Shore swimming motion evaluation method based on inertial sensor and optical motion capture system

Abstract

The invention belongs to the technical field of human motion measurement, and particularly relates to a shore swimming motion evaluation method based on an inertial sensor and an optical motion capture system. The method comprises two parts of swimming motion data acquisition and swimming motion evaluation. The motion data acquisition comprises motion scene setting and optical target spot and inertial sensor position selection; the equipment comprises a land-based swimming trainer, an optical motion capture system and a computer, calculating data measured by optical motion capture to obtain included angle time history data of the limbs at global coordinates; measuring Euler angle data of the limb by an inertial sensor, and calculating to obtain included angle time history data of the limb in global coordinates; and further calculating a Spearman correlation coefficient, an intra-class correlation coefficient and a normalized root-mean-square error between inertial sensor data and optical motion capture data of each limb under various swimming strokes, so as to evaluate the accuracy, effectiveness and precision of the measured data. According to the invention, the swimming motion can be reliably and efficiently measured, and a basis is provided for swimming training.

Description

Technical Field

[0001] This invention belongs to the field of human motion measurement technology, specifically a method for evaluating swimming movements on land. Background Technology

[0002] Swimming is a complex activity involving the coordinated movement of all joints in the body, with significant differences in movements between different swimming strokes. To effectively evaluate swimming technique and performance, precise and practical motion measurement methods are urgently needed. While existing underwater optical motion capture systems offer high accuracy, they suffer from drawbacks such as high equipment cost, complex deployment, susceptibility to water currents and obstructions, and are mostly offline, making real-time feedback difficult. In contrast, existing research indicates that inertial measurement units (IMUs) can be used to measure key indicators such as limb velocity, stroke time, and joint angles, demonstrating good feasibility and accuracy. However, most existing studies focus only on movements of specific body parts, such as installing IMUs in the lower limbs or sacral region to measure localized motion information. This distributed measurement method struggles to obtain complete data on the overall coordination of swimming movements, limiting a comprehensive analysis of the holistic nature and technical efficiency of swimming motion. Therefore, it is necessary to design a swimming motion measurement method based on inertial sensors, optical motion capture systems, and land-based swimming trainers to acquire whole-body motion data for breaststroke, freestyle, and butterfly strokes, and to propose corresponding evaluation methods for the validity, reliability, and accuracy of optical motion capture data as a benchmark. Summary of the Invention

[0003] The purpose of this invention is to provide a method for evaluating human swimming motion based on inertial sensors and optical motion capture systems, so as to overcome the defects of existing technologies.

[0004] The present invention provides a method for evaluating onshore swimming motion based on inertial sensors and optical motion capture (optical motion capture system), which specifically includes two parts: onshore swimming motion data acquisition and onshore swimming motion evaluation.

[0005] (a) Data acquisition of swimming motions on land:

[0006] (1) Setting up the sports scene, such as Figure 1 As shown, the equipment includes a land-based swimming trainer, several (e.g., 8-12) high-definition optical motion capture cameras (NOKOV), and a computer. The land-based swimming trainer is used to assist the test subject in performing various swimming strokes (such as breaststroke, freestyle, and butterfly, etc.). The high-definition optical motion capture cameras are arranged around the land-based swimming trainer to capture the motion trajectory information of target points on the test subject's body. The computer is used to acquire test data in real time.

[0007] (2) Selection of the placement positions of optical targets and inertial sensors on the test subject; optical targets are attached to the test subject's body, and inertial sensors are carried. The optical targets are arranged according to the Helen Hayes model, including the trunk, upper limbs, and lower limbs. Considering that the markers on the inner side of the lower limbs are easily obscured during actual measurement, and the positions of the knee and ankle joints can be determined by the remaining targets on the lower limbs, the number of targets is at least 25. The specific arrangement is as follows: Figure 3 As shown, the locations of the 25 target points are: top of head, front of head, back of head, left elbow, right shoulder, left elbow, right elbow, left wrist, right wrist, offset point (located below the right shoulder), left lower back, right lower back, upper left thigh, upper right thigh, left lateral knee, right lateral knee, left lateral malleolus, right lateral malleolus, left toes, right toes, left heel, right heel, left calf, right calf, and sacrum. At least 17 inertial sensors (Noitom PN series) are used, secured to the limbs with straps. Figure 4 As shown, the specific locations are: head, left scapula, right scapula, upper back, left upper arm, right upper arm, left forearm, right forearm, left hand, right hand, lower back, left thigh, right thigh, left calf, right calf, left foot, and right foot. See details. Figure 4 As shown

[0008] The optical motion capture camera acquires the motion trajectory data of the corresponding limb through optical target points and transmits it to the computer via a data cable; the inertial sensor is used to acquire the time history data of the Euler angles of the corresponding limb and transmits it via Bluetooth.

