Adaptive gait division method and device, storage medium and terminal

Abstract

The present invention discloses an adaptive gait segmentation method and device, storage medium, and terminal, wherein the method comprises: acquiring sampling point data of sensor data to be segmented in real time, and acquiring the latest sliding window sample data segment according to a preset sliding window acquisition method, calculating the variance of the latest sliding window sample data segment, and acquiring the latest data segment to be identified; acquiring the trend intensity feature of the latest data segment to be identified based on a preset trend intensity acquisition method, and using it as the target trend intensity feature; calculating the target feature intensity feature and establishing a correlation with a trend intensity feature set in a trend intensity feature library, fuzzy classifying the data segment to be identified based on the correlation information and the gait category of the feature set, and finally obtaining a re-issued label for the data segment to be identified based on the fuzzy classification. The present invention improves the accuracy of step recognition for different pedestrians at different walking speeds, and helps pedestrian indoor positioning systems meet the requirements for positioning system robustness in practical applications.

Description

Technical Field

[0001] The present invention relates to the field of indoor pedestrian inertial navigation positioning technology, and in particular to an adaptive gait division method and device, a storage medium and a terminal. Background Art

[0002] With the development of micro-electro-mechanical systems (MEMS) technology, inertial measurement units (IMUs) have achieved breakthroughs in miniaturization, accuracy, and reliability, offering advantages such as portability and cost-effectiveness. IMUs consist of a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer, and offer a sufficient sampling rate (100Hz to 300Hz) to collect real-time motion data from the wearer. As a result, IMUs are widely used in indoor pedestrian positioning applications and research.

[0003] To identify step boundaries from inertial sensor data, current approaches primarily rely on identifying the phase in the foot motion signal where the foot remains relatively stationary relative to the ground. This phase is referred to as the "zero-speed phase." Current positioning methods for detecting the "zero-speed phase" of pedestrian foot motion calculate variance to reflect the signal's fluctuations, then compare this variance with a fixed threshold to obtain a detection result. Because individual pedestrians vary in gender and height, which manifests itself in the fluctuations of sensor data, this approach often requires fine-tuning the threshold for each individual pedestrian, making step recognition highly sensitive to individual differences. For example, some dynamic phases of slow walking steps may be misidentified as static phases of normal steps, resulting in false positives. Alternatively, static phases of fast walking steps may be misidentified as dynamic phases of normal steps, leading to false negatives.

[0004] In summary, the existing processing methods of inertial sensor data are greatly affected by noise, step recognition is very sensitive to individual differences, and the accuracy and reliability of pedestrian indoor positioning are poor. Summary of the Invention

[0005] The technical problem to be solved by the present invention is that the existing method of processing inertial sensor data has a large noise impact, step recognition is very sensitive to individual differences, and the accuracy and reliability of pedestrian indoor positioning are poor.

[0006] In order to solve the above technical problems, the present invention provides an adaptive gait division method, comprising:

[0007] Acquire sampling point data of the sensor data to be divided in real time, obtain the latest sliding window sample data segment according to a preset sliding window acquisition method, calculate the variance of the latest sliding window sample data segment, and obtain the latest data segment to be identified according to a preset data segment acquisition method;

[0008] Acquire the trend strength feature of the latest data segment to be identified based on a preset trend strength acquisition method, and use the trend strength feature of the latest data segment to be identified as the target trend strength feature;

[0009] Calculating the membership between the target trend strength feature and each trend strength feature set group in the trend strength feature library, and selecting a preset number of memberships from all the memberships from largest to smallest as voting memberships, and taking the trend strength feature set group corresponding to each voting membership as a voting set group;

[0010] Based on the gait labels in each trend strength set, classify all trend strength feature sets in each voting set group to obtain a stationary phase subset and a non-stationary phase subset of each voting set group, and calculate the cluster center of each stationary phase subset and the cluster center of each non-stationary phase subset;

