Action recognition method, device and equipment

Abstract

The embodiment of the invention provides an action recognition method, device and equipment. The method comprises the following steps: acquiring acceleration information and angular velocity information of a first target object at different positions during movement; time domain features related to motion are extracted based on the acceleration information and the angular velocity information; calculating and determining a frequency domain feature and a statistical feature of the corresponding target object based on the time domain feature; constructing an initial support vector machine; training and testing the initial support vector machine based on the time domain feature, the frequency domain feature and the statistical feature to obtain an action recognition model; obtaining to-be-identified data; obtaining a motion recognition result corresponding to the second target object; determining a target feedback standard from a plurality of feedback standards stored in the field to which the corresponding to-be-recognized data belongs based on the motion recognition result; and determining actual feedback data based on the target feedback standard and the corresponding motion recognition result, and outputting the actual feedback data.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of motion recognition technology, and in particular to a motion recognition method, device and equipment. Background Art

[0002] Action recognition methods have been widely used in cross-fields such as sports training, national fitness, and artificial intelligence, such as sports posture analysis and improvement, and automatic evaluation of sports competitions. Among them, the research on running action recognition methods is relatively in-depth, but there is less research on carrying and high-platform jumping. Previous action recognition methods mainly use videos as input, and the recognition results are inaccurate and robust for complex and diverse scenes. At the same time, lighting, background interference, occlusion, etc. have a great impact on the recognition results. This method uses multiple sensors to collect human kinematic data and recognizes actions based on kinematic data. Specifically, an inertial measurement unit is used, which is also often referred to as IMU. According to the number and type of sensors included, IMU can be divided into different types, such as six-axis, nine-axis, ten-axis, etc. The six-axis IMU contains two sensors: a three-axis accelerometer and a three-axis gyroscope. The nine-axis IMU contains three sensors: a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer. Ten-axis IMU usually contains six sensors: three-axis accelerometer, three-axis gyroscope, three-axis magnetometer and temperature sensor. This patent uses a nine-axis IMU, which can collect the carrier's linear acceleration, angular velocity and the strength and direction of the geomagnetic field. According to the conversion algorithm, human motion posture data can be obtained. IMU has been used in physiological parameter monitoring and its clinical application, human motion tracking, and robust posture estimation.

[0003] In the prior art, the use of IMU includes the research of Chinese patent publication number CN114898467A, which discloses "a method, system and storage medium for human motion recognition". This patent is applied to the field of human motion recognition technology and can effectively improve the recognition performance and accuracy of human motion. However, this solution has two disadvantages. First, the hardware device used in this solution is an RGB camera, which has the phenomenon of low pixels and insufficient clarity of captured videos. Second, the scenario in which this solution is experimented and applied is only indoors and cannot be used outdoors. Another similar solution is to perform motion recognition through a neural network model based on kinematic data, but there is still a problem of poor recognition accuracy. The technologies used in the above two solutions are not applied in running, carrying and high platform jumping, and are not suitable for use in such activity scenarios.

[0004] At present, in the fields of sports training and sports competition, high-accuracy motion recognition methods are often needed to help coaches guide, monitor, analyze and study the performance of athletes. Therefore, accurate recognition of human motion in any scenario is in high demand in real-world use. Summary of the invention

[0005] The present invention provides an action recognition method, device and equipment which can quickly and accurately recognize the action of a target object, and provide matching methods for evaluating the recognized action and guiding the object to perform the action.

[0006] In order to solve the above technical problems, an embodiment of the present invention provides an action recognition method, comprising:

[0007] Collecting acceleration information and angular velocity information of the first target object at different positions during movement;

[0008] Extracting motion-related time domain features based on the acceleration information and angular velocity information;

[0009] Based on the time domain features, frequency domain features and statistical features corresponding to the target object are calculated and determined;

[0010] Construct an initial support vector machine;

[0011] Based on the time domain features, frequency domain features, and statistical features, the initial support vector machine is trained and tested to obtain an action recognition model;

[0012] obtaining data to be identified, wherein the data to be identified includes one or more of rehabilitation exercise data, running data, carrying data, and jumping data of the second target object;

[0013] Obtaining a motion recognition result corresponding to the second target object;

[0014] Determine a target feedback standard from a plurality of stored feedback standards based on the motion recognition result and the field to which the corresponding data to be recognized belongs;

[0015] Based on the target feedback standard and the corresponding motion recognition result, actual feedback data is determined and output, wherein the actual feedback data at least includes action information obtained by evaluation and recognition and information guiding the action.

