Motion data correction method and device, storage medium and computer program product

Abstract

The embodiment of the invention discloses a motion data correction method and device, a storage medium and a computer program product, and the method comprises the steps: estimating the acceleration cumulative error of the two feet of a target object in a swing phase based on the original acceleration data and original angular velocity data of the two feet of the target object in a motion period, and obtaining a first cumulative error; utilizing an error estimation model to predict an error value between the first accumulative error and a real acceleration accumulative error of the two feet of the target object in the swing phase; correcting the first cumulative error based on the error value; and correcting the acceleration data of the two feet of the target object in the swing phase in the original acceleration data based on the corrected first accumulative error.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of data processing, and in particular, to a motion data correction method and device, a storage medium, and a computer program product. BACKGROUND

[0002] With the development of smart technology of wearable devices, an Interial Measurement Unit (IMU) is widely used as a hardware accessory of a smart watch and the like to provide original detection data for motion recognition such as steps and step speed. The IMU includes 3-axis (acceleration), 6-axis (acceleration + gyroscope), 9-axis (accelerometer + gyroscope + magnetometer), and 10-axis (accelerometer + gyroscope + magnetometer + barometer) according to different sensor combinations. Among them, the accelerometer, gyroscope, magnetometer, barometer, and pressure sensor respectively output original data: acceleration, angular velocity, azimuth, height value, and pressure value, and in combination with the phase of the gait cycle and the time-frequency domain motion characteristics, the kinetic and kinematic indexes can be calculated. In theory, the kinetic and kinematic indexes calculated by the last three can be obtained by the calculation of the first two. However, the accelerometer has a reaction lag, and the gyroscope is too sensitive, so the error correction of the last three is very important.

[0003] At present, a common method for calculating the stride of the IMU is the double integration method, which divides the gait cycle and uses the acceleration to integrate twice to obtain the distance between the toe-off and the next heel-strike. The existing zero velocity update (ZVU) method exists a period from the heel-strike to the heel-off for each step, and the average acceleration value measured during this period is used as an accelerometer error term to correct the cumulative error of the previous step, thereby greatly reducing the amplification of the stride with the increase in the number of steps. However, although the error is corrected once for each step, the error accumulation can be greatly reduced, but the zero velocity update is not fine enough, and the error will be amplified under the double integration, and the accuracy of the error correction of the acceleration data is insufficient. SUMMARY

[0004] Embodiments of the present application provide a motion data correction method, device, storage medium, and computer program product, which corrects the estimated acceleration cumulative error of the swing phase by using an error estimation model trained by deep learning, so that the acceleration cumulative error of the swing phase approximates the true value for error correction, thereby improving the accuracy of error correction.

[0005] The technical solution of the embodiments of the present application is as follows:

[0006] The embodiments of the present application provide a motion data correction method, which includes:

[0007] estimating, based on original acceleration data and original angular velocity data of a double foot of a target object in a movement cycle, an acceleration cumulative error of the double foot of the target object in a swing phase, to obtain a first cumulative error;

[0008] predicting, by using an error estimation model, an error value between the first cumulative error and a true acceleration cumulative error of the double foot of the target object in the swing phase;

[0009] correcting the first cumulative error based on the error value;

[0010] correcting, based on the corrected first cumulative error, acceleration data of the double foot of the target object in the swing phase in the original acceleration data.

[0011] In the above method, further comprising:

[0012] constructing and initializing a neural network;

[0013] estimating, based on sample acceleration data and sample angular velocity data of a double foot of a sample object in a movement cycle, an acceleration cumulative error of the double foot of the sample object in a swing phase, to obtain a second cumulative error;

[0014] determining, based on movement capture data of the sample object, the sample acceleration data and the sample angular velocity data, a true acceleration cumulative error of the double foot of the sample object in the swing phase;

[0015] training the neural network by taking a difference between the second cumulative error and the true acceleration cumulative error of the double foot of the sample object in the swing phase as a label and taking the second cumulative error as a feature;

[0016] determining the trained neural network as the error estimation model.

[0017] In the above method, the estimating, based on original acceleration data and original angular velocity data of a double foot of a target object in a movement cycle, an acceleration cumulative error of the double foot of the target object in a swing phase, to obtain a first cumulative error, comprises:

[0018] preprocessing original acceleration data and original angular velocity data of the double foot of the movement object in the movement cycle to obtain first acceleration data and first angular velocity data;

[0019] determining, based on the first acceleration data, a total acceleration cumulative error of the double foot of the target object in the movement cycle;

[0020] determining, based on the crowd statistical characteristics of the target object, the first angular velocity data and the first acceleration data, acceleration cumulative error of the target object's two feet in each of the two support phases through geometric modeling analysis;

[0021] determining, as the first cumulative error, a difference between the total acceleration cumulative error and the acceleration cumulative error of the target object's two feet in the two support phases.

[0022] In the above method, the preprocessing of the original acceleration data and the original angular velocity data of the target object's two feet in a movement cycle to obtain the first acceleration data and the first angular velocity data comprises:

[0023] complementary fusion first-level correction of the original acceleration data and the original angular velocity data;

[0024] second-level correction of the first-level corrected acceleration data and the first-level corrected angular velocity data in combination with time continuity and spatial correlation of the service;

[0025] determining, based on the second-level corrected acceleration data and the second-level corrected angular velocity data, the horizontal distance data of the target object's two feet in the two support phases;

[0026] correcting the second-level corrected acceleration data based on the horizontal distance data of the two feet to obtain the first acceleration data;

[0027] determining the second-level corrected angular velocity data as the first angular velocity data.

[0028] In the above method, the determination of the horizontal distance data of the target object's two feet in the two support phases based on the second-level corrected acceleration data and the second-level corrected angular velocity data comprises:

[0029] gait phase division based on the second-level corrected angular velocity data to determine the swing phase and the two support phases of the target object's two feet;

[0030] calculating global acceleration in a non-gravitational earth coordinate system based on the second-level corrected acceleration data and the second-level corrected angular velocity data;

[0031] integrating the global acceleration to obtain movement velocity of the target object's two feet in a movement cycle;

[0032] determining positions of the target object's two feet in the two support phases based on the movement velocity, and determining the horizontal distance data of the two feet based on the determined positions.

[0033] In the method, the correcting the second-level corrected acceleration data based on the horizontal distance data of the two feet to obtain the first acceleration data comprises:

[0034] The median difference method is used to detect each abnormal distance data from the horizontal distance data of the two feet;

[0035] Each acceleration data corresponding to each abnormal distance data in the second-level corrected acceleration data is determined as an abnormal acceleration data, and each remaining acceleration data is determined as a normal acceleration data;

[0036] Each abnormal acceleration data in the second-level corrected acceleration data is corrected to a previous normal acceleration data to obtain the first acceleration data.

