A Motion Posture Recognition Method and Recognition System Based on UWB Positioning and Tracking

Abstract

The present invention relates to the technical field of motion posture recognition, and particularly relates to a motion posture recognition method and recognition system based on UWB positioning and tracking. The method includes obtaining UWB positioning data of a target user, collecting motion data of multiple parts of the target user, constructing a motion trajectory, extracting time-domain features from the collected motion data, establishing an action recognition model, outputting an action recognition result including arm-swinging actions and gait actions, constructing a multi-modal association model with time synchronization and fusion to capture the dynamic relationship between the motion trajectory and the action recognition result, and generating a motion posture and trajectory result. The generated motion posture and trajectory result is transmitted to the host computer software for parsing and generating dynamically displayed posture and trajectory information. The present invention makes full use of time series information to construct high-precision motion postures and trajectory results, and can perform action recognition and trajectory tracking more accurately.

Description

Technical Field

[0001] The present invention relates to the technical field of motion posture recognition, and in particular to a motion posture recognition method and recognition system based on UWB positioning and tracking. Background Art

[0002] With the help of computer vision or wearable devices, sports posture recognition technology can accurately measure and analyze whether the athlete's movements are correct and standardized by collecting sports action data and using machine learning methods. It can also automatically recognize movements and analyze human movement postures by capturing human movement information. This technology has been widely used in rehabilitation training, health monitoring for the elderly, somatosensory games, film and television dramas, sports, and other fields, providing people with more intelligent, personalized, and convenient services and experiences.

[0003] In the current positioning tracking and motion posture recognition methods based on ultra-wideband (UWB), although the introduction of UWB positioning can provide high-precision spatial position data, it is often difficult to achieve comprehensive motion analysis by relying solely on positioning information or motion recognition data. In addition, existing running posture recognition methods usually only focus on single motion features (such as arm swing or gait), lack multi-part motion fusion and trajectory analysis, and have limited functions. At the same time, in dynamic motion scenes, how to achieve real-time fusion and efficient analysis of multimodal data is still a key challenge that determines system performance and applicability.

[0004] In the prior art, the invention patent with the publication number CN110694252A discloses a wireless positioning and tracking system based on UWB positioning, and proposes to realize the precise positioning and tracking of the target object by the UWB signal interaction between the base station and the target object, combining one-dimensional and two-dimensional positioning technologies. However, this method only focuses on the target positioning and tracking based on the UWB signal, without visual display, and cannot pay attention to the position of the target object in real time; the invention patent with the publication number CN116112868A discloses an indoor positioning device and its usage method based on ultra-wideband positioning technology, and proposes to use the triangulation positioning technology to realize indoor precise positioning through devices such as UWB base stations, synchronization base stations, and computing engines. However, this method mainly focuses on the precise positioning in the exhibition hall and the visualization of the personnel position, without involving the real-time recognition and analysis of the human body movement posture, and the function range is relatively limited; the invention patent with the publication number CN110694252A discloses a running posture detection method based on a six-axis sensor, and proposes to collect acceleration and angular velocity data through the six-axis sensor, combine gait cycle detection and eigenvalue calculation, and use the Softmax classifier to identify the running landing mode. However, this method only analyzes the gait movement, cannot show the comprehensiveness of the human body movement posture, and does not pay attention to the position and posture changes of the human body during the movement process, lacking the comprehensive utilization of multi-dimensional sensor features; the invention patent with the publication number CN109256187A discloses a running posture suggestion system and method based on multi-mode motion sensor data, and proposes to collect acceleration and angular velocity data through the multi-mode motion sensor, analyze in combination with historical data, generate running posture suggestions and display them through a display screen. However, this method does not combine positioning information, cannot generate a motion trajectory or real-time spatial dynamic analysis, and the function range is relatively limited.

[0005] In summary, in the existing positioning and tracking and motion posture recognition methods, only a single data source is concerned through technologies such as UWB positioning or sensor technology perception, lacking the comprehensive utilization of the action data of multiple parts of the human body, and unable to realize the comprehensive action posture analysis and motion trajectory display. Therefore, it is necessary to perform multi-mode information fusion of action posture recognition by combining UWB positioning technology and sensors. Summary of the Invention

[0006] The technical problem to be solved in the embodiments of the present invention is to provide a motion posture recognition method and recognition system based on UWB positioning and tracking to solve the problem in the prior art that comprehensive action posture analysis and motion trajectory display cannot be realized.

[0007] The present invention discloses a motion posture recognition method based on UWB positioning and tracking, including:

[0008] Obtaining the positioning data of the target user based on UWB positioning, and collecting the action data of multiple parts of the target user;

[0009] Construct a motion trajectory based on the obtained positioning data, and extract time-domain features from the collected action data;

[0010] Establish an action recognition model based on the extracted time-domain features, input the collected action data into the action recognition model, and output an action recognition result including arm-swinging actions and gait actions;

[0011] Construct a multi-modal association model with time synchronization and fusion based on the motion trajectory and the action recognition result, capture the dynamic relationship between the motion trajectory and the action recognition result from the multi-modal association model, and generate a motion posture and trajectory result;

[0012] Transmit the generated motion posture and trajectory result to the host computer software for parsing to obtain the dynamically displayed posture and trajectory information.