[0009] (II) Assessment of Swimming Movements on Land

[0010] The specific steps are as follows (e.g.) Figure 2 As shown):

[0011] Step 1: Based on the test environment set up above, the inertial motion capture sensor needs to be calibrated before the test.

[0012] Step 2: The test subject performs swimming stroke tests (such as breaststroke, freestyle, and butterfly, etc.) on a land-based swimming training device; set the stroke frequency (e.g., 40-80 strokes / minute) and stroke duration (e.g., 40-50 seconds) for different strokes, and rest for several minutes (3-5 minutes) between each set; for example, choose a stroke frequency of 40 strokes per minute and a duration of 45 seconds for breaststroke; a stroke frequency of 80 strokes per minute (upper limb movement frequency) and a duration of 45 seconds for freestyle; choose a stroke frequency of 50 strokes per minute and a duration of 45 seconds for butterfly, and rest for 3 minutes between each set.

[0013] Step 3: The data obtained by optical motion capture is used to obtain the time history data of the angle between each limb and the global coordinate system through the software system deployed in the computer, as well as the Euler angle data of each limb measured by the inertial sensor. Based on the rotation matrix formula and the coordinate transformation formula, the time history data of the angle between each limb and the global coordinate system are calculated.

[0014] Step 4: Align the optical motion capture data and inertial sensor data processed in Step 3. Specifically, for all swimming strokes, the left arm is used as the reference point to determine the time point corresponding to the first peak in the left arm optical motion capture data, which is then used as the time reference point. Next, time offset processing is performed on the inertial motion capture data to align the first peak in the inertial motion capture signal with the aforementioned reference point, achieving time synchronization.

[0015] Step 5: Based on the data processed in Step 4, calculate the Spearman correlation coefficient, intraclass correlation coefficient ICC(1,1), and normalized root mean square error (NRMSE) between the inertial sensor data and optical motion capture data (optical motion capture data will be used as standard data) of each limb under various swimming strokes, so as to evaluate the accuracy, validity and precision of the measured data.

[0016] Step 3 involves calculating the time history data of the angles between each body segment (a total of 17 segments, corresponding to 17 inertial sensors) and the global coordinates based on the rotation matrix formula and coordinate transformation formula. Here, the human node model is first further simplified into a simplified model, such as... Figure 5 As shown in the figure; the 20 points marked in the figure correspond to specific limbs through the line segments between adjacent points, and the specific correspondence is shown in Table 1.

[0017] Based on the simplified human body node model ( Figure 5 As shown in Table 1, first, a table listing the inertial sensor numbers and corresponding end node numbers for each body segment is presented. Additionally, the lengths of each body segment of the test subject need to be measured according to Table 1. Then, the angle between any body segment and the global coordinate system is calculated as follows:

[0018]

[0019] In the formula, θ k φ represents the rotation angle (roll angle) of the corresponding limb about the z-axis measured by the inertial sensor numbered k. k θ represents the rotation angle (roll angle) of the corresponding limb around the x-axis measured by the sensor numbered k. k This represents the rotation angle (roll angle) along the y-axis measured by sensor number k, where k is the inertial sensor number corresponding to the segment (values ​​range from 126 to 142). Based on this, the formula for calculating the rotation matrix of any segment is:

[0020]

[0021] q in the formula k Table 1 shows the node numbering vector for the body segment corresponding to inertial sensor k. It should be noted that in this invention, the simplified model of the test subject's abdomen is assumed to be a rigid body during the test. Therefore, nodes '0', '1', '5', and '9' are on the same rigid body, and thus the rotation matrices of the connecting line segments '0-9', '0-1', and '0-5' are all the same. Based on the homogeneous transformation matrix, we can calculate Figure 5 Global coordinates of each node on a simplified human body model. The selection of global coordinates is as follows: Figure 5 The lower left corner shows the North-Sky-East coordinate system, with the x-axis pointing north, the y-axis perpendicular to the ground and pointing upwards, and the z-axis pointing east. Before the calculation, for ease of description, we introduce the index vector f again, which represents the node numbers along the motion chain from the target node to the root node. The index vector corresponding to each node is listed in Table 2. For any node num (num takes values ​​from 1 to 19), its spatial position in the global coordinate system is calculated as follows:

[0022]

[0023] In the formula f num This represents the index vector corresponding to node num, as shown in Table 2, where n is the length of the index vector. global P num This represents the position vector of the target node in global coordinates. global P0 represents the position vector of node '0' in global coordinates, i.e., [0,0,0]. Indicates that node f num (i) points to f num The segment vectors of (i+1) are shown in Table 1. It's important to note that since both node number vectors q and f are vectors composed of node numbers, therefore for... and (if q) k (2)=f num (j), q k (1)=f num If (j+1)), then these two matrices are the same homogeneous transformation matrix.