[0016] As an optional embodiment, the acquisition of acceleration information and angular velocity information of the first target object at different positions during movement includes:

[0017] The acceleration information and angular velocity information of different positions of the first target object when in motion are acquired by inertial measurement units arranged at multiple different positions of the first target object.

[0018] As an optional embodiment, the method further includes:

[0019] The acceleration information and angular velocity information are subjected to denoising and outlier removal processing using a sliding window averaging method;

[0020] The acceleration information and angular velocity information after denoising and outlier removal are normalized to scale the acceleration information and angular velocity information to a range of [0, 1].

[0021] As an optional embodiment, the extracting and obtaining the motion-related time domain features based on the acceleration information and the angular velocity information includes:

[0022] Based on the processed acceleration information and angular velocity information, the mean, standard deviation, variance, root mean square, maximum acceleration value, minimum acceleration value, maximum angular velocity value, and minimum acceleration value related to the motion are extracted.

[0023] As an optional embodiment, the calculating and determining the frequency domain feature corresponding to the target object based on the time domain feature includes:

[0024] The time domain features are converted to the frequency domain based on a Fourier transform algorithm and / or a wavelet transform algorithm to obtain frequency domain features, wherein the frequency domain features include spectrum features and frequency band energy features.

[0025] As an optional embodiment, the calculating and determining the statistical features corresponding to the target object based on the time domain features includes:

[0026] The statistical distribution of the motion features corresponding to the target object and its shape features, as well as the peak value, valley value, and slope in the statistical distribution are calculated and determined based on the time domain features.

[0027] As an optional embodiment, the constructing of the initial support vector machine includes:

[0028] Construct an initial support vector machine, and set its kernel function to a radial basis function, wherein the bandwidth parameter of the radial basis function is initially set to 0.5, and the regularization coefficient is 1;

[0029] The training and testing of the initial support vector machine based on the time domain features, frequency domain features, and statistical features to obtain an action recognition model includes:

[0030] Based on the convergence parameter of 0.001 and the maximum number of iterations of 2000, the initial support vector machine is trained in combination with some of the time domain features, frequency domain features, and statistical features to adjust the parameters of the initial support vector machine to obtain an action recognition model, wherein the parameters include determining a penalty coefficient of 1.0 and a bandwidth parameter of a radial basis function of 0.7;

[0031] The action recognition model is tested based on the time domain features, frequency domain features, and statistical features of the remaining part.

[0032] As an optional embodiment, the step of determining a target feedback standard from a plurality of stored feedback standards based on the motion recognition result and the field to which the corresponding data to be recognized belongs includes:

[0033] Determine, based on the fields to which the motion recognition results and the corresponding data to be recognized belong, a target feedback standard having the same field and including all motion recognition results arranged in time series from a plurality of stored feedback standards;

[0034] The determining of actual feedback data based on the target feedback standard and the corresponding motion recognition result includes:

[0035] Determine the number and standard degree of each action in the motion recognition results;

[0036] The content of the target feedback standard is matched and adjusted based on the quantity and standard degree to obtain the actual feedback data.

[0037] Another embodiment of the present invention also provides a motion recognition device, including:

[0038] A collection module, used to collect acceleration information and angular velocity information of the first target object at different positions during movement;

[0039] An extraction module, used for extracting time domain features related to motion according to the acceleration information and angular velocity information;

[0040] A calculation module, used to calculate and determine the frequency domain features and statistical features corresponding to the target object according to the time domain features;

[0041] Building module, used to build the initial support vector machine;

[0042] A training module, used for training and testing the initial support vector machine according to the time domain features, frequency domain features, and statistical features to obtain an action recognition model;

[0043] A first acquisition module is used to obtain data to be identified, wherein the data to be identified includes one or more of rehabilitation exercise data, running data, carrying data, and jumping data of the second target object;

[0044] A second obtaining module, used to obtain a motion recognition result corresponding to the second target object;

[0045] A first determination module is used to determine a target feedback standard from a plurality of stored feedback standards according to the motion recognition result and the corresponding field to which the data to be recognized belongs;

[0046] The second determination module is used to determine actual feedback data according to the target feedback standard and the corresponding motion recognition result, and output the actual feedback data, wherein the actual feedback data at least includes action information obtained by evaluation and recognition and information guiding the action.