[0037] In the method, the determining the acceleration accumulation error of the two support phases of the two feet of the target object based on the crowd statistical characteristics, the first angular velocity data and the first acceleration data through geometric modeling analysis comprises:

[0038] The first angular velocity data and the first acceleration data are fused into a standard quaternion, and the standard quaternion is converted into Euler angles;

[0039] For each support phase of the two feet of the target object, a corresponding foot change trajectory is constructed based on the corresponding angle information in the Euler angles and the crowd statistical characteristics;

[0040] For each support phase of the two feet of the target object, a corresponding real average acceleration value is determined based on the corresponding foot change trajectory and the distance motion equation;

[0041] For each support phase of the two feet of the target object, a corresponding measured average acceleration value is determined based on the corresponding acceleration data in the first acceleration data, and an acceleration accumulation error is determined based on the corresponding measured average acceleration value and the real average acceleration value.

[0042] Embodiments of the present application provide a motion data correction device, comprising:

[0043] An error estimation module is configured to estimate an acceleration accumulation error of a swing phase of the two feet of the target object based on original acceleration data and original angular velocity data of the two feet of the target object in a motion cycle to obtain a first accumulation error;

[0044] An error correction module is configured to predict an error value between the first accumulation error and a real acceleration accumulation error of the swing phase of the two feet of the target object by using an error estimation model, and correct the first accumulation error based on the error value.

[0045] a data correction module, configured to correct, based on the corrected first accumulated error, acceleration data of the target object in a swing phase in the original acceleration data.

[0046] An embodiment of the present application provides a motion data correction device, comprising a processor, a memory and a communication bus;

[0047] The communication bus is configured to realize communication connection between the processor and the memory.

[0048] The processor is configured to execute one or more computer programs stored in the memory, so as to realize the motion data correction method.

[0049] An embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the motion data correction method.

[0050] An embodiment of the present application provides a computer program product, comprising a computer program, and the computer program is executed by a processor to realize the motion data correction method.

[0051] An embodiment of the present application provides a motion data correction method, device, storage medium and computer program product, and the method comprises the following steps: based on original acceleration data and original angular velocity data of a target object in a motion cycle, estimating an acceleration accumulated error of the target object in a swing phase, to obtain a first accumulated error; using an error estimation model, predicting an error value between the first accumulated error and a real acceleration accumulated error of the target object in the swing phase; correcting the first accumulated error based on the error value; and correcting acceleration data of the target object in the swing phase in the original acceleration data based on the corrected first accumulated error. The technical scheme provided by the embodiment of the present application corrects the estimated swing phase acceleration accumulated error by using the error estimation model trained by deep learning, so that the swing phase acceleration accumulated error approximates to the real value for error correction, thereby improving the accuracy of error correction. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 An exemplary motion data correction hierarchical schematic diagram is provided for an embodiment of the present application.

[0053] Figure 2 An exemplary data processing process schematic diagram is provided for an embodiment of the present application.

[0054] Figure 3 A flowchart of a motion data correction method is provided for an embodiment of the present application.

[0055] Figure 4 An exemplary motion path diagram provided for an embodiment of the present application;

[0056] Figure 5 An exemplary gait phase division diagram provided for an embodiment of the present application;

[0057] Figure 6 An exemplary biped walking model diagram provided for an embodiment of the present application;

[0058] Figure 7 An exemplary geometric analysis of foot variation trajectory provided for an embodiment of the present application Figure 1 ;

[0059] Figure 8 An exemplary geometric analysis of foot variation trajectory provided for an embodiment of the present application Figure 2 ;

[0060] Figure 9 An exemplary model training process diagram provided for an embodiment of the present application Figure 1 ;

[0061] Figure 10 An exemplary position data diagram provided for an embodiment of the present application;

[0062] Figure 11 An exemplary error estimation model structure diagram provided for an embodiment of the present application;

[0063] Figure 12 An exemplary model training process diagram provided for an embodiment of the present application Figure 2 ;

[0064] Figure 13 An exemplary motion data correction device structure diagram provided for an embodiment of the present application Figure 1 ;

[0065] Figure 14 An exemplary motion data correction device structure diagram provided for an embodiment of the present application Figure 2 . DETAILED DESCRIPTION

[0066] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0067] The technical solutions of the present application and how the technical solutions solve the above technical problems will be specifically explained below through embodiments and in combination with the drawings. The following embodiments can be combined with each other, and the same or similar concepts or processes can not be described again in some embodiments.

[0068] In addition, the technical solutions described in the embodiments of the present application can be combined arbitrarily without conflict.

[0069] The embodiment of the present application provides a motion data correction method, which is realized by a motion data correction device. The specific type of the device is not limited in the embodiment of the present application, and it can be any electronic device.

[0070] The motion data correction method provided by the embodiment of the present application mines the characteristics of the two support phase double feet not leaving the ground in the double foot movement process, and proposes a zero position, zero speed, and deep learning layer-by-layer complementary filtering scheme. The contents involved include: first, on the basis of the time characteristics of the traditional zero speed (according to the zero position, the acceleration data abnormal value detection and correction are completed according to the double foot horizontal plane distance abnormal value detection method), combined with the spatial characteristics (mining the characteristics that the feet do not leave the ground, the position does not change, and the distance between the two feet does not change when the two support phases), then the geometric modeling analysis is performed on the two support phases before and after the flat foot period, the cumulative error of the swing phase and the two support phases in the movement cycle is solved respectively, the average error compensation strategy is adopted, the error of the most valuable swing phase is selected, and then the error of the swing phase is further corrected by the deep learning method, so that the error value is more close to the true value. The geometric method combines the adaptive calculation of the height of the crowd, the length of the foot and the position of the arch, and takes into account the portrait features of the business of the crowd.

[0071] Figure 1 An exemplary motion data correction level diagram is provided for the embodiment of the present application. As shown in Figure 1 , the correction process can involve multiple levels, and the motion data correction device for executing the motion data correction method can cover the terminal, mobile convergence center, server, etc. Figure 1 , as shown in Figure 2 An exemplary data processing process diagram is provided for the embodiment of the present application. The motion data correction method provided by the embodiment of the present application is described in detail below in combination with Figure 1 and Figure 2 .

[0072] Figure 3 A flowchart of a motion data correction method is provided for the embodiment of the present application. As shown in Figure 3 , in the embodiment of the present application, the motion data correction method mainly includes the following steps:

[0073] S101, estimate an acceleration cumulative error of the double feet of the target object in a swing phase based on original acceleration data and original angular velocity data of the double feet of the target object in a motion cycle, to obtain a first cumulative error;

[0074] S102, predict an error value between the first cumulative error and a true acceleration cumulative error of the double feet of the target object in the swing phase by using an error estimation model;

[0075] S103, correct the first cumulative error based on the error value;

[0076] S104, correct acceleration data of the double feet of the target object in the swing phase in the original acceleration data based on the corrected first cumulative error.