[0013] Optionally, the obtaining the positioning data of the target user based on UWB positioning includes:

[0014] Based on a scenario covered by multiple UWB base stations, calculate the time difference between the target user reaching two UWB base stations according to the UWB positioning signal sent by the device worn by the target user. The functional expression for calculating the time difference is:

[0015]

[0016] In the formula, represents the time difference, represents the time when the signal reaches base station m, represents the time when the signal reaches base station k;

[0017] Obtain the speed of the target user reaching the UWB base station, and calculate the distance between the target user and the UWB base station according to the time difference. The functional expression for calculating the distance is:

[0018]

[0019] In the formula, represents the distance difference, represents the speed of the target user reaching two UWB base stations;

[0020] Obtain the base station position coordinates of all UWB base stations. According to all the base station position coordinates and the distances between the target user and the corresponding UWB base stations, construct a system of equations, and use the least squares method to solve the system of equations to obtain the positioning data, which includes the user position coordinates of the target user at different times.

[0021] Optionally, the constructing a motion trajectory according to the obtained positioning data includes:

[0022] Filter all the user position coordinates according to the user position coordinates at different times;

[0023] Sort the filtered user position coordinates in time series, and perform smooth interpolation between the user position coordinates corresponding to adjacent times by using a trajectory fitting algorithm to generate a continuous trajectory curve to construct the motion trajectory.

[0024] Optionally, the establishing an action recognition model according to the extracted time domain features includes:

[0025] Filter the action data according to the collected action data, and perform signal drift correction on the filtered action data;

[0026] Based on signal framing processing, select a window length and an overlap rate to segment the action data after signal drift correction to obtain multiple windows containing action cycle data;

[0027] Select features associated with arm swing actions and gait actions from the action data, and statistically analyze the selected features in the action cycle data in the window;

[0028] Calculate the time domain features of the selected features according to the statistically selected features, and extract the time domain features from each window;

[0029] Screen out sample data from all the extracted time domain features, and divide the sample data into a training set and a test set;

[0030] Select a network architecture to construct a basic classification model, and batch input the time domain features in the training set into the basic classification model for training to obtain the action recognition model;

[0031] Batch input the time domain features in the test set into the trained action recognition model for testing, and update the model parameters of the action recognition model through an iterative optimization algorithm.

[0032] Optionally, the constructing a time-synchronized fusion multi-modal association model according to the motion trajectory and the action recognition result includes:

[0033] Add timestamps to the motion trajectory and the action recognition result respectively;

[0034] Calculate the sampling period of the positioning data and the sampling period of the action recognition result. The function expressions of the sampling periods of the positioning data and the action recognition result are respectively:

[0035]

[0036]

[0037] In the formula, represents the sampling period of the positioning data, represents the sampling frequency of the positioning data, represents the sampling period of the action recognition result, represents the sampling frequency of the action recognition result;

[0038] According to the consistency between the sampling period of the positioning data and the sampling period of the action recognition result, perform time alignment on the timestamps of the motion trajectory and the action recognition result, and generate a time synchronization sequence including multiple time points within the same time window. Each of the time points corresponds to the time-synchronized positioning data and the action recognition result. The functional expression of the time synchronization sequence is:

[0039]

[0040] In the formula, represents the time synchronization sequence, represents the nth time point;

[0041] Construct a time synchronization and fusion multi-modal association model according to the positioning data and the action recognition result corresponding to all time points. The functional expression of the multi-modal association model is:

[0042]

[0043] In the formula, represents the multi-modal information vector generated by fusing the positioning data and the action recognition result corresponding to the nth time point, represents the action recognition result corresponding to the nth time point, represents the positioning data corresponding to the nth time point.

[0044] Optionally, the time alignment of the timestamps of the motion trajectory and the action recognition result includes:

[0045] When the sampling period of the positioning data is the same as the sampling period of the action recognition result, unify the timestamps of the motion trajectory and the action recognition result to the same reference time and convert them to absolute timestamps for alignment;

[0046] When the sampling period of the positioning data is different from the sampling period of the action recognition result, find the two timestamps closest to a certain timestamp of the action recognition result in the timestamps of the motion trajectory. The functional expression for finding the timestamps is:

[0047]

[0048] Wherein, represents the i-th timestamp added to the action recognition result, represents the j-th timestamp added to the motion trajectory, represents the (j + 1)-th timestamp added to the motion trajectory;

[0049] According to the two found timestamps, linear interpolation is used to calculate the positioning data corresponding to a certain timestamp of the action recognition result, and the calculation function expression of the positioning data is:

[0050]

[0051] Wherein, represents the positioning data corresponding to the i-th timestamp, represents the positioning data corresponding to the j-th timestamp; represents the positioning data corresponding to the (j + 1)-th timestamp;

[0052] According to the action recognition result and the positioning data corresponding to the same timestamp, the timestamps of the motion trajectory and the action recognition result are aligned.