[0047] Another embodiment of the present invention also provides an electronic device, including:

[0048] at least one processor; and,

[0049] a memory communicatively coupled to the at least one processor;

[0050] The memory stores instructions executable by the at least one processor, and the instructions are configured to execute the action recognition method as described in any one of the embodiments above.

[0051] Based on the disclosure of the above embodiments, it can be known that the beneficial effects of the embodiments of the present invention include that the motion recognition method in the present embodiment can be applied to the fields of human-computer interaction, health detection and sports analysis, by extracting and determining the collected acceleration data and angular velocity data in the time domain, frequency domain and statistics, and then training the constructed initial support vector machine based on the extracted features to obtain a target model that can be prepared to identify the motion of the target object, and then realizing motion recognition based on the target model. At the same time, the system will analyze the recognized motion, determine the actual feedback data with guidance information and evaluation information, and then feed it back to the user for reference. The method based on the present application can not only accurately identify the motion behavior of the target object and analyze the motion behavior, but also support real-time monitoring and evaluation of the individual's motion state, posture and motion quality, and help users to conduct sports training and posture correction scientifically and rationally.

[0052] Other features and advantages of the present application will be described in the following description, and partly become apparent from the description, or be understood by practicing the present application. The purpose and other advantages of the present application can be realized and obtained by the structures specifically pointed out in the written description, claims, and drawings.

[0053] The technical solution of the present application is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The accompanying drawings are used to provide a further understanding of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings:

[0055] Figure 1 Flowchart of the action recognition method in an embodiment of the present invention.

[0056] Figure 2 4 is an application flow chart of the action recognition method in an embodiment of the present invention.

[0057] Figure 3 4 is a diagram of the relationship between the user and the acquisition device in an embodiment of the present invention.

[0058] Figure 4 4 is a structural block diagram of the action recognition device in an embodiment of the present invention. DETAILED DESCRIPTION

[0059] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings, but are not intended to limit the present invention.

[0060] It should be understood that various modifications may be made to the embodiments disclosed herein. Therefore, the following description should not be considered as limiting, but merely as an example of an embodiment. Other modifications within the scope and spirit of the present disclosure will occur to those skilled in the art.

[0061] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the present disclosure and, together with the general description of the present disclosure given above and the detailed description of the embodiments given below, serve to explain the principles of the present disclosure.

[0062] These and other characteristics of the invention will become apparent from the following description of a preferred form of embodiment given as a non-limiting example, with reference to the accompanying drawings.

[0063] It should also be understood that, although the invention has been described with reference to some specific examples, those skilled in the art will be able to realize many other equivalent forms of the invention that have the characteristics recited in the claims and are therefore within the scope of protection defined thereby.

[0064] The above and other aspects, features and advantages of the present disclosure will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings.

[0065] Specific embodiments of the present disclosure are described hereinafter with reference to the accompanying drawings; however, it should be understood that the disclosed embodiments are merely examples of the present disclosure, which may be implemented in a variety of ways. Well-known and / or repeated functions and structures are not described in detail to avoid obscuring the present disclosure with unnecessary or redundant details. Therefore, the specific structural and functional details disclosed herein are not intended to be limiting, but merely serve as a basis and representative basis for the claims to teach those skilled in the art to use the present disclosure in a variety of ways with substantially any suitable detailed structure.

[0066] This specification may use the phrases "in one embodiment," "in another embodiment," "in a further embodiment," or "in other embodiments," all of which may refer to one or more of the same or different embodiments according to the present disclosure.

[0067] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0068] like Figure 1 and Figure 2 As shown, an embodiment of the present invention provides an action recognition method, including:

[0069] S1: Collecting acceleration information and angular velocity information of a first target object at different positions during movement;

[0070] S2: extracting time domain features related to the motion based on the acceleration information and the angular velocity information;

[0071] S3: Calculate and determine frequency domain features and statistical features corresponding to the target object based on the time domain features;

[0072] S4: construct the initial support vector machine;

[0073] S5: training and testing the initial support vector machine based on the time domain features, frequency domain features, and statistical features to obtain an action recognition model;

[0074] S6: obtaining data to be identified, wherein the data to be identified includes one or more of rehabilitation exercise data, running data, carrying data, and jumping data of the second target object;

[0075] S7: Obtaining a motion recognition result corresponding to the second target object;

[0076] S8: Determine a target feedback standard from a plurality of stored feedback standards based on the motion recognition result and the corresponding field to which the data to be recognized belongs;

[0077] S9: Determine actual feedback data based on the target feedback standard and the corresponding motion recognition result, and output the actual feedback data, wherein the actual feedback data at least includes action information obtained by evaluation and recognition and information guiding the action.