[0077] In the embodiment of the present application, the motion data correction device estimates an acceleration cumulative error of the double feet of the target object in a swing phase based on original acceleration data and original angular velocity data of the double feet of the target object in a motion cycle, to obtain a first cumulative error, which comprises: preprocessing the original acceleration data and the original angular velocity data of the double feet of the target object in the motion cycle to obtain first acceleration data and first angular velocity data; determining a total acceleration cumulative error of the double feet of the target object in the motion cycle based on the first acceleration data; determining acceleration cumulative errors of the double feet of the target object in two support phases respectively by geometric modeling analysis based on crowd statistical characteristics of the target object, the first angular velocity data and the first acceleration data; the two support phases are adjacent to the swing phase, one support phase is before the swing phase and the other is after the swing phase; determining the first cumulative error as a difference between the total acceleration cumulative error and the acceleration cumulative errors of the double feet of the target object in the two support phases.

[0078] In the embodiment of the present application, the target object can be a specific person, which is not limited in the embodiment of the present application. The target object can perform a set of straight walking and turning gait tasks to collect data, which includes original acceleration data and original angular velocity data of the double feet of the target object in a motion cycle.

[0079] For example, in the embodiment of the present application, the target object is required to first perform a static test, standing still with the double feet 15 cm apart. In the walking test, the target object performs a “straight walking-turning-straight walking-turning-straight walking” task as shown in Figure 4 and repeats 20 laps at a comfortable speed. The path length is 7 meters, and the subject must turn 180 degrees during walking. The original acceleration data, the original angular velocity data, the original attitude angle data and the like are collected by the IMU of the target object.

[0080] In the embodiment of the present application, the motion data correction device pre-processes the original acceleration data and the original angular velocity data of the target object's two feet in a motion cycle to obtain first acceleration data and first angular velocity data, including: performing first-level correction on the original acceleration data and the original angular velocity data based on a complementary fusion strategy; performing second-level correction on the first-level corrected acceleration data and the first-level corrected angular velocity data in combination with the time continuity and spatial correlation of the service; determining the horizontal distance data of the target object's two feet in two support phases based on the second-level corrected acceleration data and the second-level corrected angular velocity data; correcting the second-level corrected acceleration data based on the horizontal distance data of the two feet to obtain the first acceleration data; and determining the second-level corrected angular velocity data as the first angular velocity data.

[0081] In the embodiment of the present application, the motion data correction device can first perform first-level correction on the original acceleration data and the original angular velocity data through complementary fusion, which may involve multiple corrections, which will be described below.

[0082] In the embodiment of the present application, the motion data correction device can select the real-time change of speed and direction represented by the quaternion mode according to the fusion of the original acceleration data and the angular velocity data to the quaternion mode, which only occupies 4 bits, reduces the storage and calculation amount, and reduces the time overhead and conversion error of the multi-coordinate system conversion under the unified coordinate system to the earth coordinate system.

[0083] It should be noted that, in the embodiment of the present application, the quaternion The direction of the IMU coordinate system relative to the earth coordinate system is described, which can be obtained by integrating the quaternion change rate

[0084]

[0085] Wherein, the quaternion change rate The calculation is as follows:

[0086]

[0087] In the embodiment of the present application, the motion data correction device performs first error correction on acceleration and angular velocity, specifically, initializes the calibration key event point, and converges to determine the fusion gain coefficient K between the sensors included in the IMU, wherein the original angular velocity data ω collected by the gyroscope and the fusion gain coefficient K are related and calculated as follows:

[0088]

[0089] Wherein, the initial is a fixed value, which is inaccurate at the initial calculation. Through the initialization time t init ​The fusion gain coefficient K is dynamically converged from a larger initial value to a regular value, so that slowly converges to the accurate value.

[0090]

[0091] where t init is the initialization time in seconds, K init is the initial value of the fusion gain coefficient, K normal is the fusion gain coefficient when the normal quaternion is calculated, and K init >K normal , and t is the actual sampling time point.

[0092] In the embodiments of the present application, the motion data correction device performs second error correction of acceleration, corrects the determined acceleration error term using the fusion coordinate system, and determines the acceleration error component e from the original acceleration data a as follows:

[0093]

[0094] where q x , q y , q z , q ω are elements of the normalized quaternion .

[0095] In the embodiments of the present application, the motion data correction device performs second error correction of angular velocity, corrects the determined angular velocity error term using the fusion coordinate system. Linear error is generated in the measurement process of the gyroscope, and the estimated error value ω bias can be subtracted before the complementary filter estimation calculation, and the ω in the quaternion rate of change calculation uses the corrected ω':

[0096] ω' = ω - ω bias (6)

[0097] where the estimated error value ω bias is calculated using the low-pass filtering result:

[0098] ω bias = 2πf c ∫pω·dt (7)

[0099]

[0100] where f c represents the cutoff frequency, and f b represents the bandwidth of the low-pass filtering.

[0101] In the embodiment of the present application, the motion data correction device performs third error correction of acceleration, removes the influence of gravity acceleration, and calculates the acceleration in the earth coordinate system according to the current attitude quaternion. The zero gravity acceleration a zero is the accelerometer measurement value after removing gravity. The gravity component in the IMU coordinate system is removed from the original acceleration data a:

[0102]

[0103] By normalizing the change of the quaternion, the global acceleration a global in the earth coordinate system can be calculated:

[0104]

[0105] In the embodiment of the present application, the motion data correction device performs fourth error correction of acceleration, linear acceleration suppression of inclination, and linear acceleration suppression to reduce errors caused by linear and rotational motion. The working principle of acceleration suppression is to compare the instantaneous inclination measurement value provided by the accelerometer with the current inclination measurement value g d output by the algorithm. If the angle difference between the two inclinations is greater than a threshold t a , the accelerometer will be ignored in this algorithm update, as shown in the following formula:

[0106]

[0107] -g d <(‖a‖-1)<g d (12)

[0108] In the embodiment of the present application, the above series of processes can be used to realize the first-level correction of complementary fusion of the original acceleration data and the original angular velocity data. Then, the first-level corrected acceleration data and angular velocity data are combined with the time continuity and spatial correlation of the service to perform second-level correction, which will be described in detail below.

[0109] In the embodiment of the present application, the motion data correction device can determine the gait phase of the key gait event to continue correcting the first-level corrected acceleration data and angular velocity data in combination with the time continuity and spatial correlation of the service to obtain the second-level corrected acceleration data and angular velocity data. The gait phase of the key gait event is determined by the following method:

[0110] 1. According to the attitude angle signal, use peak detection to identify the toe-off (TO) and heel strike (HS) gait events;

[0111] 2. According to the attitude angle signal, perform first-order differentiation to obtain the pitch angle velocity;

[0112] 3. Filtering the pitch angle velocity through a third-order zero-lag high-pass filter to obtain the filtered pitch angle velocity;

[0113] 4. Using threshold detection and peak detection methods to identify the heel-off (HO) and foot-flat (FF) gait events from the angular velocity data;

[0114] 5. Calculating the gait time parameters according to the HO, TO, HS and FF gait events, and then dividing the gait phase, calculating the stance phase time and swing phase time.