[0053] The present invention also discloses an identification system, which adopts the above-mentioned motion posture identification method based on UWB positioning and tracking. The identification system includes:

[0054] A motion posture tracking module, configured to obtain the positioning data of the target user based on UWB positioning and collect the action data of multiple parts of the target user;

[0055] A data processing module, configured to construct a motion trajectory according to the obtained positioning data and extract time-domain features from the collected action data;

[0056] An action recognition module, configured to establish an action recognition model according to the extracted time-domain features, input the collected action data into the action recognition model, and output an action recognition result including arm-swinging actions and gait actions;

[0057] A joint recognition module, configured to construct a time-synchronized fusion multi-modal association model according to the motion trajectory and the action recognition result, capture the dynamic relationship between the motion trajectory and the action recognition result from the multi-modal association model, and generate a motion posture and trajectory result;

[0058] A visualization module, configured to transmit the generated motion posture and trajectory result to the host computer software.

[0059] Optionally, the motion posture tracking module is connected to a wearing device, which includes a wearing shell, a display unit arranged on the wearing shell, and a main control chip, a power supply unit, a UWB positioning unit, and a posture sensing unit integrated in the wearing shell. The display unit, the power supply unit, the UWB positioning unit, and the posture sensing unit are respectively electrically connected to the main control chip, so that the UWB positioning unit collects and sends the UWB positioning signal of the target user, and the posture sensing unit collects and sends the motion data of the wearing part of the target user.

[0060] The present invention also discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned motion posture recognition method based on UWB positioning and tracking are realized.

[0061] The present invention also discloses a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned motion posture recognition method based on UWB positioning and tracking are realized.

[0062] Compared with the prior art, the beneficial effects of the motion posture recognition method and recognition system based on UWB positioning and tracking provided by the embodiments of the present invention are as follows:

[0063] By combining UWB positioning and multi-part motion sensing data, a motion trajectory is constructed according to the obtained target user positioning data. At the same time, time-domain features of the collected motion data are extracted, an action recognition model is established according to the extracted time-domain features, and the collected motion data is input into the trained action recognition model to obtain action recognition results including arm-swinging actions and gait actions. By constructing a time-synchronized fusion multi-modal association model, the dynamic relationship between the motion trajectory and the action recognition result is captured, and accurate motion posture and trajectory results are generated. Thus, the time series information is fully utilized to construct high-precision motion posture and trajectory results, which can perform action recognition and trajectory tracking more accurately and has broad application potential in the fields of sports health monitoring, behavior analysis, and intelligent interaction. Description of the Drawings

[0064] The technical solutions of the present invention will be further described in detail below in conjunction with the drawings. In the drawings:

[0065] Figure 1 It is a schematic block diagram of the steps of the motion posture recognition method based on UWB positioning and tracking provided by the embodiments of the present invention;

[0066] Figure 2 It is a schematic diagram of the process of collecting the positioning data and motion data of the target user provided by the embodiments of the present invention;

[0067] Figure 3 It is a schematic structural diagram of the LMS adaptive filtering algorithm provided by an embodiment of the present invention;

[0068] Figure 4 It is a schematic diagram of window sliding for windowing processing of motion data provided by an embodiment of the present invention;

[0069] Figure 5 It is a schematic diagram of the convolutional neural network algorithm model provided by an embodiment of the present invention;

[0070] Figure 6 It is a schematic diagram of the accuracy comparison of the iterative algorithm for updating the action recognition model provided by an embodiment of the present invention;

[0071] Figure 7 It is a schematic diagram of the training loss comparison of the iterative algorithm for updating the action recognition model provided by an embodiment of the present invention;

[0072] Figure 8 It is a schematic structural diagram of the wearing device provided by an embodiment of the present invention.

[0073] Each reference numeral in the figure is:

[0074] 1. UWB base station; 2. Computing server; 3. Host computer software; 4. Motion attitude tracking module; 41. Display unit; 42. Main control chip; 43. Power supply unit; 44. UWB positioning unit; 45. Attitude sensing unit. Detailed implementation manners

[0075] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. Now, with reference to the accompanying drawings, the preferred embodiments of the present invention will be described in detail.

[0076] The present invention discloses a motion attitude recognition method based on UWB positioning and tracking, as Figure 1 shown, including:

[0077] S1. Obtain the positioning data of the target user based on UWB positioning, and collect the motion data of multiple parts of the target user;

[0078] S2. Construct a motion trajectory according to the obtained positioning data, and extract time-domain features from the collected motion data;

[0079] S3. Establish an action recognition model according to the extracted time-domain features, input the collected motion data into the action recognition model, and output the action recognition result including arm swing action and gait action;

[0080] S4. Construct a time-synchronized and fused multi-modal association model based on the motion trajectory and action recognition result, capture the dynamic relationship between the motion trajectory and the action recognition result from the multi-modal association model, and generate the motion posture and trajectory result;

[0081] S5. Transmit the generated motion posture and trajectory result to the host computer software 3 for parsing and generating the posture and trajectory information for dynamic display.