[0078] The target feedback standard includes standard feedback information corresponding to the current recognition result, such as the overall scoring range of the motion posture, the range of the motion posture correction amount, and the action guidance standard corresponding to the current motion posture. The actual feedback data is adjusted and determined based on the target feedback standard and the corresponding motion recognition result, that is, each actual feedback data is matched with the corresponding action recognition result, which can positively and accurately reflect the action problem of the target object and give matching feedback.

[0079] Based on the contents disclosed in the above embodiments, it can be known that the beneficial effects of this embodiment include that the motion recognition method can be applied to the fields of human-computer interaction, health detection and sports analysis, by extracting and determining the collected acceleration data and angular velocity data in the time domain, frequency domain and statistics, and then training the constructed initial support vector machine based on the extracted features to obtain a target model that can be prepared to identify the motion of the target object, and then realizing motion recognition based on the target model. At the same time, the system will analyze the recognized motion, determine the actual feedback data with guidance information and evaluation information, and then feed it back to the user for reference. The method based on this application can not only accurately identify the motion behavior of the target object and analyze the motion behavior, but also support real-time monitoring and evaluation of the individual's motion state, posture and motion quality, and help users to conduct sports training and posture correction scientifically and rationally.

[0080] Furthermore, the acquisition of acceleration information and angular velocity information of the first target object at different positions during movement includes:

[0081] S10: Acceleration information and angular velocity information of different positions of the first target object when in motion are collected by inertial measurement units arranged at multiple different positions of the first target object.

[0082] For example, taking the first target object as a user, 9 inertial measurement unit (IMU) sensors can be selected to collect the acceleration and angular velocity kinematic data of the athlete's skeletal segments. The 9 IMU units are distributed as follows: 5 for the upper limbs (such as the left and right upper arms, left and right forearms, and the middle of the lower back), and 4 for the lower limbs (such as the left and right thighs and the left and right calves). The setting positions can be referred to Figure 3 As shown. After the motion data is collected, you can further choose to manually annotate the data. For example, when the user is running, the label of the recorded data is set as "running". In this embodiment, three labels are obtained: "running", "carrying", and "high platform jumping". Since 9 IMUs were used first, each IMU has 9 axes, which record the data of running, carrying, and high platform jumping respectively. Each state lasts for 3 seconds, the frequency of the sensor is 100Hz, and each behavior is measured ten times. Therefore, 243,000 dimensions of data are obtained.

[0083] In order to make the data more standardized, the method further includes:

[0084] S11: using a sliding window averaging method to perform denoising and outlier removal on the acceleration information and angular velocity information;

[0085] S12: normalizing the acceleration information and angular velocity information after denoising and outlier removal, so as to scale the acceleration information and angular velocity information to a range of [0, 1].

[0086] The normalization formula is:

[0087]

[0088] Where x is the original data sample, that is, the measured kinematic data, including the original acceleration data and angular velocity data, x norm is the normalized data sample.

[0089] After the data processing is completed, the time domain features related to the motion are extracted based on the acceleration information and the angular velocity information, including:

[0090] S13: extracting the mean, standard deviation, variance, root mean square, maximum acceleration value, minimum acceleration value, maximum angular velocity value, and minimum acceleration value related to the motion based on the processed acceleration information and angular velocity information.

[0091] The calculating and determining the frequency domain feature corresponding to the target object based on the time domain feature includes:

[0092] S14: Convert the time domain features to the frequency domain based on a Fourier transform algorithm and / or a wavelet transform algorithm to obtain frequency domain features, where the frequency domain features include spectrum features and frequency band energy features.

[0093] The calculating and determining the statistical features corresponding to the target object based on the time domain features includes:

[0094] S15: Determine the statistical distribution of the motion features corresponding to the target object and its shape features, as well as the peak value, valley value, and slope in the statistical distribution based on the time domain features.

[0095] For example, the mean in this embodiment is characterized by: the average value of the signal at the beginning and end of each step frequency when the user is running. The average value of the signal at the beginning and end of lifting and putting down a heavy object during transportation. The average value of the signal at the beginning and end of a high platform jump. It is used to indicate the overall intensity of the activity.

[0096] The mean can be expressed as:

[0097]

[0098] Where μ represents the mean value, N represents the total number of data points, and x i Represents the value of the i-th data point.