[0024] Then, the angle between any segment (i.e., corresponding to any inertial sensor k) and the x-axis, y-axis, and z-axis of the global coordinate system.

[0025] The calculation formula is as follows:

[0026]

[0027] In the above formula, arccos is the inverse cosine function, · is the dot product operation between vectors, and || is the vector modulo operation (to calculate the length of a vector).

[0028] Table 1. Correspondence between test subject's body nodes and inertial sensors

[0029]

[0030] Table 2. Correspondence between test subject's body nodes and inertial sensors

[0031]

[0032]

[0033] Step 5 involves calculating the Spearman correlation coefficient, intra-class correlation coefficient (ICC(1,1)), and normalized root mean square error between the inertial sensor data and optical motion capture data for each limb under various swimming strokes.

[0034] The Spearman correlation coefficient is calculated as follows:

[0035]

[0036] In the formula, This represents the i-th data value in the time history data column corresponding to the segment of the inertial sensor with number k (k ranges from 126 to 142) and the γ-axis (values ​​x, y, or z), where n represents the total number of data points. This represents the average value of the corresponding inertial sensor data. This represents the i-th data value in the optical motion capture time history data column corresponding to the segment. This represents the corresponding average value. The closer the Spearman correlation coefficient is to 1, the more positively correlated the two sets of data are.

[0037] The reliability of inertial sensor data is evaluated by calculating the intraclass correlation coefficient ICC(1,1), as shown in the following formula.

[0038]

[0039] Similarly, in the formula and The values ​​represent the i-th data value in the time history data column of the inertial sensor and optical motion capture corresponding to the segment k and the γ-axis, respectively, and n represents the total number of data in the data column. and These represent the data column vectors of the inertial sensor and the optical motion capture, respectively. D() represents the variance of the calculated vectors. The closer the intraclass correlation coefficient (ICC(1,1)) is to 1, the higher the reliability of the inertial sensor data.

[0040] The accuracy of inertial sensor data is evaluated using the normalized root mean square error (NRMSE), calculated as follows:

[0041]

[0042] In the formula, and The definition is the same as in Formula 9. This represents the maximum value in the column vector of optical motion capture data for the corresponding segment. This represents the minimum value in the column vector of the corresponding segment optical motion capture data. The smaller the normalized root mean square error, the better the accuracy of the inertial sensor data.

[0043] The analysis results show that for the three swimming strokes, most body parts have good validity (Spearman correlation coefficient > 0.75), reliability (ICC > 0.75), and accuracy (NRMSE < 25%).

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] (1) This invention proposes a swimming motion measurement method and a complete testing scheme based on inertial sensors, an optical motion capture system, and a land-based swimming trainer. Compared with existing underwater swimming motion measurement methods based solely on optical motion capture systems, this method is lower in cost and can also improve the measurement accuracy of the motion. In addition, the arrangement of whole-body inertial sensors can also help us obtain motion data of all body segments.

[0046] (2) An inertial sensor data processing method is proposed, which obtains the angle data of the test subject's limb relative to the global coordinates through rotation matrix and coordinate transformation.

[0047] (3) An evaluation method for inertial sensor data was proposed, which uses optical motion capture data as a benchmark and calculates the Spearman correlation coefficient, linear regression determination coefficient, intra-class correlation coefficient and normalized root mean square error of inertial sensor data for verification. Attached Figure Description

[0048] Figure 1 A schematic diagram of the swimming motion test scenario.

[0049] Figure 2 Flowchart of testing and data evaluation methods.

[0050] Figure 3Schematic diagram of the location of optical target points on the human body.

[0051] Figure 4 Schematic diagram of the human body inertial sensor layout.

[0052] Figure 5 Human body node model diagram.

[0053] Figure 6 Data graph showing the alignment of the angle between the left upper arm and the z-axis in freestyle swimming.