[0082] Through the implementation of the above embodiments of the motion posture recognition method, combining UWB positioning and multi-site motion sensing data, a motion trajectory is constructed according to the obtained target user positioning data. At the same time, time-domain features are extracted from the collected action data, an action recognition model is established according to the extracted time-domain features, and the collected action data is input into the trained action recognition model to obtain the action recognition result including arm-swinging actions and gait actions. By constructing a time-synchronized and fused multi-modal association model, the dynamic relationship between the motion trajectory and the action recognition result is captured, and accurate motion posture and trajectory results are generated. Thus, by making full use of time series information to construct high-precision motion posture and trajectory results, action recognition and trajectory tracking can be performed more accurately, and it has broad application potential in the fields of sports health monitoring, behavior analysis, and intelligent interaction.

[0083] Further, based on UWB positioning, the positioning data of the target user is obtained, including:

[0084] Based on the coverage scenario of multiple UWB base stations 1, according to the UWB positioning signal sent by the device worn by the target user, calculate the time difference between the target user arriving at two UWB base stations 1. The function expression for calculating the time difference is:

[0085]

[0086] In the formula, represents the time difference, represents the time when the signal arrives at base station m, represents the time when the signal arrives at base station k;

[0087] Obtain the speed of the target user reaching the UWB base station 1, and calculate the distance between the target user and the UWB base station 1 according to the time difference. The function expression for calculating the distance is:

[0088]

[0089] In the formula, represents the distance difference, represents the speed of the target user reaching two UWB base stations;

[0090] Obtain the base station position coordinates of all UWB base stations 1. According to all the base station position coordinates and the distances between the target user and the corresponding UWB base stations 1, construct a system of equations, and use the least squares method to solve the system of equations to obtain the positioning data, where the positioning data includes the user position coordinates of the target user at different times.

[0091] Through the implementation of the embodiments of the above-mentioned motion gesture recognition method, as Figure 2 shown, the wearable devices worn on the upper arm and ankle of the target user can be used to collect the motion data of multiple parts such as the target user's arm swing and gait, and obtain the UWB positioning signal where the target user is located. By sending and receiving UWB positioning signals through the wearable devices, establish communication with the UWB base stations 1 arranged at known fixed positions, and use positioning algorithms such as time of flight or time difference of arrival to accurately calculate the three-dimensional position coordinates of the target user in space. That is, assume there are n UWB base stations 1 with known positions, and the position coordinates of the kth UWB base station 1 are , use the distance differences between multiple UWB base stations 1 and the target user to construct a system of equations, and use the least squares method to solve the system of equations to obtain the position coordinates of the target user to complete the positioning and tracking of the target user.

[0092] Further, construct a motion trajectory based on the obtained positioning data, including:

[0093] Perform filtering processing on all user position coordinates according to the user position coordinates at different times;

[0094] Perform time series sorting on the filtered user position coordinates, and use the trajectory fitting algorithm to perform smooth interpolation processing between the user position coordinates corresponding to adjacent times to generate a continuous trajectory curve to construct the motion trajectory.

[0095] Through the implementation of the embodiments of the above-mentioned motion gesture recognition method, for the user position coordinates obtained at different times , preferably use the Kalman filter algorithm for denoising and outlier processing. Then, perform time series sorting on the positioning data containing user position coordinates, and use the trajectory fitting algorithm to smooth the coordinate points. Preferably use cubic spline interpolation to perform smooth interpolation between adjacent coordinate points to generate a continuous trajectory curve to construct the motion trajectory of the target user, thereby completing trajectory fitting.

[0096] Further, establish an action recognition model according to the extracted time domain features, including:

[0097] According to the collected action data, perform filtering processing on the action data and perform signal drift correction on the filtered action data;

[0098] Based on signal frame processing, select the window length and overlap rate to segment and intercept the action data after signal drift correction, and obtain multiple windows containing action cycle data;

[0099] Select the features associated with the arm swing action and gait action from the action data, and statistically analyze the selected features in the action cycle data in the window;

[0100] Calculate the time-domain features of the selected features according to the statistically selected features, and extract the time-domain features from each window;

[0101] Screen out the sample data from all the extracted time-domain features, and divide the sample data into a training set and a test set;

[0102] Select a network architecture to construct a basic classification model, batch input the time-domain features in the training set into the basic classification model for training, and obtain an action recognition model;

[0103] Batch input the time-domain features in the test set into the trained action recognition model for testing, and update the model parameters of the action recognition model through an iterative optimization algorithm.

[0104] Through the implementation of the above embodiments of the motion posture recognition method, preferably use the LMS (Least Mean Squares) adaptive filtering algorithm to filter the action data. This algorithm will adaptively adjust the structural parameters of the filter according to the minimum mean square error between the output signal and the desired signal so that the filtering algorithm can adapt to the time-domain changes of random signals.

[0105] Assume the input signal is:

[0106]

[0107] The output signal of the filter is , is the filter weight coefficient, then:

[0108]

[0109]

[0110] In the formula, L represents the filter order, represents the input signal at time represents the vector transpose operation, represents the th weight coefficient of the adaptive filter, represents the index of the sum of the weight coefficients,​ represents the transpose of the filter weight coefficient vector, and represents the last element in the filter weight coefficient vector;

[0111] Suppose is the error signal, is the desired signal, then there is:

[0112]

[0113] Define the cost function of the filter as the mean square value of the error signal , then:

[0114]

[0115] Adopt the steepest descent method to adjust the weight vector along the negative gradient direction of the performance surface , and solve the optimal weight vector, and we can get:

[0116]

[0117] Then the weight vector has the following update formula:

[0118]

[0119] In the formula, , represents the step size factor, E represents the mean square value of the error signal, represents the gradient of the cost function, represents the partial derivative.