[0011] Calculating the Euclidean distance between the target trend intensity feature and each of the stationary phase subset cluster centers, and the Euclidean distance between the target trend intensity feature and each of the non-stationary phase subset cluster centers, and obtaining a stationary phase gait probability estimate value and a non-stationary phase gait probability estimate value corresponding to each of the voting set groups based on the Euclidean distance corresponding to each of the voting set groups;

[0012] Using the corresponding voting membership as weight, weighting and summing all the static phase gait probability estimation values ​​respectively to obtain the static phase decision value of the latest data segment to be identified, and weighting and summing the non-static phase gait probability estimation values ​​respectively to obtain the non-static phase decision value of the latest data segment to be identified;

[0013] Determine whether the static phase decision value is greater than the non-static phase decision value; if so, determine that the gait of the latest data segment to be identified is the static phase; otherwise, determine that the gait of the latest data segment to be identified is the non-static phase.

[0014] Preferably, the process of acquiring the trend strength feature library is as follows:

[0015] Acquire a reference gait dataset, and group all reference gait data segments in the reference gait dataset based on an influencing parameter to obtain a plurality of reference gait data groups, wherein the reference gait dataset includes a plurality of reference gait data segments and a gait label corresponding to each reference gait data segment;

[0016] Acquire all reference sliding window sample data segments of each reference gait data segment by a preset sliding window acquisition method, calculate the variance of each reference sliding window sample data segment, and acquire a reference data segment of each reference gait data segment by a preset data segment acquisition method;

[0017] Acquire the trend strength feature of each reference data segment based on a preset trend strength acquisition method, and group the trend strength feature corresponding to each reference gait data segment and its corresponding gait label into a trend strength feature set, and group all trend strength feature sets corresponding to each group of reference gait data into a trend strength feature set group, and all trend strength feature set groups form a trend strength feature library;

[0018] Wherein, the influencing parameters include gender, height and walking speed.

[0019] Preferably, the preset sliding window is obtained in the following manner:

[0020] Calculating the acceleration modulus value of the target sampling point data as the target acceleration modulus value, and adding the target acceleration modulus value to the end of the sliding window queue, and sequentially selecting w acceleration modulus values ​​from the sliding window queue starting from the target acceleration modulus value as the sliding window sample data segment;

[0021] Wherein, when the targeted data is sensor data to be divided, the target sampling point data is the latest sampling point data of the sensor data to be divided, and the obtained sliding window sample data segment is the latest sliding window sample data segment;

[0022] When the targeted data is a reference gait data segment, the sampling point data in the reference gait data segment is sequentially used as the target sampling point data, and the obtained sliding window sample data segment is the reference sliding window sample data segment;

[0023] When the targeted data is a target data segment, the sampling point data in the target data segment is sequentially used as the target sampling point data, and the obtained sliding window sample data segment is a sliding window data segment.

[0024] Preferably, the preset data segment is obtained in the following manner:

[0025] The target variance is placed at the end of the target variance queue. If the target variance is less than a preset variance threshold, and the variance of the previous bit of the target variance in the current target variance queue is greater than the preset variance threshold, the sampling point data corresponding to the acceleration modulus value at the end point of the sliding window sample data segment corresponding to the target variance is determined to be the starting point of the target data segment.

[0026] If the target variance is greater than a preset variance threshold, and the variance of the previous target variance in the current target variance queue is less than the preset variance threshold, then the sampling point data corresponding to the acceleration modulus value of the end point of the sliding window sample data segment corresponding to the target variance is determined to be the end point of the target data segment;

[0027] Whenever an end point of the target data segment is determined, the sampling point data from the most recently determined start point of the target data segment to the most recently determined end point of the target data segment in the currently targeted data is used as the target data segment;

[0028] Wherein, when the targeted data is sensor data to be divided, the target variance is the variance of the latest sliding window sample data segment, the target variance queue is the latest variance queue, and the target data segment is the latest data segment to be identified;

[0029] When the targeted data is a reference gait data segment, the target variance is the variance of the reference sliding window sample data segment, the target variance queue is a reference variance queue, and the target data segment is a reference data segment.