[0115] In the embodiments of the present application, after obtaining the second-level corrected acceleration data and angular velocity data, the motion data correction device determines the horizontal distance data of the target object's two feet in two support phases based on the second-level corrected acceleration data and angular velocity data, including: dividing the gait phase based on the second-level corrected angular velocity data to determine the swing phase and the two support phases of the target object's two feet; calculating the global acceleration in the non-gravitational earth coordinate system based on the second-level corrected acceleration data and the second-level corrected angular velocity data; integrating the global acceleration to obtain the motion velocity of the target object's two feet in the motion cycle; determining the positions of the target object's two feet in the two support phases based on the motion velocity, and determining the horizontal distance data of the two feet based on the determined positions.

[0116] In the embodiments of the present application, the gait phase can be divided according to the above-mentioned gait phase division method according to the second-level corrected angular velocity data, the swing phase and the two support phases before and after it are determined, and the positions of the two feet in the two support phases are calculated to determine the horizontal distance data of the two feet. According to the gait events, the stationary period is defined as: Terminal Swing (TS) → Heel Off (HO), and the motion cycle is defined as: HO → TS, wherein TS corresponds to the above-mentioned FF gait event. The motion cycle can be refined into three stages: support phase HO → TO, swing phase TO → HS and support phase HS → TS, as shown in Figure 5 , that is, the swing phase and the two support phases before and after it in the present application.

[0117] Referring to Figure 6 , according to the analysis of the human walking rule, the horizontal distance amplitude of the two feet in the two-foot support phase (one foot is in the support phase in the motion cycle: HO → TO, and the other foot is in the support phase in the motion cycle: HS → TS) changes slightly, based on which, the positions of the two feet in the earth coordinate system are preliminarily calculated as follows:

[0118]

[0119] wherein, tstart is the start time of walking, t end is the end time of walking, p0 represents the initial position (0, 0, 0) when walking, and v is the movement speed.

[0120] The formula for calculating the horizontal distance of the biped within the support phase is as follows:

[0121]

[0122] where T DLS is the duration of the biped support phase, P represents the position data, R / L respectively represent the right foot and the left foot, and D t represents the distance data of the biped on the horizontal plane at time t. The biped horizontal distance data can include the distance data of the biped on the horizontal plane at different times in two support phases, i.e., multiple distance data.

[0123] In an embodiment of the present application, the motion data correction device corrects the acceleration data of the second level correction based on the biped horizontal distance data to obtain first acceleration data, including: using the median deviation method to detect each abnormal distance data from the biped horizontal distance data; determining the acceleration data corresponding to each abnormal distance data in the second level corrected acceleration data as an abnormal acceleration data, and determining the remaining each acceleration data as a normal acceleration data; correcting each abnormal acceleration data in the second level corrected acceleration data to the previous normal acceleration data to obtain the first acceleration data.

[0124] In an embodiment of the present application, the biped horizontal distance data is used as the basis for detecting abnormal acceleration data in the second level corrected acceleration data according to the small change of the biped horizontal distance data. When a person is in a walking state, the abnormal acceleration data is detected in combination with the abnormal distance data in the biped horizontal distance data. For the calculated biped horizontal distance data, the absolute median deviation method (Mad-Median Rule) is used to detect the abnormal distance data therein.

[0125] First, the median of the biped horizontal distance data is calculated:

[0126]

[0127] The biped horizontal distance data can include the distance data of the biped on the horizontal plane at different times, i.e., multiple distance data. If the number n of distance data is odd, the median is the middle number; if n is even, the median is the average of the middle two numbers, where X represents all sorted distance data, and median represents the method for solving the median.

[0128] The absolute deviation Deviation of each distance data from the median is calculatedi :

[0129] Deviation i = |X i -median(X) | (16)

[0130] where X i represents the ith distance data in the horizontal distance data of the biped.

[0131] Calculate the absolute median deviation MAD of all the absolute deviations obtained above:

[0132] MAD = median(Deviation1, Deviation2, …, Deviation n ) (17)

[0133] Convert the MAD to a scale similar to the standard deviation, using a constant factor 0.6745 (the inverse of the 25th percentile of the standard normal distribution), to calculate the modified score Z-Score corresponding to each distance data:

[0134]

[0135] where, if the modified score Z-Score i of the ith distance data exceeds the threshold value 2.5, it can be considered that the ith distance data is an abnormal distance data, and the acceleration data corresponding to it in the second-level modified acceleration data is determined as an abnormal acceleration data, and is modified to the previous normal acceleration data, so as to realize the modification of the second-level modified acceleration data, and obtain the first acceleration data.

[0136] In the embodiments of the present application, after obtaining the first acceleration data, the motion data correction device can use the zero velocity update method to determine the total acceleration cumulative error of the biped of the target object in the motion cycle based on the first acceleration data.

[0137] In the embodiments of the present application, the linear velocity of the foot is zero when the foot is stationary, and the velocity at rest is set to zero, and the acceleration data of the biped of the target object in the motion cycle is integrated to calculate the velocity.

[0138] Thus, the global acceleration under the gravity-free earth coordinate system is integrated to obtain the motion velocity of the biped of the target object in the motion cycle:

[0139]

[0140] where aglobal(t) is the global acceleration under the gravity-free earth coordinate system, t HO is the time point of the HO gait event, t TSis the time point of the TS gait event.

[0141] Due to bias error in acceleration data measurement, the estimated angular velocity can not be zero when the foot is in the stance phase. The difference between the actual velocity (known to be zero) and the integrated velocity is used to correct the bias error in acceleration. The drift error resulting from the integration of acceleration is further eliminated by calculating the accumulated error of the motion period.

[0142] For the first acceleration data of the motion period:

[0143]

[0144] T = t TS -t HO (21)

[0145] wherein, is the first acceleration data, e represents the bias error in the earth coordinate system, is the true acceleration value resulting from the motion, e is the bias error, and T is the duration of the motion period, t HO is the time point of the HO gait event, t TS is the time point of the HS gait event.

[0146] The bias error is a variable e(t) that fluctuates slightly over time within a short time of the motion period. And at the beginning of the motion period, the velocity is zero. Then the solution for the motion velocity is rewritten as follows:

[0147]

[0148] wherein, represents the integrated velocity, which is composed of the true velocity and the accumulated error function e(t), t ∈ [0, T].