[0054] Figure 7 Time history data of the angle between the left upper arm and the x-axis and the angle between the left thigh and the z-axis in freestyle swimming.

[0055] The diagram is labeled as follows: 1 is the test subject, 2 is the land-based swimming trainer, 3 is the first optical motion capture camera, 4 is the second optical motion capture camera, 5 is the third optical motion capture camera, 6 is the fourth optical motion capture camera, 7 is the fifth optical motion capture camera, 8 is the sixth optical motion capture camera, 9 is the seventh optical motion capture camera, 10 is the eighth optical motion capture camera, 11 is the ninth optical motion capture camera, 12 is the tenth optical motion capture camera, 13 is the computer, 101 is the first optical target (located in front of the head), 102 is the second optical target (located in the right shoulder), 103 is the third optical target (located in the left shoulder), 104 is the fourth optical target (located in the right elbow), and 105 is the fifth optical target (located in the left elbow). 106 is the sixth optical target point (located in the right forearm), 107 is the seventh optical target point (located in the left forearm), 108 is the eighth optical target point (located in the upper right thigh), 109 is the ninth optical target point (located in the upper left thigh), 110 is the tenth optical target point (located in the left lateral knee), 111 is the eleventh optical target point (located in the right lateral knee), 112 is the twelfth optical target point (located in the left calf), 113 is the thirteenth optical target point (located in the right calf), 114 is the fourteenth optical target point (located in the right lateral malleolus), 115 is the fifteenth optical target point (located in the left lateral malleolus), 116 is the sixteenth optical target point (right toe), 117 is the seventeenth optical target point (left toe), and 118 is the eighteenth optical target point. Point 119 is the 19th optical target point (located on the back of the head), point 120 is the 20th optical target point (located on the lower side of the right shoulder), point 121 is the 21st optical target point (located on the sacrum), point 122 is the 22nd optical target point (located on the left hand), point 123 is the 23rd optical target point (located on the right hand), point 124 is the 24th optical target point (located on the left heel), point 125 is the 25th optical target point (located on the right heel), point 126 is the first inertial sensor (located on the head), point 127 is the second inertial sensor (located on the left upper arm), point 128 is the third inertial sensor (located on the left forearm), point 129 is the fourth inertial sensor (located on the left hand), point 130 is the fifth inertial sensor (located on the right upper arm), and point 131 is the... Six inertial sensors (located in the right forearm), 132 is the seventh sensor (located in the right hand), 133 is the eighth inertial sensor (located in the left thigh), 134 is the ninth inertial sensor (located in the left calf), 135 is the tenth inertial sensor (located in the left foot), 136 is the eleventh inertial sensor (located in the right thigh), 137 is the twelfth inertial sensor (located in the right calf), 138 is the thirteenth inertial sensor (located in the right foot), 139 is the fourteenth inertial sensor (located in the left scapula), 140 is the fifteenth inertial sensor (located in the right scapula), 141 is the sixteenth inertial sensor (located in the upper back), 142 is the seventeenth inertial sensor (located in the lower back), and 143 is the shape of the inertial sensor. Detailed Implementation

[0056] The present invention will be further described below with reference to the embodiments and accompanying drawings. However, the present invention is not limited to the embodiments described below.

[0057] The dimensions and thicknesses of each component shown in the accompanying drawings are schematic and not intended to limit the invention. To make the illustrations clearer and show the mating relationships between the various components, some areas in the drawings have been appropriately scaled down, and the distances between the components have been increased or decreased.

[0058] The swimming assessment method proposed in this invention includes the following steps:

[0059] Step 1: Test environment setup and tester device donning. The test environment setup is as follows: Figure 1 As shown, the land-based swimming training device 2 is positioned in the center of the field, with ten optical motion capture cameras 3-12 (NOKOV) arranged around it. The field measures 11 meters long and 7 meters wide. A desktop computer 13 is used to receive and process data. Next, the lengths of each limb of the test subject need to be measured, as shown in Table 1 (limb lengths, i.e., the distance between the two ends of a segment). Test subject 1 needs to first... Figure 3 Complete the application of optical target points 101-125. The locations of the 25 target points are: top of head, front of head, back of head, left armpit, right shoulder, left elbow, right elbow, left wrist, right wrist, offset point (located below the right shoulder), left lower back, right lower back, upper left thigh, upper right thigh, left lateral knee, right lateral knee, left lateral malleolus, right lateral malleolus, left toes, right toes, left heel, right heel, left calf, right calf, and sacrum. Then, the tester needs to... Figure 4 The inertial sensors 126-142 were successfully donned and secured to the limbs using straps. A total of 17 inertial sensors were attached to the limbs using straps. Figure 4 As shown, the locations are head, left scapula, right scapula, upper back, left upper arm, right upper arm, left forearm, right forearm, left hand, right hand, lower back, left thigh, right thigh, left calf, right calf, left foot, and right foot.