[0030] Preferably, obtaining the trend strength feature of the target data segment based on a preset trend strength obtaining method includes:

[0031] Using multiple time scales as sliding window widths, the target data segments are respectively subjected to sliding sampling by a preset sliding window acquisition method to obtain sliding window data segments of each time scale;

[0032] Based on all the sliding window data segments of each time scale, the moving average sequence of each time scale is obtained in time sequence; so that the moving averages at the same position in the moving average sequence of all time scales correspond to each other, and the moving average sequences corresponding to all time scales have the same data length

[0033] All trend data segments are intercepted from the target data segment based on the signal trend condition, and the one with the longest data length is selected from all the trend data segments as the most obvious trend data segment, wherein all the sampling point data in the trend data segment must satisfy: all the moving average values ​​corresponding to a single sampling point data must satisfy the signal trend condition, and all the moving average values ​​corresponding to the sampling point data include: when the sampling point data is used as the target sampling point data, the moving average values ​​are obtained based on the sliding window data segments of each time scale;

[0034] Calculating a trend strength feature of the most obvious trend data segment, the trend strength feature including a ratio of the time length of the most obvious trend data segment to the target data segment, a signal energy of the most obvious trend data segment, an average energy value of the most obvious trend data segment, and a slope of the most obvious trend data segment;

[0035] The signal trend condition is as follows: all the moving average sequences increase or decrease in sequence according to the order of increasing or decreasing time scales; and the target data segment is the latest data segment to be identified or the reference data segment.

[0036] Preferably, the moving average sequence of a single time scale is obtained based on all sliding window data segments of the time scale:

[0037] ;

[0038] in, represents the kth moving average in a moving average sequence with time scale w, Indicates the acceleration modulus value of the kth sampling point data in the target data segment.

[0039] Preferably, calculating the membership between the target trend strength feature and a single trend strength feature set group in the trend strength feature library comprises:

[0040]

[0041] in, Indicates the target trend strength characteristic, Represents the cluster center of the current trend strength feature set, represents the Euclidean distance between the target trend intensity feature and the cluster center of the current trend intensity feature set, Indicates the trend strength characteristics, Grouped separately with all trend strength feature sets The sum of the membership degrees between represents the cluster center of the wth group of trend strength feature sets in the trend strength feature library, O is the number of trend strength feature sets in the trend strength feature library, represents the Lagrange multiplier.

[0042] In order to solve the above technical problems, the present invention also provides an adaptive gait segmentation device, which includes a latest data segment acquisition module to be identified, a target trend strength feature acquisition module, a voting set group acquisition module, a cluster center acquisition module, a gait probability estimation value acquisition module, a decision value acquisition module and a stage judgment module;

[0043] The latest data segment to be identified acquisition module is used to acquire the sampling point data of the sensor data to be divided in real time, and obtain the latest sliding window sample data segment according to a preset sliding window acquisition method, calculate the variance of the latest sliding window sample data segment, and obtain the latest data segment to be identified according to the preset data segment acquisition method;

[0044] The target trend strength feature acquisition module is used to acquire the trend strength feature of the latest data segment to be identified based on a preset trend strength acquisition method, and use the trend strength feature of the latest data segment to be identified as the target trend strength feature;

[0045] The voting set group acquisition module is used to calculate the membership between the target trend strength feature and each trend strength feature set group in the trend strength feature library, and select a preset number of memberships from all the memberships from large to small as voting memberships, and use the trend strength feature set group corresponding to each voting membership as the voting set group;

[0046] The cluster center acquisition module is used to classify all trend strength feature sets in each voting set group based on the gait labels in each trend strength set, so as to obtain a stationary phase subset and a non-stationary phase subset of each voting set group, and respectively calculate the cluster center of each stationary phase subset and the cluster center of each non-stationary phase subset;