[0149] At the end of the motion period, i.e. when t = T, the actual velocity is zero when the sole is fully in contact with the ground. Therefore, the bias error in the acceleration measurement can be calculated as follows:

[0150]

[0151] Based on this, the value of e(T) is an accumulated error value of the integration of acceleration in the motion period. Then the accumulated error of the accelerometer of the IMU between two double-foot stance phases, i.e. the total acceleration accumulated error of the target object in the motion period, can be obtained as follows:

[0152]

[0153] It should be noted that the total acceleration cumulative error of the movement cycle includes the acceleration cumulative errors of the swing phase and the two support phases before and after the swing phase, i.e., the acceleration cumulative errors of the three stages of the support phase HO→TO, the swing phase TO→HS, and the support phase HS→TS. The determination method of the acceleration cumulative error of each support phase of the target object's two feet is described in detail below.

[0154] In the embodiments of the present application, the motion data correction device determines the acceleration cumulative error of each support phase of the target object's two feet based on the population statistical characteristics of the target object, the first angular velocity data and the first acceleration data through geometric modeling analysis, including: fusing the first angular velocity data and the first acceleration data into a standard quaternion, and converting the standard quaternion into Euler angles; for each support phase of the target object's two feet, constructing a corresponding foot change trajectory based on the corresponding angle information in the Euler angles and the population statistical characteristics; for each support phase of the target object's two feet, determining a corresponding true average acceleration value based on the corresponding foot change trajectory and the distance motion equation; for each support phase of the target object's two feet, determining a corresponding measured average acceleration value based on the corresponding acceleration data in the first acceleration data, and determining the acceleration cumulative error based on the corresponding measured average acceleration value and the true average acceleration value.

[0155] In the embodiments of the present application, the population statistical characteristics can be obtained, including obtaining gender and height data, calculating foot length in proportion according to the following table, and using average value calculation in the case where it cannot be obtained, as involved in the following formulas (26) to (28): tiptoe , as involved in the following formulas (32) to (34): heel and the two parameters have a linear relationship with the length of the worn insole. The length of the foot has a subtle relationship with the height, so an implicit mapping relationship between the relevant attributes and the shoe length can be constructed based on large-scale data, for example:

[0156] Table 1

[0157] Gender Height Foot length Male 170 24.5 Female 158 22.5

[0158] In the embodiments of the present application, the motion data correction device can fuse the first angular velocity data and the first acceleration data into a standard quaternion, and convert the standard quaternion into Euler angles. The conversion of the standard quaternion into Euler angles is a common attitude representation conversion, which is used to convert the attitude information of the quaternion into a more understandable Euler angle representation. The standard quaternion is converted into Euler angles, where q w , q x , q y and q z are the elements of the standard quaternion:

[0159]

[0160] where the results of arctan and arcsin are This does not cover all orientations (for q pitch angle The range of values has been satisfied, so atan2 can be used instead of arctan.

[0161] In the embodiments of the present application, for one of the two support phases HO→TO, i.e. Figure 5 Phase 1, the changing trajectory of the foot is similar to a circle, through geometric analysis, see Figure 7 The distance changes of the x, y, and z axes in this support phase can be calculated as follows:

[0162]

[0163]

[0164]

[0165] where △z1, △y1, and △x1 represent the distances of the z, y, and x axes in the support phase HO→TO, respectively, L tiptoe represents the distance from the IMU placement position to the insole tip, represents the pitch angle corresponding to the TO gait event, represents the pitch angle corresponding to the TO gait event.

[0166] According to the distance motion equation:

[0167]

[0168] where △z1, △y1, and △x1 are known, the real average acceleration value in the support phase HO→TO can be solved:

[0169] The average value of the acceleration data corresponding to the support phase HO→TO in the second acceleration data is obtained as Since can be represented as:

[0170]

[0171] where is the measured average acceleration value corresponding to the support phase HO→TO, is the real average acceleration value corresponding to the support phase HO→TO, is the average error value in the support phase HO→TO.

[0172] Based on this, the acceleration cumulative error corresponding to the support phase HO→TO can be calculated:

[0173]

[0174] wherein e(T1) represents the acceleration cumulative error corresponding to the support phase HO→TO, and T1 is the duration of the support phase HO→TO.

[0175] In the embodiments of the present application, for the other support phase HS→TS in the two support phases, i.e. Figure 5 As shown in stage three, the changing trajectory of the foot is similar to a circle, through geometric analysis, see Figure 8 The distance changes of the x, y, and z axes in this support phase can be calculated as follows:

[0176]

[0177]

[0178]

[0179] wherein △z3, △y3, and △x3 represent the distances of the z, y, and x axes in the support phase HS→TS, respectively, and L heel represents the distance between the IMU placement position and the heel of the insole, represents the pitch angle corresponding to the HS gait event, represents the pitch angle corresponding to the HS gait event.

[0180] The support phase HS→TS is the reverse process of the support phase HO→TO, and the speed gradually increases from zero in the support phase HO→TO, while the speed gradually decreases to zero in the support phase HS→TS. The motion equation of the distance in the support phase HS→TS is also as follows:

[0181]

[0182] Given △z3, △y3, and △x3, the real average acceleration value in the support phase HS→TS can be solved:

[0183] The average value of the acceleration data corresponding to the support phase HS→TS in the second acceleration data is obtained Since can be represented as:

[0184]

[0185] wherein, is the measured average acceleration value corresponding to the support phase HS→TS, is a real average acceleration value corresponding to the supporting phase HS TS, is an average error value in the supporting phase HS TS.

[0186] Based on this, the acceleration cumulative error corresponding to the supporting phase HS TS can be calculated:

[0187]

[0188] wherein e(T3) represents the acceleration cumulative error corresponding to the supporting phase HS TS, and T3 is the duration of the supporting phase HS TS.

[0189] It can be understood that in the embodiments of the present application, the total acceleration cumulative error is composed of the acceleration cumulative errors of the supporting phase HO TO, the swing phase TO HS, and the supporting phase HS TS, based on which, the difference between the total acceleration cumulative error and the acceleration cumulative errors of the two supporting phases is the acceleration cumulative error of the swing phase:

[0190] e(T2) = e(T) - e(T1) - e(T3) (38)

[0191] wherein e(T2) represents the acceleration cumulative error of the swing phase TO HS, that is, the first cumulative error in the present application, and T2 is the duration of the swing phase TO HS.

[0192] In the embodiments of the present application, after estimating the acceleration cumulative error of the double feet of the target object in the swing phase, that is, the first cumulative error, the estimated acceleration cumulative error of the swing phase can be further corrected by using the error estimation model, so that the acceleration cumulative error of the swing phase approximates the real value.

[0193] It should be noted that in the embodiments of the present application, considering that the neural network has excellent generalization performance, its special learning ability can effectively extract data features, and the approximation of nonlinear functions is the advantage of neural networks, therefore, the neural network can be used to train the error estimation model, so as to predict the error value between the first cumulative error and the real acceleration cumulative error of the double feet of the target object in the swing phase.