[0060] After completing the donner, the inertial sensor is calibrated: After donning the inertial sensor, the tester needs to complete the following actions in sequence: 1) T-shaped movement: The tester needs to stand straight, extend both arms, perpendicular to the body's upward position, with palms facing down. 2) A-shaped movement: The tester needs to stand straight, with arms down, palms facing the body; arms should be as straight as possible downward, perpendicular to the ground; feet should be hip-width apart, keeping them parallel and upright. 3) W-shaped movement: After the A-shaped movement, walk slowly forward. 4) B-shaped movement: Bring both hands together in front of the body, with four fingers parallel to the ground, and the thumb and four fingers forming a 45-degree angle. After completing the actions, the software will automatically adjust the inertial sensor. It should be noted that the initial position of the inertial sensor is when the tester is in the T-shaped movement.

[0061] Step 2: The test subject performs breaststroke, freestyle, and butterfly strokes on a land-based swimming training device. For breaststroke, the stroke rate is 40 strokes per minute, with three sets of 45 seconds each, and a 3-minute rest between sets. For freestyle, the stroke rate is 80 strokes per minute (upper limb movement frequency), with each set lasting 45 seconds, and a 3-minute rest between sets. For butterfly, the stroke rate is 50 strokes per minute, with each set lasting 45 seconds, and a 3-minute rest between sets.

[0062] Step 3: Here, we'll take the left upper arm in a freestyle swimming stroke as an example to perform the corresponding data operations. First, we need to further simplify the human body model as follows: Figure 5 The simplified model shown in the figure has 20 points marked on it. The line segments between adjacent points correspond to specific limbs, as shown in Table 1. These 20 points also correspond to the following joints: '0' sacrum, '1' right hip joint, '2' right knee joint, '3' right ankle joint, '4' right toe, '5' left ankle joint, '6' left knee joint, '7' left ankle joint, '8' left toe, '9' first lumbar vertebra, '10' center point of the line connecting the left and right shoulder joints, '11' left shoulder joint, '12' left elbow joint, '13' left wrist joint, '14' left distal end of the hand, '15' right shoulder joint, '16' right elbow joint, '17' right wrist joint, '18' right distal end of the hand, '19' top of the head. It should be noted that node '0' is the set origin [0,0,0], i.e., the root node.

[0063] Taking the left upper arm in freestyle swimming as an example, its corresponding inertial sensor number is 127, and the two end nodes of the left upper arm are '12' and '11'. Assume that the Euler angle data measured at any time are: the rotation angle around the x-axis, i.e., the yaw angle φ; the rotation angle around the y-axis, i.e., the yaw pitch angle ψ; and the rotation angle around the z-axis, i.e., the roll angle θ. Then, according to the above formulas (1)-(3), the three Euler angles are respectively transposed according to the following formulas to obtain the corresponding rotation matrices.

[0064]

[0065]

[0066] Multiplying the three matrices together yields the homogeneous transformation matrix of the corresponding segment.

[0067] The superscript and subscript of the matrix are the node numbers at both ends of the body segment, respectively (for example, in this case, the node numbers are 12 and 11 for the upper left arm). Based on the homogeneous transformation matrix, the following can be calculated: Figure 5 The global coordinates of each node on the simplified human body model shown are as follows. The selection of global coordinates is as follows: Figure 5 The bottom left corner shows the North-East coordinate system, with the x-axis pointing north, the y-axis perpendicular to the ground and pointing upwards, and the z-axis pointing east.

[0068] Based on the homogeneous transformation matrix and formula (5) mentioned above, the method for calculating the coordinate positions of the two nodes on the upper left arm is based on the following formula.

[0069]

[0070] In the formula, f represents an index vector containing the node numbers of all nodes along the kinematic chain from the target node to the root node. For example, for the left forearm, the nodes to be determined are '12' and '11', so the corresponding index vectors are f and f, respectively. 11 =[0,9,10,11] and f 12 = [0, 9, 10, 11, 12]. n is the length of the corresponding vector f. global P 11 and global P 12 This represents the position vectors of target nodes '11' and '12' in global coordinates. global P0 represents the position vector of node '0' in global coordinates. This represents the segment vector pointing from node f(i) to f(i+1), as detailed in Table 1. Then, the angles between the upper left arm (inertial sensor number 127) and the x-axis, y-axis, and z-axis of the global coordinate system are...