[0120] As described above, the schematic structural diagram of the LMS adaptive filtering algorithm provided by the embodiment of the present invention is as shown in Figure 3 .

[0121] In addition, considering the continuity, periodicity, and integrity of the motion actions, the window length is preferably set to N = 200 sampling data points, the overlap rate of the window is set to 50%, and an example diagram of the sliding window in the windowing process is as shown in Figure 4As shown in the figure. The horizontal axis represents the sampling points, which refer to each time point in the time series data, and each vertical line represents a sampling moment; the vertical axis represents the acceleration value, that is, the acceleration value of each sampling point has a corresponding height on this axis. The process of the sliding window is to first define the size of a window. For example, the window contains five sampling points. Slide this window from the beginning of the data sequence, moving one sampling point each time. After each window slide, the sampling points within the window form a new data subset. Thus, through the sliding window, the local features of the data sequence can be analyzed. Thus, by extracting the time-domain features of the motion data in each window, the differential information between different swing-arm motions can be established. Statistical methods are used for time-domain feature extraction, and considering the time-domain features of the data in each motion window, features associated with swing-arm motions and gait motions are selected from the motion data information, such as: Y-axis acceleration, X-axis angular velocity, Z-axis angular velocity, pitch angle, roll angle as features. During the time-domain feature extraction process, time-domain features such as mean, standard deviation, kurtosis, and skewness are selected for extraction. The algorithm complexity of these time-domain features is relatively low and is easy to implement within wearable devices.

[0122] More preferably, classification machine learning algorithms such as XGBoost (Extreme Gradient Boosting), Bayes (Bayes Classifier), CNN (Convolutional Neural Network), SVM (Support Vector Machine), etc. can be selected as the basic classification model to train the training set, and the hyperparameters and model structure of the algorithm can be adjusted to optimize the model performance. Among them, the schematic diagram of the convolutional neural network algorithm model is as Figure 5 shown. After the collected motion data is filtered, windowed, and feature-extracted, it is input into the convolutional neural network algorithm model for training. Among them, the input layer is designed according to the shape of the motion data after processing such as filtering. If the data is two-dimensional, the input layer should accept two-dimensional input. The convolutional layer extracts local features from the input data by adding one or more convolutional layers, and the appropriate convolutional kernel size and number can be selected. A pooling layer (such as max pooling) is added after each convolutional layer to reduce the spatial dimension of the features and reduce the computational amount. Usually, multiple convolutional layers and pooling layers are stacked to form multiple convolutional blocks. After the convolutional layer, one or more fully connected layers are added to learn high-level features and perform classification. The number of neurons in the fully connected layer depends on the needs of the output layer. For example, for a multi-classification problem, the number of neurons in the last layer is equal to the number of classes. After the last fully connected layer, the Softmax function (normalized exponential function) is used to convert the output into a probability distribution. Thus, high-precision motion recognition is achieved using the classification algorithm.

[0123] At this time, it is preferable to select 80% of the sample data as the training set and 20% as the test set. There are 3,600 groups of training sets and 900 groups of test sets for the training and testing of the model. Commonly used iterative algorithms such as Momentum (Momentum Gradient Descen, improved gradient descent method), RMSProp (Root Mean Square Propagation, adaptive learning rate optimization algorithm), and Adam (Adaptive Moment Estimation, adaptive learning rate optimization algorithm) are selected for testing respectively. The comparison of the accuracy (Accuracy) and training loss (Training Function) of the provided iterative algorithms is as Figure 5 and Figure 6 shown. Through the above steps, a trained action recognition model is finally obtained. Among them, the division of the sample data comprehensively considers randomness and representativeness, and then divides the training set and the test set to avoid affecting the model performance due to data distribution deviation.

[0124] The collected action data is input into the action recognition model, and the action recognition results including arm swing actions and gait actions are output. The action recognition results usually include arm swing actions and gait actions. For example: the arm swing actions are divided into excessive arm swing, crossed arm swing, outward arm swing, and standard arm swing: when the forward swing amplitude of the arm is too large or the reverse swing arc increases significantly, it is an excessive arm swing; when the arms cross frequently on the body center line, it is a crossed arm swing; when the outward deviation amplitude of the arm is large, it is an outward arm swing; in other cases, it is a standard arm swing. The gait actions include full-foot landing, heel landing, and toe landing: when the outer sides of the soles touch the ground simultaneously, it is a full-foot landing; when the heel lands first, it is a heel landing; when the front sole touches the ground first and the center of gravity gradually transitions to the heel, it is a toe landing. The recognition of the action results is completed by dividing the action categories.