[0194] In the embodiments of the present application, the motion data correction device can further perform the following steps: constructing and initializing a neural network; estimating the acceleration cumulative error of the sample object's double feet in the swing phase based on the sample acceleration data and the sample angular velocity data of the sample object's double feet in the sample motion cycle, to obtain a second cumulative error; determining the real acceleration cumulative error of the sample object's double feet in the swing phase based on the motion capture data, the sample acceleration data and the sample angular velocity data of the sample object; training the neural network by taking the difference between the second cumulative error and the real acceleration cumulative error of the sample object's double feet in the swing phase as a label and taking the second cumulative error as a feature; and determining the trained neural network as an error estimation model.

[0195] In the embodiments of the present application, the error estimation model training data set can be made by calculating the real acceleration cumulative error of the sample object's double feet in the swing phase, and taking the difference between the real acceleration cumulative error and the second cumulative error as a label, and taking the second cumulative error as a feature, wherein the determination method of the second cumulative error, i.e., the acceleration cumulative error of the sample object's double feet in the swing phase, is similar to the determination method of the first cumulative error, which will not be described here again, and the determination method of the real acceleration cumulative error of the sample object's double feet in the swing phase is shown in the model training process of Figure 9 In addition, the sample object can be a certain user, and the specific sample object can be one or more, which is not limited in the embodiments of the present application.

[0196] According to the change trend of the position data in the visualization walking, the characteristics corresponding to the gait events are mined, as shown in Figure 10 , wherein the HS gait event (RTOE_Z axis data is a peak value) and the TO gait event (RTOE_Z axis data peak valley) are determined based on rules, and RTOE_z represents the Z axis data of the right toe-off.

[0197] According to the HS and TO gait events, the gait cycle can be divided into a support phase and a swing phase. Thus, the average acceleration value of the swing phase can be calculated according to the motion capture data (position data) collected by the motion capture system (vicon) in the swing phase and the motion characteristics (the initial speed of the TO and HS gait events is approximately equal to 0) of each step of a person walking.

[0198] In combination with the steps shown in Figure 9 , the distance D of the foot movement of the sample object's double feet in the swing phase is calculated swing :

[0199] D swing =p HS -p TO (39)

[0200] wherein p HS and p TOThese represent the foot positions during HS gait events and TO gait events, respectively.

[0201] Calculate the oscillation phase time interval t swing :

[0202] t swing =t HS -t TO (40)

[0203] Among them, t HS and t TO These represent the time points of the HS gait event and the TO gait event, respectively.

[0204] To calculate the distance traveled based on acceleration, using basic kinematic formulas, given that an object accelerates from rest, the distance formula is:

[0205]

[0206] Based on this, and combining the above formulas with the motion capture data (position data) collected by the motion capture system within the swing phase, the average acceleration of the swing phase is calculated. for:

[0207]

[0208] The average value of the raw acceleration data within the swing phase is calculated based on the raw acceleration data collected by the accelerometers included in the IMU.

[0209]

[0210] according to and The cumulative error Δe of the true acceleration of the sample object's two feet during the swing phase can be calculated. true :

[0211]

[0212] △e zvu Based on the cumulative acceleration error of the two feet during the swing phase estimated for the target object according to step S101 above, Δe can then be calculated. zvu and △e true The difference between the two values ​​is Δe = Δe true -△e zvu Thus, they can be used as labels for model training, such as Figure 9 As shown.

[0213] It can be understood that the cumulative error compensation of swing phase in the dynamic process of human walking can be caused by the dynamic changes of the foot and external disturbances. In order to correct this error, the neural network needs to be able to identify and predict the pattern of these errors and generate the corresponding compensation difference.

[0214] In the embodiments of the present application, the error estimation model is a multi-layer feedforward neural network, that is, a backpropagation (BP) neural network. From Figure 11 It can be seen from the above that the network comprises an input layer, two hidden layers and an output layer. The input layer inputs the data of the cumulative error of the swing phase calculated by the optimization zero velocity update scheme. The hidden layer converts these inputs and passes through the weights and biases in the network, which are updated during the training process to minimize the difference between the output and the target value. The output layer generates a cumulative error difference compensation signal to adjust the cumulative error of the swing phase in the optimization zero velocity scheme, so that the final swing phase error value is closer to the true value.

[0215] In the embodiments of the present application, during the model training process, the input signal is first forward propagated through the weight connections between layers until the output signal is generated. The difference between the predicted result of the output layer and the actual result forms the error signal. This error is then used in the backpropagation process, starting from the output layer and passing through each layer back to the input layer. During the backpropagation process, the weights and bias parameters of the network are adjusted according to the error gradient, which is usually implemented using the gradient descent method or its variants. The goal of adjustment is to minimize the loss function of the network, which is a quantitative indicator of the prediction error. Through multiple iterations of this process, the neural network can learn the characteristics of the data and gradually reduce the error between the prediction result and the target value. The backpropagation of error and the update of weights rely on the chain rule to calculate the gradient, which is usually combined with the derivative of the loss function and the derivative of the activation function to execute. This method gradually adjusts the network weights to minimize the loss function, allowing the network to learn complex function mappings. The related steps of the above content are shown in Figure 12 .

[0216] In the embodiments of the present application, after the motion data device implements the estimated acceleration accumulation error of the double feet of the target object in the swing phase, i.e., the first cumulative error correction, the corresponding acceleration data in the original acceleration data can be corrected, and of course, the determination of the acceleration accumulation error of the double feet of the target object in the two support phases is also involved in the implementation process of the above step S101, so the corresponding acceleration data in the original acceleration data can also be corrected according to the acceleration accumulation error of the support phases HO→TO and HS→TS, and then the corrected acceleration data is used to calculate the stride and position information. Based on this, the technical scheme provided by the embodiments of the present application can use the acceleration accumulation error values of the swing phase and the two support phases before and after the motion cycle to correct the acceleration data of the three phases, more accurately correct the bias error of the original acceleration data and the original angular velocity data measured by the double foot IMU, and through the ordered fusion of the dimensional feature information, the data analysis result accuracy is considered on the basis of simplifying the hardware and reducing the cost to meet the business needs.

[0217] It should be noted that in the embodiments of the present application, the average error value can be obtained according to the calculation of the acceleration accumulation error And So the cumulative error caused by the acceleration integration when solving the speed corresponding to the swing phase and the two support phases of the motion cycle can be compensated respectively.

[0218] It can be understood that in the embodiments of the present application, the compensated speed according to the motion cycle geometric modeling analysis is more close to the linear speed generated by the foot when walking than the traditional zero speed updated compensated speed. Therefore, according to the motion cycle geometric modeling analysis compensated speed, the stride and position information data calculated by the integral method are more accurate.

[0219] It should be noted that in the embodiments of the present application, the stride is calculated by integrating the speed in the swing phase of the motion cycle, as follows:

[0220]

[0221] Wherein represents the speed after correction of error, t HS is the time point of the HS gait event, t TO is the time point of the TO gait event.