[0047] The gait probability estimation value acquisition module is used to calculate the Euclidean distance between the target trend intensity feature and each of the stationary phase subset cluster centers, and the Euclidean distance between the target trend intensity feature and each of the non-stationary phase subset cluster centers, and obtain the stationary phase gait probability estimation value and the non-stationary phase gait probability estimation value corresponding to each of the voting set groups based on the Euclidean distance corresponding to the voting set group;

[0048] The decision value acquisition module is configured to weight and sum all the estimated gait probabilities of the stationary phase using the corresponding voting membership as weights to obtain the stationary phase decision value of the latest data segment to be identified, and to weight and sum all the estimated gait probabilities of the non-stationary phase to obtain the non-stationary phase decision value of the latest data segment to be identified;

[0049] The stage judgment module is used to judge whether the static stage decision value is greater than the non-static stage decision value. If so, the gait of the latest data segment to be identified is judged to be the static stage; otherwise, the gait of the latest data segment to be identified is judged to be the non-static stage.

[0050] In order to solve the above technical problems, the present invention further provides a storage medium on which a computer program is stored, and when the program is executed by a processor, an adaptive gait division method is implemented.

[0051] In order to solve the above technical problem, the present invention further provides a terminal, comprising: a processor and a memory, wherein the memory is communicatively connected to the processor;

[0052] The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the terminal performs an adaptive gait division method.

[0053] Compared with the prior art, one or more embodiments of the above solutions may have the following advantages or beneficial effects:

[0054] The adaptive gait segmentation method provided by the embodiment of the present invention is applied. Based on a large number of reference gait data sets containing a wide range of influencing parameter data, a trend intensity feature library with gait labels is obtained. Then, based on the trend intensity feature, the sensor data to be segmented is classified, that is, the degree of correlation between the sensor data to be segmented and the different feature groups in the trend intensity feature library is obtained. Finally, the correlation degree is used as a weight parameter to make a centralized decision on the results of the fuzzy classification, so that the step segmentation algorithm can meet the requirements of cross-individual and wide-speed domains in actual application scenarios. The present invention improves the accuracy of step recognition for different pedestrian individuals at different walking speeds, and helps pedestrian indoor positioning systems meet the requirements for positioning system robustness in actual applications.

[0055] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained through the structures particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0057] Figure 1 FIG2 shows a flow chart of an adaptive gait division method according to an embodiment of the present invention;

[0058] Figure 2 Schematic diagram showing acceleration data of a stationary stage in the first embodiment of the present invention;

[0059] Figure 3 Schematic diagram showing the moving average of acceleration data in a stationary phase in the first embodiment of the present invention;

[0060] Figure 4 FIG1 shows a schematic diagram of acceleration data in a non-stationary stage in the first embodiment of the present invention;

[0061] Figure 5 A schematic diagram showing a moving average of acceleration data in a non-stationary phase in the first embodiment of the present invention is shown;

[0062] Figure 6 It shows a schematic structural diagram of an adaptive gait division device according to a second embodiment of the present invention;

[0063] Figure 7 It shows a schematic structural diagram of a terminal according to a fourth embodiment of the present invention. DETAILED DESCRIPTION

[0064] The following describes the embodiments of the present invention in detail with reference to the accompanying drawings and examples, so that the present invention can fully understand how to apply technical means to solve technical problems and achieve technical effects, and thus implement the invention accordingly. It should be noted that, as long as no conflict exists, the various embodiments of the present invention and the various features of the embodiments can be combined with each other, and the resulting technical solutions are all within the scope of protection of the present invention.

[0065] Small inertial measurement units are susceptible to signal noise, often resulting in cumulative errors in distance and direction calculations, making them unsuitable for long-term use. Due to individual differences (gender, height, age, etc.), pedestrians' walking patterns indoors are diverse and uncertain. Inertial measurement unit-based systems exhibit poor stability across different pedestrians and at varying walking speeds, often requiring customized parameter design for each individual user. This limits the system's stability and versatility. Therefore, reducing the impact of noise in sensor-collected data and improving the stability of positioning systems across individual pedestrians have become key technical areas for improving the accuracy and reliability of indoor pedestrian positioning.