[0125] According to the action recognition results output by the action recognition model, the arm swing actions and gait actions can be fused to obtain twelve action categories, as shown in the action category table in Table 1:

[0126] Table 1 Action Category Table

[0127]

[0128] Furthermore, a multi-modal association model with time synchronization fusion is constructed according to the motion trajectory and the action recognition results, including:

[0129] Adding timestamps to the motion trajectory and the action recognition results respectively;

[0130] Calculating the sampling periods of the positioning data and the action recognition results. The function expressions of the sampling periods of the positioning data and the action recognition results are respectively:

[0131]

[0132]

[0133] In the formula, represents the sampling period of the positioning data, represents the sampling frequency of the positioning data, represents the sampling period of the action recognition result, represents the sampling frequency of the action recognition result;

[0134] According to the consistency of the sampling period of the positioning data and the sampling period of the action recognition result, perform time alignment on the timestamps of the motion trajectory and the action recognition result, and generate a time synchronization sequence containing multiple time points within the same time window. Each time point corresponds to time-synchronized positioning data and action recognition results. The functional expression of the time synchronization sequence is:

[0135]

[0136] In the formula, represents the time synchronization sequence, represents the nth time point;

[0137] Construct a time synchronization fusion multi-modal association model based on the positioning data and action recognition results corresponding to all time points. The functional expression of the multi-modal association model is:

[0138]

[0139] In the formula, represents the multi-modal information vector generated by fusing the positioning data and action recognition results corresponding to the nth time point, represents the action recognition result corresponding to the nth time point, represents the positioning data corresponding to the nth time point.

[0140] Furthermore, performing time alignment on the timestamps of the motion trajectory and the action recognition result includes:

[0141] When the sampling period of the positioning data is the same as the sampling period of the action recognition result, unify the timestamps of the motion trajectory and the action recognition result to the same reference time and convert them to absolute timestamps for alignment;

[0142] When the sampling period of the positioning data is different from the sampling period of the action recognition result, find the two timestamps closest to a certain timestamp of the action recognition result in the timestamps of the motion trajectory. The functional expression for finding the timestamps is:

[0143]

[0144] In the formula, represents the i-th timestamp added to the action recognition result, represents the j-th timestamp added to the motion trajectory, represents the (j + 1)-th timestamp added to the motion trajectory;

[0145] According to the two timestamps found, linear interpolation is used to calculate the positioning data corresponding to a certain timestamp of the action recognition result. The calculation function expression of the positioning data is:

[0146]

[0147] In the formula, represents the positioning data corresponding to the i-th timestamp, represents the positioning data corresponding to the j-th timestamp; represents the positioning data corresponding to the (j + 1)-th timestamp;

[0148] According to the action recognition result and the positioning data corresponding to the same timestamp, the timestamps of the motion trajectory and the action recognition result are aligned.

[0149] Through the implementation of the above embodiments of the motion posture recognition method, unified timestamps are added to each group of motion trajectories and action recognition results, and data with different sampling rates is processed using the linear interpolation method to complete the synchronous fusion of the motion trajectory and the action recognition result within the same time window. And by constructing a multi-modal association model based on time series, the motion trajectory and the action recognition result are feature-fused to capture the dynamic relationship between the motion trajectory and the action recognition result, so as to generate the motion posture and trajectory result by combining timestamp matching and the multi-modal association model.

[0150] The present invention also discloses an identification system, which adopts the above-mentioned motion posture recognition method based on UWB positioning and tracking. The identification system includes:

[0151] A motion posture tracking module 4, which is used to obtain the positioning data of the target user based on UWB positioning and collect the action data of multiple parts of the target user;

[0152] A data processing module, which is used to construct a motion trajectory according to the obtained positioning data and extract time-domain features from the collected action data;

[0153] An action recognition module, which is used to establish an action recognition model according to the extracted time-domain features, input the collected action data into the action recognition model, and output an action recognition result including arm-swinging actions and gait actions;

[0154] A joint recognition module, which is used to construct a multi-modal association model with time synchronization fusion based on the motion trajectory and the action recognition result, capture the dynamic relationship between the motion trajectory and the action recognition result from the multi-modal association model, and generate the motion posture and trajectory result;

[0155] A visualization module, which is used to transmit the generated motion posture and trajectory result to the host computer software 3, and the host computer software 3 obtains the dynamically displayed posture and trajectory information through parsing.

[0156] Through the implementation of the above recognition system embodiment, the motion posture tracking module 4, the data processing module, the action recognition module, and the joint recognition module are all executed by the computing server 2. The host computer software 3 receives the motion posture and trajectory result from the computing server 2 through the Ethernet to dynamically display the real-time motion posture and trajectory information in a visualized manner (refer to Figure 2 ).