[0222] It can be understood that the traditional zero velocity update adopts the average error of the entire motion cycle to compensate for the speed cumulative error in the swing phase, and the compensation is not accurate. In the present application, according to the motion cycle geometric modeling analysis, the average error of the swing phase is used to compensate for the speed cumulative error in the swing phase, so that the speed after compensation is more consistent with the true value, and the calculated step is more accurate.

[0223] In the embodiment of the present application, the position at each time is calculated by integrating the speed in the walking time as follows:

[0224]

[0225] Where t start is the start time of walking, t end is the end time of walking, and p0 represents the initial position during walking.

[0226] It can be understood that according to the motion cycle geometric modeling analysis, the average error of each phase in the motion cycle is used to compensate for the speed cumulative error in each phase, and the compensated speed is closer to the true speed value of each phase in the motion cycle, so that the accuracy of the position data at each time is higher. The error compensation is performed for each step, and the speed compensation accuracy of each step is improved compared with the traditional zero velocity update. The final position data is the integral of the speed during the entire walking process, so the accuracy of the final calculated position is higher than that of the traditional zero velocity update.

[0227] The embodiment of the present application provides a motion data correction device, Figure 13 The structure of the motion data correction device provided by the embodiment of the present application is shown in Figure 1 . As shown in Figure 13 , the motion data correction device comprises:

[0228] The error estimation module 201 is configured to estimate the acceleration cumulative error of the double legs of the target object in the swing phase based on the original acceleration data and the original angular velocity data of the double legs of the target object in the motion cycle, and obtain a first cumulative error.

[0229] The error correction module 202 is configured to predict the error value between the first cumulative error and the true acceleration cumulative error of the double legs of the target object in the swing phase by using the error estimation model, and correct the first cumulative error based on the error value.

[0230] The data correction module 203 is configured to correct the acceleration data of the double legs of the target object in the swing phase in the original acceleration data based on the corrected first cumulative error.

[0231] In an embodiment of the present application, the error correction module 202 is further configured to construct and initialize a neural network; estimate an acceleration cumulative error of a double foot of a sample object in a swing phase based on sample acceleration data and sample angular velocity data of the double foot of the sample object in a motion cycle, to obtain a second cumulative error; determine a real acceleration cumulative error of the double foot of the sample object in the swing phase based on motion capture data of the sample object, the sample acceleration data and the sample angular velocity data; train the neural network by taking a difference between the second cumulative error and the real acceleration cumulative error of the double foot of the sample object in the swing phase as a label and taking the second cumulative error as a feature; and determine the trained neural network as the error estimation model.

[0232] In an embodiment of the present application, the error estimation module 201 is configured to preprocess original acceleration data and original angular velocity data of a double foot of a motion object in a motion cycle to obtain first acceleration data and first angular velocity data; determine a total acceleration cumulative error of a double foot of a target object in a motion cycle based on the first acceleration data; determine acceleration cumulative errors of the double foot of the target object in two support phases respectively through geometric modeling analysis based on crowd statistical characteristics of the target object, the first angular velocity data and the first acceleration data; the two support phases are adjacent to a swing phase, one support phase is located before the swing phase and the other is located after the swing phase; and determine the first cumulative error by taking a difference between the total acceleration cumulative error and the acceleration cumulative errors of the double foot of the target object in the two support phases.

[0233] In an embodiment of the present application, the error estimation module 201 is configured to perform complementary fusion first-level correction on the original acceleration data and the original angular velocity data; perform second-level correction on the first-level corrected acceleration data and the first-level corrected angular velocity data in combination with time continuity and spatial correlation of a service; determine double foot horizontal distance data of the double foot of the target object in the two support phases based on the second-level corrected acceleration data and the second-level corrected angular velocity data; correct the second-level corrected acceleration data based on the double foot horizontal distance data to obtain the first acceleration data; and determine the second-level corrected angular velocity data as the first angular velocity data.

[0234] In an embodiment of the present application, the error estimation module 201 is configured to divide the gait phase based on the second corrected angular velocity data, determine the swing phase and two support phases of the target object's biped, calculate the global acceleration in the non-gravitational earth coordinate system based on the second corrected acceleration data and the second corrected angular velocity data, and integrate the global acceleration to obtain the movement speed of the target object's biped in a movement cycle.

[0235] In an embodiment of the present application, the error estimation module 201 is configured to detect each abnormal distance data from the biped horizontal distance data by using the median difference method, determine each acceleration data corresponding to each abnormal distance data in the second corrected acceleration data as an abnormal acceleration data, and determine each acceleration data other than the abnormal acceleration data as a normal acceleration data, and correct each abnormal acceleration data in the second corrected acceleration data to a previous normal acceleration data to obtain the first acceleration data.

[0236] In an embodiment of the present application, the error estimation module 201 is configured to fuse the first angular velocity data and the first acceleration data into a standard quaternion, and convert the standard quaternion into Euler angles, construct a corresponding foot change trajectory based on the corresponding angle information in the Euler angles and the crowd statistical characteristics for each support phase of the target object's biped, determine a corresponding real average acceleration value based on the corresponding foot change trajectory and the distance movement equation for each support phase of the target object's biped, determine a corresponding measured average acceleration value based on the corresponding acceleration data in the first acceleration data for each support phase of the target object's biped, and determine an acceleration cumulative error based on the corresponding measured average acceleration value and the real average acceleration value.

[0237] Figure 14 A structure of a motion data correction device provided in an embodiment of the present application Figure 2 As shown in Figure 14 , the motion data correction device includes a processor 301, a memory 302, and a communication bus 303.

[0238] The communication bus 303 is configured to realize the communication connection between the processor 301 and the memory 302.

[0239] The processor 301 is configured to execute one or more computer programs stored in the memory 302 to realize the motion data correction method described above.

[0240] The embodiment of the present application provides a computer program product, comprising a computer program, which realizes the motion data correction method when executed by a processor.

[0241] The embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program realizes the motion data correction method when executed by a processor. The computer readable storage medium can be a volatile memory (volatile memory), such as a random access memory (Random-Access Memory, RAM); or a non-volatile memory (non-volatile memory), such as a read-only memory (Read-Only Memory, ROM), a flash memory (flash memory), a hard disk (Hard Disk Drive, HDD) or a solid state disk (Solid-State Drive, SSD); or a device including one or any combination of the above memories, such as a mobile phone, a computer, a tablet device, a personal digital assistant, etc.

[0242] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can be in the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can be in the form of a computer program product implemented on one or more computer usable storage media (including but not limited to magnetic disk storage and optical storage, etc.) containing computer usable program code.