[0066] Inertial measurement units are usually worn on the human head, waist, legs, backpacks, pockets, and feet. Compared with other wearing methods, the feet can obtain the most direct and relevant motion information about walking, and therefore are one of the main data types for pedestrian indoor positioning.

[0067] Pedestrian foot motion data includes both dynamic and static walking patterns. However, these two patterns also reflect individual differences and the influence of individual walking patterns. When processing foot motion data, while the signal exhibits significant periodicity, it is crucial to minimize the impact of noise in the data and maintain cross-individual adaptability.

[0068] In actual application scenarios, the foot movements of pedestrians vary from individual to individual and at different walking speeds, making it impossible for the above methods to guarantee the accuracy of step recognition. For example, due to individual differences, the above methods often have problems such as identifying false steps and missing steps. The reason is that when pedestrians walk, the amplitude of the foot movement signal collected by the sensor varies greatly in scenarios with different individuals and different walking speeds. Due to the limitation of the threshold, the above methods often identify the dynamic phase of some slow-walking steps as the static phase of normal steps, which leads to the erroneous result of false steps; or identify the static phase of fast-walking steps as the dynamic phase of normal steps, which leads to the erroneous result of missing steps.

[0069] Example 1

[0070] In order to solve the technical problems existing in the prior art, an embodiment of the present invention provides an adaptive gait division method.

[0071] Figure 1 FIG1 shows a flow chart of an adaptive gait division method according to an embodiment of the present invention; ... Figure 1 As shown, the adaptive gait division method according to an embodiment of the present invention includes the following steps.

[0072] Step S101, acquiring sampling point data of sensor data to be divided in real time, and acquiring the latest sliding window sample data segment according to a preset sliding window acquisition method, calculating the variance of the latest sliding window sample data segment, and acquiring the latest data segment to be identified according to a preset data segment acquisition method.

[0073] Specifically, this embodiment can acquire sampling point data in real time for the sensor data to be segmented. Based on the real-time acquired sampling point data, the sensor data to be segmented can be segmented into gaits, and gait labels for the segmented gaits can be determined in real time. Furthermore, embodiments of the present invention can also perform adaptive gait segmentation on the already acquired sensor data to be segmented. In this case, the sampling point data from the fully acquired sensor data to be segmented can be sequentially provided in chronological order as the real-time acquired sampling point data for the sensor data to be segmented. Furthermore, the sensor data to be segmented includes acceleration data, meaning that the real-time acquired sampling point data is acceleration data at the sampling points, and the acceleration data is represented by data along three perpendicular axes. These three axes are the longitudinal axis running vertically through the human body (i.e., the vertical axis, perpendicular to the horizontal plane, represented by the Z-axis); the horizontal axis running vertically through the human body (i.e., the horizontal axis, perpendicular to the coronal plane, represented by the X-axis); and the horizontal axis running horizontally through the human body (i.e., the coronal axis, perpendicular to the sagittal plane, represented by the Y-axis). Furthermore, the sensor data to be divided in the embodiment of the present invention is sensor data recording footstep movements, and the data format of the sensor data in the reference gait data set below is the same as the data format of the sensor data to be divided.

[0074] The sensor data to be divided can be expressed in the following form:

[0075] (1);

[0076] Where acc represents acceleration data, t represents time, x, y, and z represent the three axis directions of acceleration data, and j represents the jth sampling point in the sensor data to be divided. Indicates the data length of the sensor data to be divided.

[0077] After acquiring the sampling point data for the sensor data to be segmented, the latest sliding window sample data segment is acquired using a preset sliding window acquisition method based on the latest acquired sampling point data. The variance of the latest sliding window sample data segment is calculated, and then the latest data segment to be identified is acquired using the preset data segment acquisition method based on the variance of the latest sliding window sample data segment. Using the preset sliding window acquisition method to process the sensor data to be segmented can remove high-frequency noise from the sensor data to be segmented.