[0157] Preferably, the computer upper computer software 3 is used to display the attitude and trajectory information. Among them, the operation execution process of the upper computer software 3 is preferably as follows: first, open the upper computer software 3 and perform network settings, including setting network protocols, the local IP (Internet Protocol), and port numbers, and open the UDP Server (User Datagram Protocol Serve, a server program running on a specific host that uses network communication protocols to receive and respond to client requests). The upper computer software simultaneously receives the network access request frames from the motion attitude tracking module 4 and the UWB base station, completes device registration, and respectively sends out network access response frames. After the device registration is completed, it starts to continuously receive the action data and positioning data collected by the motion attitude tracking module 4 at the upper arm and ankle of the target user in real time, and saves these data into the database to achieve data persistence processing, providing support for the next feature extraction and fusion processing. Further, the upper computer software 3 supports two modes: the analysis mode and the recognition mode. The analysis mode is mainly used for training classifiers and constructing classification models, and performing data analysis and model optimization based on the sample data selected from the collected motion data; the recognition mode is used for real-time motion action recognition and trajectory tracking, that is, the system reads the collected action data from the database and inputs it into the trained action recognition model, and finally outputs the classification result and generates the motion trajectory at the same time. In addition, the upper computer software 3 can display the attitude and trajectory information in real time. In the visualization interface, the end user can dynamically observe the recognition results of the arm swinging action and gait action, as well as the real-time drawing of the motion trajectory. Through the UWB positioning data, the system can accurately draw the three-dimensional trajectory of the target user in space, realizing the comprehensive analysis of the motion attitude and spatial motion. Enabling the upper computer software 3 to receive and parse data in real time, dynamically display the attitude and trajectory information, including the line chart of key parameters and the motion path trajectory, providing precise support for motion training and attitude optimization.

[0158] Preferably, it can also be visually presented by transmitting the data to the corresponding mobile phone APP.

[0159] Furthermore, in combination with Figure 2 and Figure 8 As shown, the motion attitude tracking module 4 is connected with a wearing device. The wearing device includes a wearing shell, a display unit 41 arranged on the wearing shell, and a main control chip 42, a power supply unit 43, a UWB positioning unit 44, and an attitude sensing unit 45 integrally arranged in the wearing shell. The display unit 41, the power supply unit 43, the UWB positioning unit 44, and the attitude sensing unit 45 are respectively electrically connected to the main control chip 42, so that the UWB positioning unit 44 collects and sends the UWB positioning signal of the target user, and the attitude sensing unit 45 collects and sends the action data of the wearing part of the target user.

[0160] Through the implementation of the above embodiments of the recognition system, the main control chip 42, as the core of the system, is responsible for coordinating the operation between various modules and connecting the UWB positioning unit 44 and the attitude sensing unit 45 through the USART (Universal Synchronous and Asynchronous Receiver and Transmitter) interface. The attitude sensing unit 45 collects real-time action data of the target user's movement, including various parameters such as acceleration and angular velocity. The UWB positioning unit 44 is responsible for collecting and sending the positioning signal of the target user. The display unit 41 is used to display the collected motion data and processing results in real time. The power supply unit 43 is used to provide stable electrical energy to ensure the efficient operation of the motion attitude tracking module during data collection and wireless transmission. The motion attitude tracking module 4 performs wireless data transmission with external devices through the UWB positioning unit 44, combines multi-mode sensor technology and UWB positioning technology to achieve high-precision motion attitude recognition and real-time positioning functions.

[0161] The present invention also discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-described motion attitude recognition method based on UWB positioning and tracking are implemented.

[0162] The present invention also discloses a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the steps of the above-described motion attitude recognition method based on UWB positioning and tracking are implemented.

[0163] The present invention is described in terms of flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to specific embodiments. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented 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 devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in Figure 1 one or more of the flows and / or Figure 1 blocks or multiple blocks.

[0164] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device that implements the functions in Figure 1 one or more of the flows and / orFigure 1 The functions specified in one or more boxes.

[0165] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide for implementing the steps of the functions specified in one Figure 1 process or multiple processes and / or boxes Figure 1 or more boxes.

[0166] It should be understood that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. For those skilled in the art, the technical solutions recorded in the above embodiments can be modified, or some of the technical features can be equivalently replaced; and all such modifications and replacements should fall within the protection scope of the present invention.