[0099] Standard Deviation: The dispersion of the overall signal value in each cycle of running, lifting and lowering of heavy objects during carrying, and the start and end of high platform jumping is used to describe the volatility of the signal. The standard deviation can be expressed as:

[0100]

[0101] Where σ represents the standard deviation, μ represents the mean, N represents the total number of data points, and x i Represents the value of the i-th data point.

[0102] Maximum and Minimum: The maximum and minimum values ​​of the start and end signals of each cycle of running, lifting and lowering of heavy objects during carrying, and high platform jumping can identify the extreme states of the activity.

[0103] The maximum value can be expressed as:

[0104] The minimum value can be expressed as:

[0105] Root Mean Square (RMS): The root mean square value of the overall signal in each cycle of running, lifting and lowering of heavy objects during carrying, and the start and end of high platform jumping is used to represent the overall amplitude of the signal.

[0106] The root mean square can be expressed as:

[0107]

[0108] Where RMS stands for root mean square, N stands for the total number of data points, and represents the value of the i-th data point.

[0109] Next, the signal is converted to the frequency domain by applying Fourier transform or wavelet transform, or other algorithms to extract the spectrum characteristics and frequency band energy. The Fourier transform formula is X(f) = ∫x(t)e -2πift dt, where x(t) is the sample of the signal in the time domain, and X(f) is the spectrum of the signal in the frequency domain. At the same time, the system also calculates the statistical distribution and shape characteristics of the signal, including the maximum value, minimum value, peak value, and slope, to obtain statistical characteristics.

[0110] In another embodiment, constructing an initial support vector machine includes:

[0111] S16: construct an initial support vector machine, and set its kernel function to a radial basis function, wherein the bandwidth parameter of the radial basis function is initially set to 0.5, and the regularization coefficient is 1;

[0112] The training and testing of the initial support vector machine based on the time domain features, frequency domain features, and statistical features to obtain an action recognition model includes:

[0113] S17: Based on the convergence parameter of 0.001 and the maximum number of iterations of 2000, the initial support vector machine is trained in combination with some of the time domain features, frequency domain features, and statistical features to adjust the parameters of the initial support vector machine to obtain an action recognition model, wherein the parameters include determining a penalty coefficient of 1.0 and a bandwidth parameter of a radial basis function of 0.7;

[0114] S18: Testing the action recognition model based on the time domain features, frequency domain features, and statistical features of the remaining part.

[0115] In this embodiment, the data set is divided into 30% and 70%, that is, 70% test set. 30% is validation set and test set. Among them, the validation set and test set are divided into 50% and 50%. That is, the final ratio is 70% of the data is used for training, with 218,700 data. 15% of the data is used for validation, with 109,350 data. 15% of the data is used for testing, with 109,350 data.

[0116] The kernel function is an RBF kernel. RBF (Radial Basis Function) is the abbreviation of radial basis function, which is one of the kernel functions commonly used in machine learning and pattern recognition, and is usually used for modeling nonlinear data. In this embodiment, the mathematical expression of the kernel function is:

[0117]

[0118] Among them, K(x, x′) represents the output of the RBF kernel function, x and x' are the feature vectors between the input samples, ||xx′|| represents the Euclidean distance between the two feature vectors, and σ is a control parameter of the kernel function, usually called bandwidth or kernel function width. The bandwidth determines the shape of the RBF kernel function and affects the mapping of samples in the feature space. A smaller bandwidth will make the kernel function sharper, and a larger bandwidth will make the kernel function smoother.

[0119] The application of the RBF kernel function in the support vector machine in this embodiment allows the model to perform nonlinear classification tasks in high-dimensional space because it maps the input samples to an infinite-dimensional feature space, making it easier to separate the samples in the space. This allows SVM to handle a variety of complex data distributions and perform well in many machine learning tasks.

[0120] After the initial support vector machine is built and the data for training and testing is prepared, the system will train the initial support vector machine based on the training data, and then input the test data into the trained model to further adjust the model parameters. In this embodiment, according to the results of the iteration, we finally choose the penalty coefficient to be 1.0, the bandwidth parameter σ of the RBF kernel to be 0.7, and the probability estimation is not enabled. The contraction heuristic calculation is turned on, the convergence parameter of the model construction is 0.001, and the maximum number of iterations is 2000. Of course, the above parameters are not unique and can also be other values, depending on the actual situation.