[0078] Specifically, the variance of the latest sliding window sample data segment is calculated as follows:

[0079] (2);

[0080] (3);

[0081] in, Represents the variance of the latest sliding window sample data segment, Indicates the acceleration modulus value of the j-th sampling point data in the latest sliding window sample data segment, Represents the average value of the acceleration modulus of all sampling point data in the latest sliding window sample data segment, N represents the number of sampling point data in the latest sliding window sample data segment, and j represents the jth sampling point in the latest sliding window sample data segment. Indicates the data length of the latest sliding window sample data segment.

[0082] Furthermore, the process of obtaining the latest sliding window sample data segment based on the latest acquired sampling point data according to a preset sliding window acquisition method specifically includes: using the latest sampling point data of the sensor data to be segmented as the target sampling point data, calculating the acceleration modulus of the target sampling point data as the target acceleration modulus value, and adding the target acceleration modulus value to the end of the sliding window queue. From the sliding window queue, sequentially selecting w acceleration modulus values ​​starting from the target acceleration modulus value as the sliding window sample data segment. The sliding window sample data segment is the latest sliding window sample data segment. The time scale represented by w is the sliding window width. The sliding window width must be between 1 / 4 and 1 / 2 of the shortest step time. In the embodiment of the present invention, the sliding step of the sliding window is set to one sampling point data. For example, the time it takes for a pedestrian to take a single step when walking is 250 milliseconds to 2000 milliseconds. When the sensor sampling frequency is 100 Hz (i.e., the interval between each sample is 10 milliseconds), the value range of w can be [7, 13].

[0083] The acceleration modulus of a certain sampling point data is calculated as follows:

[0084] (4);

[0085] in, Indicates the acceleration modulus value of the j-th sampling point data, 、 and Represents the acceleration data in three directions of the j-th sampling point data respectively.

[0086] It should be noted that since each acceleration modulus value corresponds to a sampling point data, and the latest sliding window sample data segment is obtained based on a certain acceleration modulus value, we take the latest sliding window sample data segment with the acceleration modulus value as the starting point as the latest sliding window sample data segment corresponding to the acceleration modulus value. Further, one sampling point data corresponds to one latest sliding window sample data segment, and other sliding window sample data segments obtained through the preset sliding window acquisition method are also regarded as the same setting.

[0087] Furthermore, the process of obtaining the latest data segment to be identified according to the preset data segment acquisition method based on the variance of the latest sliding window sample data segment specifically includes: taking the variance of the latest sliding window sample data segment as the target variance, after obtaining the target variance, placing the target variance at the end of the target variance queue, and then comparing the target variance with the preset variance threshold; if the target variance is less than the preset variance threshold, and the variance of the previous bit of the target variance in the current target variance queue is greater than the preset variance threshold, that is, the variance has negatively crossed the preset variance threshold, at this time, it is determined that the sampling point data corresponding to the endpoint acceleration modulus value of the latest sliding window sample data segment corresponding to the target variance is the target data segment. Starting point; If the target variance is greater than the preset variance threshold, and the variance of the previous target variance in the current target variance queue is less than the preset variance threshold, that is, the variance has positively crossed the preset variance threshold, then the sampling point data corresponding to the acceleration modulus value of the endpoint of the latest sliding window sample data segment corresponding to the target variance is determined as the target data segment endpoint; the target data segment starting point and target data segment endpoint in this step are both sampling point data in the sensor data to be divided, and each time a target data segment endpoint is determined, the sampling point data from the most recently determined target data segment starting point to the most recently determined target data segment endpoint in the sensor data to be divided is used as the target data segment. For ease of distinction, we identify the target variance queue in the preset data segment acquisition method as the most recent variance queue, and the acquired target data segment as the most recently identified data segment.

[0088] Step S102: obtaining the trend strength feature of the latest data segment to be identified based on a preset trend strength acquisition method, and using the trend strength feature of the latest data segment to be identified as the target trend strength feature.

