and its X-axis motion curve is used as a reference line for human motion. The introduction of the CoF curve enables better analysis of turn timing. Combining the three-dimensional coordinate structure characteristics of each joint captured by monocular video, the characteristics of ski slalom motion, and the characteristics of the ski slalom dynamics model, key evaluation indicators for assessing ski slalom performance based on monocular video are constructed by analyzing the kinematic parameters obtained from monocular video. These key evaluation indicators include turn switch TS, forward lean mean FM, vertical drop VD, and symmetry ratios SR, VR, and FR. The key evaluation indicator, Turn Switch (TS), is calculated as follows: Based on the characteristics of ski slalom motion and the three-dimensional coordinate structure of each joint captured by monocular video, the human body is projected onto the XOZ plane and the start and end of the turn are defined. The intersection of the X-axis curves of CoF and CoM is the Turn Switch (TS). The Turn Switch is divided into Left Turn Switch and Right Turn Switch, indicating the next turn direction will be left or right, respectively. The key evaluation indicator, the mean forward tilt value FM, is obtained as follows: During the ski slalom process, the kinematics of the center of mass in the forward direction of the human body is analyzed and applied to the monocular video to evaluate the ski slalom level system, that is, the human body is projected onto the XOZ plane to define the forward tilt value, which represents the average forward displacement of the left and right turns in one rotation cycle; the calculated value of the forward tilt value is the average value of the CoM in the forward direction, i.e., the Z-axis direction, in each rotation cycle (divided by the turn switch). Among them, f1 and f2 represent the forward tilt peaks of the center of mass when turning left and right respectively; The key evaluation indicator vertical drop VD is obtained as follows: the difference Δy between the maximum and minimum values ​​of the center of mass in the Y-axis direction within a rotation cycle is defined as the vertical drop; The key evaluation indicators, symmetry ratio SR, VR, and FR, are obtained as follows: the complete rotation process is divided into two parts, left rotation and right rotation, through the turning switch; due to the different sliding strategies and uneven muscle strength of athletes, different athletes have different symmetry ratios of left and right rotations during their rotations. The sports performance parameters related to the symmetry ratio have evaluation and guidance significance for the training of athletes; the relevant parameters are the rotation ratio SR, that is, the ratio of the left and right rotation durations divided according to TS; the vertical ratio VR, that is, the ratio of the left and right rotation descent peaks in a single rotation cycle; the forward tilt ratio FR, that is, the ratio of the left and right rotation forward tilt peaks in a single rotation cycle.

9. The method for evaluating ski slalom skill based on monocular video according to claim 8, characterized in that: The implementation method of step seven is: The spatial position of the human center of mass calculated through steps 1 to 4 and the kinematic parameters of the ski slalom obtained in step 5 are used to evaluate the ski slalom level using the key evaluation indicators for evaluating the ski slalom level based on monocular video constructed in step 6, thereby improving the accuracy and efficiency of the ski slalom level evaluation: the higher the athletic level, the better the active force control of the athlete. When the dynamic state of the human center of mass on the X and Y axes remains unchanged, the smaller the FM, the greater the active force effect of the human body on the Z axis. When assuming that the dynamic state of the human body's center of mass on the X and Z axes remains unchanged, the larger the VD is, the greater the active force of the human body on the Y axis; Athletes with higher athletic levels have better control over their active force; the smaller the FR, the better the effect of the human body's active force on the Z axis; the larger the VR, the better the effect of the human body's active force on the Y axis; in ski strength training, SR, VR and FR should be kept close to 1 as much as possible. At this time, the training effects of left and right turns during the turn are symmetrical, and the strength training is balanced.

10. The method for evaluating ski slalom skill based on monocular video according to claim 9, characterized in that: In step 1, based on the body segment inertia parameters BSIPs, the mass distribution and center of mass spatial position of each human body segment applicable to males are defined as shown in Table 1; Table 1 Mass distribution and center of mass spatial position of each segment of the human body 。

Images