[0071]

[0072] In the above formula, arccos is the inverse cosine function, · is the dot product operation between vectors, and || is the vector modulo operation (to calculate the length of a vector).

[0073] Step 4: Alignment of optical motion capture data with inertial sensor data. The inertial sensor data is processed according to Step 3 and converted into the angle between the body segment and the global coordinate system. The optical motion capture data can directly obtain the angle between the corresponding body segment and the global coordinate system using software. The alignment process is based on the angle between the upper left arm and the z-axis. Figure 6 The specific data alignment method is demonstrated. The left image shows the original data (the solid black line represents data obtained from the inertial sensor, and the dashed black line represents data obtained from optical motion capture). It can be seen that the time difference between the first peak (the dot in the image) of each set of data is 2.17 seconds. It's important to note that the first peak is defined here as not being less than 90% of the maximum data value, thus excluding data whose peaks have not yet stabilized within the first 3 seconds. Then, the entire sensor motion capture data is shifted 2.17 seconds in the negative x-axis direction to complete the data alignment operation. Figure 7 As shown in the right figure, for all limbs except the left upper arm, the data obtained from the inertial sensor needs to be shifted 2.17 seconds in the negative x-axis direction. It is important to note that the data obtained from each experiment requires corresponding alignment.

[0074] Step 5: After aligning the inertial sensor data and the optical motion capture data, the next step is to analyze the validity, reliability and accuracy of the inertial sensor data. Here, we continue to take the angle between the left upper arm and the z-axis in the freestyle swimming motion as an example (i.e., the inertial sensor number is 127).

[0075] Validity is determined by analyzing the Spearman correlation coefficient between the two sets of data, and the specific calculation method is as follows:

[0076]

[0077] In the formula, This represents the i-th data value in the time history data column read by the inertial sensor, where n represents the total number of data points (2250, sampling frequency 25 data points per second, for a total of 45 seconds). This represents the average value of the inertial sensor data. This represents the i-th data value in the time history data column read by optical motion capture, and the total number of corresponding data is also 2250. This represents the average value of the optical motion capture data. The closer the Spearman correlation coefficient is to 1, the more positively correlated the two sets of data are.

[0078] The reliability of inertial sensor data is evaluated by calculating the intraclass correlation coefficient ICC(1,1), as shown in the following formula.

[0079]

[0080] Similarly, in the formula and represents the i-th data value in the time history data column of the inertial sensor and the optical motion capture, respectively, and n also represents the total number of data in the data column, which is 2250. and Let represent the data column vectors of the inertial sensor and the optical motion capture, respectively. D() represents the variance of the calculated vectors. The closer the intraclass correlation coefficient (ICC(1,1)) is to 1, the higher the reliability of the inertial sensor data.

[0081] The accuracy of inertial sensor data is evaluated using the normalized root mean square error (NRMSE), calculated as follows:

[0082]

[0083] In the formula, and represents the i-th data value in the time history data column of the inertial sensor and the optical motion capture, respectively, and n also represents the total number of data in the data column, which is 2250. This represents the maximum value in the data column vector of optical motion capture. This represents the minimum value in the column vector of data from optical motion capture. The smaller the normalized root mean square error, the better the accuracy of the inertial sensor data.

[0084] For the left upper arm in freestyle strokes, the Spearman correlation coefficient is 0.995, the intraclass correlation coefficient (ICC(1,1)) is 0.985, and the normalized root mean square error (NRMSE) is 5.766%, demonstrating that the inertial sensor data possesses very good validity, reliability, and accuracy. Using the steps mentioned above, a comparison of data on the angle between the left thigh and the z-axis, and the angle between the left upper arm and the x-axis during freestyle strokes is also presented (e.g., Figure 7 As shown, the data from the inertial sensor and the optical motion capture data fit very well. For the angle between the left lower leg and the z-axis, the Spearman correlation coefficient is 0.965, the intra-class correlation coefficient ICC(1,1) is 0.913, and the normalized root mean square error (NRMSE) is 13.359%. For the angle between the left upper arm and the x-axis, the Spearman correlation coefficient is 0.927, the intra-class correlation coefficient ICC(1,1) is 0.933, and the normalized root mean square error (NRMSE) is 13.359%. This demonstrates that the inertial motion capture sensor data has high validity (Spearman correlation coefficient > 0.85), reliability (ICC > 0.85), and accuracy (NRMSE < 15%).