[0089] Specifically, multiple time scales are set as sliding window widths. The maximum sliding window width range must not exceed 1 / 4 of the shortest step time. In this embodiment of the present invention, the sliding window step is set to one sampling point. Furthermore, to reflect the difference in signal trend strength between sliding windows of different scales, the various sliding window widths set in this step must have a 2-fold relationship from small to large. For example, the sliding window width limits the time it takes for a pedestrian to take a single step to between 250 milliseconds and 2000 milliseconds. The sensor sampling frequency used in this solution is 100 Hz, meaning that the interval between each sample is 10 milliseconds. Therefore, the maximum sliding window width used is no more than 13. Based on the constraints set on the various sliding window widths, the various time scales selected in this embodiment can be 3, 6, and 12, respectively.

[0090] Then, using each of the aforementioned time scales as the sliding window width, the target data segment is slidingly sampled using a preset sliding window acquisition method to obtain all sliding window data segments for each time scale. That is, each time scale has a corresponding sliding window data segment group. It should be noted that in this step, when using the preset sliding window acquisition method to obtain sliding window data segments for each time scale, the sampling point data in the target data segment is sequentially used as the target sampling point data, and the sliding window sample data segments obtained using the preset sliding window acquisition method are referred to as sliding window data segments. Furthermore, the process of obtaining sliding window sample data segments using the sampling point data in the target data segment as the target sampling point data and using the preset sliding window acquisition method is nearly identical to the aforementioned process of obtaining the latest sliding window sample data segments using the preset sliding window acquisition method based on the latest acquired sampling point data. The only difference is that the sampling point data in the target data segment is used as the target sampling point data, and the corresponding sliding window sample data segments are referred to as sliding window data segments.

[0091] After obtaining all sliding window data segments corresponding to each time scale, a moving average sequence of each time scale is obtained based on all sliding window data segments of each time scale. All moving averages in the moving average sequence can be arranged in chronological order.

[0092] Furthermore, obtaining the moving average sequence of a single time scale based on all sliding window data segments of the time scale can be achieved by the following formula:

[0093] (4)

[0094] in, represents the kth moving average in a moving average sequence with time scale w, Indicates the acceleration modulus value of the kth sampling point data in the target data segment, This means that the average value of all acceleration moduli in the kth sliding window data segment of this time scale is the same. The above formula shows that the moving average sequences for all time scales have the same data length, and the moving average values ​​at the same position in the moving average sequences for all time scales are corresponding.

[0095] After obtaining a moving average sequence for each time scale, the most obvious trend data segment can be obtained from the target data segment using the moving average sequence for each time scale. Specifically, the process for obtaining the most obvious trend data segment includes: extracting all trend data segments from the target data segment based on a signal trend condition, and selecting the segment with the longest data length as the most obvious trend data segment. Furthermore, the process for obtaining the trend data segment requires that all sampling point data in the trend data segment meet a fixed condition: the fixed condition is that all moving average values ​​corresponding to a single sampling point data must meet the signal trend condition. Since the trend data segment is a data segment extracted from the target data segment, the sampling point data in the trend data segment is guaranteed to be continuous within the target data segment. Furthermore, all moving average values ​​corresponding to the sampling point data include: when the sampling point data is used as the target sampling point data, the moving average values ​​are obtained based on the sliding window data segment for each time scale. The signal trend condition is that all moving average value sequences increase or decrease in order of increasing or decreasing time scales.