[0243] The present application is described with reference to the implementation flowchart and / or block diagram of the method, device (system), and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram and the combination of the flows and / or blocks in the flowchart and / or block diagram can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device that implements the functions specified in the flowchart and / or block diagram. Figure 1 The device for implementing the functions specified in one or more flows and / or blocks Figure 1 The device for implementing the functions specified in one or more flows and / or blocks

[0244] These computer program instructions can also be stored in a computer readable storage medium that can guide the computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer readable storage medium produce a product including instruction devices, which implement the functions specified in the flowchart and / or block diagram. Figure 1one or more processes and / or blocks Figure 1 the function specified in the one or more blocks.

[0245] These computer program instructions can also be loaded into computer or other programmable data processing devices, so that a series of operation steps are performed on the computer or other programmable data processing devices to generate computer-implemented processes, so that the instructions executed on the computer or other programmable devices provide a process for implementing the flowchart Figure 1 one or more processes and / or blocks Figure 1 the function specified in the one or more blocks.

[0246] The above description is merely one specific implementation of the application. However, one of ordinary skill in the art should, in light of the above description, appreciate that many changes and substitutions can be made to the present application. Therefore, the present application should not be limited to the above description but should be accorded a full scope of the claims.

Claims

1. A method of motion data correction, characterized by, The method comprises: estimating, based on original acceleration data and original angular velocity data of a target object's biped during a motion cycle, an acceleration cumulative error of the target object's biped during a swing phase, to obtain a first cumulative error; predicting, using an error estimation model, an error value between the first cumulative error and a true acceleration cumulative error of the target object's biped during the swing phase; correcting the first cumulative error based on the error value; correcting, based on the corrected first cumulative error, acceleration data of the target object's biped during the swing phase in the original acceleration data.

2. The method of claim 1, wherein, The method further comprises: constructing and initializing a neural network; estimating, based on sample acceleration data and sample angular velocity data of a sample object's biped during a motion cycle, an acceleration cumulative error of the sample object's biped during a swing phase, to obtain a second cumulative error; determining, based on motion capture data of the sample object, the sample acceleration data and the sample angular velocity data, a true acceleration cumulative error of the sample object's biped during the swing phase; training the neural network by taking the difference between the second cumulative error and the true acceleration cumulative error of the sample object's biped during the swing phase as a label and taking the second cumulative error as a feature; determining the trained neural network as the error estimation model.

3. The method of claim 1, wherein, The method of estimating, based on original acceleration data and original angular velocity data of a target object's biped during a motion cycle, an acceleration cumulative error of the target object's biped during a swing phase, to obtain a first cumulative error, comprises: preprocessing the original acceleration data and the original angular velocity data of the target object's biped during a motion cycle to obtain first acceleration data and first angular velocity data; determining, based on the first acceleration data, a total acceleration cumulative error of the target object's biped during a motion cycle; determining, based on population statistical characteristics of the target object, the first angular velocity data and the first acceleration data, acceleration cumulative errors of the target object's biped during two support phases respectively through geometric modeling analysis; the two support phases are adjacent to the swing phase, one support phase is before the swing phase and the other is after the swing phase; determining, as the first cumulative error, the difference between the total acceleration cumulative error and the acceleration cumulative errors of the target object's biped during the two support phases.

4. The method of claim 3, wherein, The method of preprocessing the original acceleration data and the original angular velocity data of the target object's biped during a motion cycle to obtain first acceleration data and first angular velocity data, comprises: complementarily fusing the original acceleration data and the original angular velocity data for first-level correction; performing second-level correction on the first-level corrected acceleration data and the first-level corrected angular velocity data in combination with time continuity and spatial correlation of a service; determining, based on the second-level corrected acceleration data and the second-level corrected angular velocity data, biped horizontal distance data of the target object's biped during the two support phases; correcting the second-level corrected acceleration data based on the biped horizontal distance data to obtain the first acceleration data; determining the second-level corrected angular velocity data as the first angular velocity data.

5. The method of claim 4, wherein, The second-level corrected acceleration data and the second-level corrected angular velocity data are used to determine the horizontal distance between the two supporting phases of the target object's feet, including: Based on the second-level corrected angular velocity data, the swing phase and the two supporting phases of the target object's feet are determined by gait phase division; Based on the second-level corrected acceleration data and the second-level corrected angular velocity data, the global acceleration in the non-gravitational earth coordinate system is calculated; The global acceleration is integrated to obtain the movement speed of the target object's feet in the movement cycle; Based on the movement speed, the positions of the target object's feet in the two supporting phases are determined, and the horizontal distance between the two supporting phases is determined based on the determined positions.

6. The method of claim 4, wherein, The second-level corrected acceleration data is corrected based on the horizontal distance between the two supporting phases to obtain the first acceleration data, including: Using the median difference method, each abnormal distance data is detected from the horizontal distance between the two supporting phases; In the second-level corrected acceleration data, the acceleration data corresponding to each abnormal distance data is determined as an abnormal acceleration data, and the remaining acceleration data is determined as normal acceleration data; In the second-level corrected acceleration data, each abnormal acceleration data is corrected to the previous normal acceleration data to obtain the first acceleration data.

7. The method of claim 4, wherein, Based on the population statistical characteristics of the target object, the first angular velocity data and the first acceleration data, the acceleration cumulative error of the target object's feet in each supporting phase is determined through geometric modeling analysis, including: The first angular velocity data and the first acceleration data are fused into a standard quaternion, and the standard quaternion is converted into Euler angles; For each supporting phase of the target object's feet, based on the corresponding angle information in the Euler angles and the population statistical characteristics, a corresponding foot change trajectory is constructed; For each supporting phase of the target object's feet, based on the corresponding foot change trajectory and the distance motion equation, a corresponding true average acceleration value is determined; For each supporting phase of the target object's feet, based on the corresponding acceleration data in the first acceleration data, a corresponding measured average acceleration value is determined, and based on the corresponding measured average acceleration value and the true average acceleration value, an acceleration cumulative error is determined.

8. A motion data correction apparatus characterized by comprising: It includes: An error estimation module is configured to estimate the acceleration cumulative error of the target object's feet in the swing phase based on the original acceleration data and the original angular velocity data of the target object's feet in the movement cycle, to obtain a first cumulative error; An error correction module is configured to predict an error value between the first cumulative error and the true acceleration cumulative error of the target object's feet in the swing phase using an error estimation model; The first cumulative error is corrected based on the error value; A data correction module is configured to correct the acceleration data of the target object's feet in the swing phase in the original acceleration data based on the corrected first cumulative error.

9. A motion data correction apparatus characterized by comprising: It includes: A processor, a memory and a communication bus; The communication bus is configured to realize communication connection between the processor and the memory. The processor is configured to execute one or more computer programs stored in the memory to realize the motion data correction method according to any one of claims 1-7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by the processor, realizes the motion data correction method according to any one of claims 1-7.

11. A computer program product comprising a computer program, characterized in that, The computer program, when executed by the processor, realizes the motion data correction method according to any one of claims 1-7. The computer program, when executed by the processor, realizes the motion data correction method according to any one of claims 1-7.

Images