Claims

1. A method for recognizing motion postures based on UWB positioning and tracking, characterized in that, The motion gesture recognition method based on UWB positioning and tracking includes: Obtaining positioning data of a target user based on UWB positioning, including: Based on a scenario covered by multiple UWB base stations, according to the UWB positioning signals sent by the device worn by the target user, calculating the time difference between the target user reaching two UWB base stations. The functional expression for calculating the time difference is: wherein, represents the time difference, represents the time when the signal arrives at base station m, represents the time when the signal arrives at base station k; Obtaining the speed of the target user reaching the UWB base station, and calculating the distance between the target user and the UWB base station according to the time difference. The functional expression for calculating the distance is: In the formula, represents the distance difference, represents the speeds of the target user reaching two UWB base stations; Obtaining the base station position coordinates of all UWB base stations, constructing a system of equations based on all the base station position coordinates and the distances between the target user and the corresponding UWB base stations, and solving the system of equations using the least squares method to obtain the positioning data. The positioning data includes the user position coordinates of the target user at different times; Collecting motion data of multiple parts of the target user; Constructing a motion trajectory based on the obtained positioning data, and extracting time-domain features from the collected motion data; Establishing an action recognition model according to the extracted time-domain features, including: According to the collected motion data, performing filtering processing on the motion data, and performing signal drift correction on the motion data after filtering processing; Based on signal frame processing, selecting a window length and an overlap rate to segment and intercept the motion data after signal drift correction, obtaining multiple windows containing action cycle data; Selecting features associated with arm swing actions and gait actions from the motion data, and statistically analyzing the selected features in the action cycle data in the window; Calculating the time-domain features of the selected features according to the statistical selected features, and extracting the time-domain features from each window; Screening out sample data from all the extracted time-domain features, and dividing the sample data into a training set and a test set; Selecting a network architecture to construct a basic classification model, batch inputting the time-domain features in the training set into the basic classification model for training to obtain the action recognition model; Batch inputting the time-domain features in the test set into the trained action recognition model for testing, and updating the model parameters of the action recognition model through an iterative optimization algorithm; Inputting the collected motion data into the action recognition model, and outputting an action recognition result including arm swing actions and gait actions; Constructing a time-synchronized fusion multi-modal association model according to the motion trajectory and the action recognition result, including: Adding timestamps to the motion trajectory and the action recognition result respectively; Calculating the sampling period of the positioning data and the sampling period of the action recognition result. The functional expressions for the sampling periods of the positioning data and the action recognition result are: In the formula, represents the sampling period of the positioning data, represents the sampling frequency of the positioning data, represents the sampling period of the action recognition result, represents the sampling frequency of the action recognition result; According to the consistency of the sampling periods of the positioning data and the action recognition result, performing time alignment on the timestamps of the motion trajectory and the action recognition result, and generating a time-synchronized sequence containing multiple time points within the same time window. Each time point corresponds to the time-synchronized positioning data and action recognition result. The functional expression of the time-synchronized sequence is: In the formula, represents the time synchronization sequence, represents the nth time point; Construct a time-synchronized fusion multi-modal association model based on the positioning data and the action recognition results corresponding to all time points. The functional expression of the multi-modal association model is as follows: Wherein, represents the multi-modal information vector generated by fusing the positioning data and action recognition result corresponding to the nth time point, represents the action recognition result corresponding to the nth time point, represents the positioning data corresponding to the nth time point; Capture the dynamic relationship between the motion trajectory and the action recognition result from the multi-modal association model, and generate a motion posture and trajectory result; Transmit the generated motion posture and trajectory result to the host computer software for parsing and generating dynamically displayed posture and trajectory information.

2. The method for recognizing motion postures based on UWB positioning and tracking according to claim 1, wherein, The construction of the motion trajectory based on the obtained positioning data includes: Filter all the user position coordinates according to the user position coordinates at different times; Perform time series sorting on the filtered user position coordinates, and use a trajectory fitting algorithm to perform smooth interpolation processing between the user position coordinates corresponding to adjacent times to generate a continuous trajectory curve to construct the motion trajectory.

3. The motion posture recognition method based on UWB positioning and tracking according to claim 1, wherein The time alignment of the timestamps of the motion trajectory and the action recognition result includes: When the sampling periods of the positioning data and the action recognition result are the same, unify the timestamps of the motion trajectory and the action recognition result to the same reference time and convert them into absolute timestamps for alignment; When the sampling periods of the positioning data and the action recognition result are different, find the two timestamps closest to a certain timestamp of the action recognition result in the timestamps of the motion trajectory. The functional expression of finding the timestamps is as follows: In the formula, represents the i-th timestamp added to the action recognition result, represents the j-th timestamp added to the motion trajectory, represents the (j + 1)-th timestamp added to the motion trajectory; According to the two found timestamps, use linear interpolation to calculate the positioning data corresponding to a certain timestamp of the action recognition result. The calculation functional expression of the positioning data is as follows: In the formula, represents the positioning data corresponding to the i-th timestamp, represents the positioning data corresponding to the j-th timestamp; represents the positioning data corresponding to the (j + 1)-th timestamp; Align the timestamps of the motion trajectory and the action recognition result according to the action recognition result and the positioning data corresponding to the same timestamp.

4. An identification system, adopting the motion posture identification method based on UWB positioning and tracking described in any one of claims 1-3, wherein, The recognition system includes: A motion posture tracking module for obtaining positioning data of a target user based on UWB positioning and collecting action data of multiple parts of the target user; A data processing module for constructing a motion trajectory according to the obtained positioning data and extracting time-domain features from the collected action data; An action recognition module for establishing an action recognition model according to the extracted time-domain features, inputting the collected action data into the action recognition model, and outputting an action recognition result including arm swing actions and gait actions; A joint recognition module for constructing a time-synchronized fusion multi-modal association model according to the motion trajectory and the action recognition result, capturing the dynamic relationship between the motion trajectory and the action recognition result from the multi-modal association model, and generating a motion posture and trajectory result; A visualization module for transmitting the generated motion posture and trajectory result to the host computer software.

5. The recognition system according to claim 4, wherein: The motion attitude tracking module is connected to a wearing device, which includes a wearing shell, a display unit arranged on the wearing shell, and a main control chip, a power supply unit, a UWB positioning unit and an attitude sensing unit integrally arranged in the wearing shell. The display unit, the power supply unit, the UWB positioning unit and the attitude sensing unit are respectively electrically connected to the main control chip, so that the UWB positioning unit collects and sends the UWB positioning signal of the target user, and the attitude sensing unit collects and sends the motion data of the wearing part of the target user.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the motion attitude recognition method based on UWB positioning and tracking according to any one of claims 1-3.

7. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the motion attitude recognition method based on UWB positioning and tracking according to any one of claims 1-3.

Images