[0121] Further, as mentioned above, the target feedback standard includes standard feedback information corresponding to the current recognition result, such as the overall scoring range of the motion posture, the range of the correction amount of the motion posture, and the action guidance standard corresponding to the current motion posture. For example, different motion fields have different target feedback standards, and different action sets have different target feedback standards, such as the running field has a first feedback standard, the jumping field has a second feedback standard, and the carrying field has a third feedback standard. The first feedback standard, the second feedback standard, and the third feedback standard can also be divided into object types, such as the first feedback standard, the second feedback standard, and the third feedback standard all have two different standards corresponding to human bodies and mechanical objects. Alternatively, if the recognized action includes running and jumping, then the fourth feedback standard corresponds to it, and so on. Multiple feedback standards can be stored in the system. When the target feedback standard is determined from the multiple feedback standards stored in the field to which the corresponding data to be identified belongs based on the motion recognition result, it includes:

[0122] S19: Based on the motion recognition result and the corresponding field to which the to-be-recognized data belongs, a target feedback standard having the same field and including all motion types arranged in time series is determined from a plurality of stored feedback standards.

[0123] That is, first determine the corresponding action field according to the action recognition results, such as running, carrying, rehabilitation exercises, etc., then determine the candidate feedback standards based on the action field, and then determine the target feedback standards that include all movement types arranged in time series based on the action recognition results. That is, the target feedback standards include all types of action recognition results of the target object that appear in time series. For example, the target object runs first, then jumps, and then continues to run, etc. Based on this, the target feedback standards also include information on this kind of movement combination, and have evaluation criteria and action guidance information for different time periods and different types of movements.

[0124] Next, determining actual feedback data based on the target feedback standard and the corresponding motion recognition result includes:

[0125] S20: Determine the number and standard degree of each action in the motion recognition result;

[0126] S21: Matching and adjusting the content of the target feedback standard based on the quantity and standard degree to obtain the actual feedback data.

[0127] For example, the system will identify and determine the number of occurrences of each action in each motion recognition result, and compare the parameters of each recognized action based on the pre-stored standard action parameters to determine its standardization. The standardization of the same action can then be processed to obtain the mean standardization. After the number and mean standardization of each action are determined, the system will adjust the content of the target feedback standard based on the number and mean standardization, such as adjusting which type of action guidance is emphasized, the specific guidance content, for example, it may include the amount of leg height adjustment, etc., to obtain actual feedback data. The actual feedback data obtained matches the user's actual movement situation, providing a positive and valuable reference for the user's actual movement.

[0128] like Figure 4 As shown, another embodiment of the present invention also provides a motion recognition device 100, including:

[0129] A collection module, used to collect acceleration information and angular velocity information of the first target object at different positions during movement;

[0130] An extraction module, used for extracting time domain features related to motion according to the acceleration information and angular velocity information;

[0131] A calculation module, used to calculate and determine the frequency domain features and statistical features corresponding to the target object according to the time domain features;

[0132] Building module, used to build the initial support vector machine;

[0133] A training module, used for training and testing the initial support vector machine according to the time domain features, frequency domain features, and statistical features to obtain an action recognition model;

[0134] A first acquisition module is used to obtain data to be identified, wherein the data to be identified includes one or more of rehabilitation exercise data, running data, carrying data, and jumping data of the second target object;

[0135] A second obtaining module, used to obtain a motion recognition result corresponding to the second target object;

[0136] A first determination module is used to determine a target feedback standard from a plurality of stored feedback standards according to the motion recognition result and the corresponding field to which the data to be recognized belongs;

[0137] The second determination module is used to determine actual feedback data according to the target feedback standard and the corresponding motion recognition result, and output the actual feedback data, wherein the actual feedback data at least includes action information obtained by evaluation and recognition and information guiding the action.

[0138] As an optional embodiment, the acquisition of acceleration information and angular velocity information of the first target object at different positions during movement includes:

[0139] The acceleration information and angular velocity information of different positions of the first target object when in motion are acquired by inertial measurement units arranged at multiple different positions of the first target object.

[0140] As an optional embodiment, the method further includes:

[0141] The acceleration information and angular velocity information are subjected to denoising and outlier removal processing using a sliding window averaging method;

[0142] The acceleration information and angular velocity information after denoising and outlier removal are normalized to scale the acceleration information and angular velocity information to a range of [0, 1].

[0143] As an optional embodiment, the extracting and obtaining the motion-related time domain features based on the acceleration information and the angular velocity information includes:

[0144] Based on the processed acceleration information and angular velocity information, the mean, standard deviation, variance, root mean square, maximum acceleration value, minimum acceleration value, maximum angular velocity value, and minimum acceleration value related to the motion are extracted.