[0085] In this invention, the test subject wears 17 inertial measurement units (IMUs) to acquire Euler angle data for corresponding limbs. Simultaneously, the test subject attaches 25 optical target points to their body according to the Helen Hayes model, and an optical camera (optical motion capture system) records the motion data of each limb. Based on rotation matrices and coordinate transformation formulas, the measured segment Euler angle data can be converted into angle data between the segment and the global coordinate system, facilitating data evaluation. Finally, using the human data captured by optical motion capture as a standard, the accuracy, validity, and precision of the inertial sensor data are evaluated using the Spearman correlation coefficient, intraclass correlation coefficient ICC(1,1), and normalized root mean square error (NRMSE). Taking freestyle swimming as an example, the results show that this inertial sensor measurement scheme has high validity (Spearman correlation coefficient > 0.85), reliability (ICC > 0.85), and precision (NRMSE < 15%) for most body parts. Overall, this invention enables reliable and efficient measurement of swimming movements on land and can provide a basis for swimmer training.

[0086] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for evaluating onshore swimming motion based on inertial sensors and an optical motion capture system, characterized in that, Specifically, it includes two parts: acquiring onshore swimming motion data and evaluating onshore swimming motion. (a) Data acquisition of swimming motions on land: (1) Setting up the sports scene: The equipment used includes a land-based swimming trainer, several optical motion capture high-definition cameras, and a computer. The land-based swimming trainer is used to assist the test subject in performing various swimming strokes. The optical motion capture high-definition cameras are arranged around the land-based swimming trainer to capture the motion trajectory information of the target points on the test subject's body. The computer is used to acquire test data in real time. (2) Selection of the locations of optical target points and inertial sensors on the test subject: The test subject had optical targets attached to their body and was equipped with inertial sensors. The optical targets were arranged according to the Helen Hayes model, including the torso, upper limbs, and lower limbs. There were at least 25 targets, specifically located at: top of head, front of head, back of head, left arm, right shoulder, left elbow, right elbow, left wrist, right wrist, offset point, left lower back, right lower back, upper left thigh, upper right thigh, left lateral knee, right lateral knee, left lateral malleolus, right lateral malleolus, left toe, right toe, left heel, right heel, left calf, right calf, and sacrum. There were at least 17 inertial sensors, which were fixed to the limbs with straps, specifically located at: head, left scapula, right scapula, upper back, left upper arm, right upper arm, left forearm, right forearm, left hand, right hand, lower back, left thigh, right thigh, left calf, right calf, left foot, and right foot. The optical motion capture camera acquires the motion trajectory data of the corresponding limb through optical target points and transmits it to the computer via a data cable; the inertial sensor is used to acquire the time history data of the Euler angles of the corresponding limb and transmits it via Bluetooth. (II) Assessment of Swimming Movements on Land: The specific steps are as follows: Step 1: Before testing, calibrate the inertial motion capture sensor; Step 2: The test subject performs various swimming stroke tests on a land-based swimming training device; the stroke frequency and stroke duration are set for different swimming strokes, and several minutes of rest are required between each set; Step 3: The data obtained by optical motion capture is used to obtain the time history data of the angle between each limb and the global coordinates through the software system deployed in the computer, as well as the Euler angle data of each limb measured by the inertial sensor. Based on the rotation matrix formula and the coordinate transformation formula, the time history data of the angle between each limb and the global coordinates is calculated. Step 4: Align the optical motion capture data and inertial sensor data processed in Step 3. Specifically, for all swimming strokes, the left arm is used as the reference to determine the time point corresponding to the first peak in the left arm optical motion capture data, which is then used as the time reference point. Then, the inertial motion capture data is time-shifted to align the first peak in the inertial motion capture signal with the above reference point, thus achieving time synchronization. Step 5: Based on the data processed in Step 4, calculate the Spearman correlation coefficient, intraclass correlation coefficient ICC(1,1), and normalized root mean square error between the inertial sensor data and optical motion capture data of each limb under various swimming strokes, and evaluate the accuracy, validity, and precision of the data measured by the inertial sensor.

2. The method for evaluating onshore swimming movements according to claim 1, characterized in that, The test subjects performed swimming style tests on a land-based swimming training device. For breaststroke, the stroke rate was 40 strokes per minute for 45 seconds; for freestyle, the stroke rate was 80 strokes per minute for 45 seconds; and for butterfly, the stroke rate was 50 strokes per minute for 45 seconds. There was a 3-minute rest between each set.