[0145] As an optional embodiment, the calculating and determining the frequency domain feature corresponding to the target object based on the time domain feature includes:

[0146] The time domain features are converted to the frequency domain based on a Fourier transform algorithm and / or a wavelet transform algorithm to obtain frequency domain features, wherein the frequency domain features include spectrum features and frequency band energy features.

[0147] As an optional embodiment, the calculating and determining the statistical features corresponding to the target object based on the time domain features includes:

[0148] The statistical distribution of the motion features corresponding to the target object and its shape features, as well as the peak value, valley value, and slope in the statistical distribution are calculated and determined based on the time domain features.

[0149] As an optional embodiment, the constructing of the initial support vector machine includes:

[0150] Construct an initial support vector machine, and set its kernel function to a radial basis function, wherein the bandwidth parameter of the radial basis function is initially set to 0.5, and the regularization coefficient is 1;

[0151] The training and testing of the initial support vector machine based on the time domain features, frequency domain features, and statistical features to obtain an action recognition model includes:

[0152] Based on the convergence parameter of 0.001 and the maximum number of iterations of 2000, the initial support vector machine is trained in combination with some of the time domain features, frequency domain features, and statistical features to adjust the parameters of the initial support vector machine to obtain an action recognition model, wherein the parameters include determining a penalty coefficient of 1.0 and a bandwidth parameter of a radial basis function of 0.7;

[0153] The action recognition model is tested based on the time domain features, frequency domain features, and statistical features of the remaining part.

[0154] As an optional embodiment, the step of determining a target feedback standard from a plurality of stored feedback standards based on the motion recognition result and the field to which the corresponding data to be recognized belongs includes:

[0155] Determine, based on the fields to which the motion recognition results and the corresponding data to be recognized belong, a target feedback standard having the same field and including all motion recognition results arranged in time series from a plurality of stored feedback standards;

[0156] The determining of actual feedback data based on the target feedback standard and the corresponding motion recognition result includes:

[0157] Determine the number and standard degree of each action in the motion recognition results;

[0158] Based on the quantity and standard degree, the content of the target feedback standard is matched and adjusted to obtain the actual feedback data.

[0159] Another embodiment of the present invention further provides an electronic device, including:

[0160] at least one processor; and,

[0161] a memory communicatively coupled to the at least one processor;

[0162] The memory stores instructions executable by the at least one processor, and the instructions are configured to execute the action recognition method as described in any one of the embodiments above.

[0163] Another embodiment of the present invention further provides a storage medium, which includes a stored program, wherein when the program is running, a device including the storage medium is controlled to execute the high-dimensional prediction and classification method for class-imbalanced data as described in any of the embodiments above.

[0164] The embodiment of the present invention also provides a computer program product, which is tangibly stored on a computer-readable medium and includes computer-readable instructions, and when the computer-executable instructions are executed, at least one processor executes the action recognition method in the embodiment described above. It should be understood that each solution in this embodiment has the corresponding technical effect in the above method embodiment, which will not be repeated here.

[0165] It should be noted that the computer storage medium of the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access storage medium (RAM), a read-only storage medium (ROM), an erasable programmable read-only storage medium (EPROM or flash memory), an optical fiber, a portable compact disk read-only storage medium (CD-ROM), an optical storage medium, a magnetic storage medium, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate or transmit a program configured to be used by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, antenna, optical cable, RF, etc., or any suitable combination of the above.

[0166] The above embodiments are only exemplary embodiments of the present invention and are not intended to limit the present invention. The protection scope of the present invention is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present invention within the essence and protection scope of the present invention, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present invention.

Claims

1. A method for motion recognition, characterized in that: include: Collecting acceleration information and angular velocity information of the first target object at different positions during movement; Extracting motion-related time domain features based on the acceleration information and angular velocity information; Based on the time domain features, frequency domain features and statistical features corresponding to the target object are calculated and determined; Construct an initial support vector machine; Training and testing the initial support vector machine based on the time domain features, frequency domain features, and statistical features to obtain an action recognition model; obtaining data to be identified, wherein the data to be identified includes one or more of rehabilitation exercise data, running data, carrying data, and jumping data of the second target object; Obtaining a motion recognition result corresponding to the second target object; Determine a target feedback standard from a plurality of stored feedback standards based on the motion recognition result and the field to which the corresponding data to be recognized belongs; Based on the target feedback standard and the corresponding motion recognition result, actual feedback data is determined and output, wherein the actual feedback data at least includes action information obtained by evaluation and recognition and information guiding the action.

2. The action recognition method according to claim 1, characterized in that: The acquisition of acceleration information and angular velocity information of the first target object at different positions during movement includes: The acceleration information and angular velocity information of different positions of the first target object when in motion are acquired by inertial measurement units arranged at multiple different positions of the first target object.

3. The action recognition method according to claim 1, characterized in that: The method further comprises: The acceleration information and angular velocity information are subjected to denoising and outlier removal processing using a sliding window averaging method; The acceleration information and angular velocity information after denoising and outlier removal are normalized to scale the acceleration information and angular velocity information to a range of [0, 1].

4. The action recognition method according to claim 3, characterized in that: The extracting of motion-related time domain features based on the acceleration information and the angular velocity information includes: Based on the processed acceleration information and angular velocity information, the mean, standard deviation, variance, root mean square, maximum acceleration value, minimum acceleration value, maximum angular velocity value, and minimum acceleration value related to the motion are extracted.

5. The action recognition method according to claim 1, characterized in that: The calculating and determining the frequency domain feature corresponding to the target object based on the time domain feature includes: The time domain features are converted to the frequency domain based on a Fourier transform algorithm and / or a wavelet transform algorithm to obtain frequency domain features, wherein the frequency domain features include spectrum features and frequency band energy features.

6. The action recognition method according to claim 1, characterized in that: The calculating and determining the statistical features corresponding to the target object based on the time domain features includes: The statistical distribution of the motion features corresponding to the target object and its shape features, as well as the peak value, valley value, and slope in the statistical distribution are calculated and determined based on the time domain features.

7. The action recognition method according to claim 1, characterized in that: The constructing of the initial support vector machine comprises: Construct an initial support vector machine, and set its kernel function to a radial basis function, wherein the bandwidth parameter of the radial basis function is initially set to 0.5, and the regularization coefficient is 1; The training and testing of the initial support vector machine based on the time domain features, frequency domain features, and statistical features to obtain an action recognition model includes: Based on the convergence parameter of 0.001 and the maximum number of iterations of 2000, the initial support vector machine is trained in combination with some of the time domain features, frequency domain features, and statistical features to adjust the parameters of the initial support vector machine to obtain an action recognition model, wherein the parameters include determining a penalty coefficient of 1.0 and a bandwidth parameter of a radial basis function of 0.7; The action recognition model is tested based on the time domain features, frequency domain features, and statistical features of the remaining part.

8. The action recognition method according to claim 1, characterized in that: The step of determining a target feedback standard from a plurality of stored feedback standards based on the motion recognition result and the field to which the corresponding data to be recognized belongs includes: Determine, based on the motion recognition result and the corresponding field to which the to-be-recognized data belongs, a target feedback standard having the same field and including all motion types arranged in time series from a plurality of stored feedback standards; The determining of actual feedback data based on the target feedback standard and the corresponding motion recognition result includes: Determine the number and standard degree of each action in the motion recognition results; Based on the quantity and standard degree, the content of the target feedback standard is matched and adjusted to obtain the actual feedback data.

9. A motion recognition device, characterized in that: include: A collection module, used to collect acceleration information and angular velocity information of the first target object at different positions during movement; An extraction module, used for extracting time domain features related to motion according to the acceleration information and angular velocity information; A calculation module, used to calculate and determine the frequency domain features and statistical features corresponding to the target object according to the time domain features; Building module, used to build the initial support vector machine; A training module, used for training and testing the initial support vector machine according to the time domain features, frequency domain features, and statistical features to obtain an action recognition model; A first acquisition module is used to obtain data to be identified, wherein the data to be identified includes one or more of rehabilitation exercise data, running data, carrying data, and jumping data of the second target object; A second obtaining module, used to obtain a motion recognition result corresponding to the second target object; A first determination module is used to determine a target feedback standard from a plurality of stored feedback standards according to the motion recognition result and the corresponding field to which the data to be recognized belongs; The second determination module is used to determine actual feedback data according to the target feedback standard and the corresponding motion recognition result, and output the actual feedback data, wherein the actual feedback data at least includes action information obtained by evaluation and recognition and information guiding the action.

10. An electronic device, comprising: at least one processor; as well as, a memory communicatively coupled to the at least one processor; The memory stores instructions executable by the at least one processor, and the instructions are configured to execute the action recognition method as described in any one of claims 1-8.

Images