3. The method for evaluating onshore swimming movements according to claim 1, characterized in that, Step 3 involves calculating the time history data of the angle between each body segment and the global coordinate system using the rotation matrix formula and coordinate transformation formula. Here, this corresponds to 17 inertial sensors and 17 body segments. Based on the human body node model, a table is first created showing the correspondence between the inertial sensor number and the corresponding end node numbers for each body segment (see Table 1). The length of each body segment of the test subject is then measured according to Table 1. Therefore, the angle between any body segment and the global coordinate system is calculated as follows: In the formula, θ k φ represents the rotation angle of the corresponding limb about the z-axis measured by the inertial sensor numbered k. k θ represents the rotation angle of the corresponding limb around the x-axis measured by the sensor numbered k. k Let represent the rotation angle along the y-axis measured by sensor k, where k is the inertial sensor number corresponding to the segment, with a value ranging from 126 to 142; Therefore, the formula for calculating the rotation matrix of any segment is: In the formula, q k Let be the node number vector corresponding to the segment of inertial sensor k; assuming the simplified model of the test subject's waist and abdomen is a rigid body during the test, nodes '0', '1', '5', and '9' are on the same rigid body, and the rotation matrices of the connecting line segments '0-9', '0-1', and '0-5' are all the same. Based on the homogeneous transformation matrix, the global coordinates of each node on the human body model are calculated, specifically using a North-Sky-East coordinate system, with the x-axis pointing north, the y-axis perpendicular to the ground and pointing east. For ease of description, an index vector f is introduced, which represents the node numbers along the motion chain from the target node to the root node. The index vector corresponding to each node is listed in Table 2. For any node num, where num takes values ​​from 1 to 19, its spatial position in the global coordinate system is calculated as follows: In the formula, f num The index vector corresponding to node num is shown in Table 2, where n is the length of the index vector; global P num This represents the position vector of the target node in global coordinates. global P0 represents the position vector of node '0' in global coordinates, i.e., [0,0,0]; Indicates that node f num (i) points to f num The segment vectors of (i+1) are shown in Table 1; since the node number vectors q and f are both vectors composed of node numbers, for and They are the same homogeneous transformation matrix; Then, for any body segment, the formulas for calculating the angles between any inertial sensor k and the x-axis, y-axis, and z-axis of the global coordinate system are as follows: In the formula, arccos is the inverse cosine function, · is the dot product operation between vectors, and || is the vector modulo operation; Table 1. Correspondence between test subject's body nodes and inertial sensors Table 2. Correspondence between test subject's body nodes and inertial sensors 4. The method for evaluating onshore swimming movements according to claim 1, characterized in that, Step 5 involves calculating the Spearman correlation coefficient, intra-class correlation coefficient (ICC(1,1)), and normalized root mean square error (RMSE) between the inertial sensor data and optical motion capture data for each limb under various swimming strokes; where: The Spearman correlation coefficient is calculated as follows: In the formula, This represents the i-th data value in the time history data column corresponding to the segment of the inertial sensor with number k (k ranges from 126 to 142) and the γ-axis (values ​​x, y, or z), where n represents the total number of data points. This represents the average value of the corresponding inertial sensor data; This represents the i-th data value in the optical motion capture time history data column corresponding to the segment. This represents the corresponding average value; the closer the Spearman correlation coefficient is to 1, the more positively correlated the two sets of data are. The reliability of inertial sensor data is evaluated by calculating the intraclass correlation coefficient ICC(1,1), as shown in the following formula: In the formula, and , respectively represent the i-th data value in the time history data column of the inertial sensor and optical motion capture of the segment corresponding to k and the γ axis, and n represents the total number of data in the data column; and These represent the data column vectors of the corresponding inertial sensor and optical motion capture, respectively; D(·) represents the variance of the calculated vector; the closer the intraclass correlation coefficient ICC(1,1) is to 1, the higher the reliability of the inertial sensor data. The accuracy of inertial sensor data is evaluated using the normalized root mean square error, calculated as follows: In the formula, and The definition is the same as in Formula 9; This represents the maximum value in the column vector of optical motion capture data for the corresponding segment. This represents the minimum value in the column vector of the corresponding segment optical motion capture data; the smaller the value of the normalized root mean square error, the better the accuracy of the inertial sensor data.

Images