[0096] That is, if the time scales are 3, 6, and 12 hours respectively, the sampling point data in the trend data segment must meet the following requirements:

[0097] (5)

[0098] Or the sampling point data in the trend data segment must meet the following requirements:

[0099] (6)

[0100] When analyzing existing footstep sensor data, we found that pedestrian foot data is periodic when walking normally, and the acceleration data in the static phase and the non-static phase are significantly different in amplitude and distribution. Figure 2 Schematic diagram showing acceleration data of a stationary stage in the first embodiment of the present invention; Figure 3 FIG1 shows a schematic diagram of the moving average of acceleration data in a stationary phase in the first embodiment of the present invention; ... Figure 2 and Figure 3 As shown in the figure, the acceleration vertical axis data in the static stage presents random fluctuations around the 0 axis. At the same time, in order to reflect the randomness of the fluctuation, Figure 3 Mean filtering with sliding windows of 3, 6, and 12 is used. It can be seen that the three signal curves after filtering can hardly reflect the acceleration trend. Figure 4 FIG1 shows a schematic diagram of acceleration data in a non-stationary stage in the first embodiment of the present invention; Figure 5 FIG1 shows a schematic diagram of the moving average of acceleration data in a non-stationary stage in the first embodiment of the present invention; ... Figure 4 and Figure 5As shown, the acceleration data during the non-stationary phase sequentially crosses between different mean signal curves, indicating a clear increase or decrease in the signal. Comparing the trend signal lines for the stationary and non-stationary phases, we can see a significant magnitude difference between the static and non-stationary phases.

[0101] Based on the analysis of the above content, we propose the following signal-based trend strength features; trend strength features include (1) the ratio of the time length of the signal segment to the entire step signal segment; (2) the signal energy of the signal segment; (3) the average energy value of the signal segment; and (4) the slope of the signal segment. The above trend strength features cover the intensity of the signal trend reflected by acceleration, the average energy value, the shape of the signal line (slope), and the time ratio relative to the entire step signal in the non-stationary phase of the foot sensor data, basically describing and recording the spatiotemporal characteristics of the signal trend. The ratio of time length and energy relative to the entire step can avoid the influence of individual differences on signal amplitude. Therefore, this set of features has better adaptability to individual differences than traditional statistical features (peak value, variance, etc.).

[0102] After obtaining the most obvious trend data segment, the trend strength characteristics of the most obvious trend data segment can be calculated. The trend strength characteristics of the most obvious trend data segment include:

[0103] (1) The ratio of the time length of the most obvious trend data segment to the target data segment. The specific calculation formula is as follows:

[0104] (7);

[0105] in, Indicates the ratio of the time length of the most obvious trend data segment to the target data segment. Indicates the most obvious trend data segment, Indicates the target data segment, Indicates the number of sampling point data in the most obvious trend data segment, Indicates the number of sampling point data in the target data segment.

[0106] (2) The signal energy of the most obvious trend data segment is calculated as follows:

[0107] (8);

[0108] in, The signal energy of the data segment with the most obvious trend, Indicates the data of the pth sampling point in the most obvious trend data segment, Indicates the number of sampling point data in the most obvious trend data segment.

[0109] (3) The average energy value of the most obvious trend data segment is calculated as follows:

[0110] (9)

[0111] in, Indicates the average energy value of the most obvious trend data segment, The signal energy of the data segment with the most obvious trend, Indicates the number of sampling point data in the most obvious trend data segment.

[0112] (4) The slope of the most obvious trend data segment is calculated as follows:

[0113] (10)

[0114] in, The slope of the data segment that represents the most obvious trend, Indicates the maximum value in the most obvious trend data segment. Indicates the minimum value in the data segment with the most obvious trend. Indicates the number of sampling point data in the most obvious trend data segment.

[0115] After obtaining the trend strength feature of the latest data segment to be identified, the trend strength feature of the latest data segment to be identified is used as the target trend strength feature.

[0116] Step S103, calculate the membership between the target trend strength feature and each trend strength feature set group in the trend strength feature library, and select a preset number of memberships from all the memberships from large to small as voting memberships, and use the trend strength feature set group corresponding to each voting membership as the voting set group.

[0117] Specifically, before performing adaptive gait segmentation on the sensor data to be segmented, it is necessary to first obtain a trend strength feature library. The specific process of obtaining the trend strength feature library is as follows:

[0118] A reference gait data set is obtained based on the data actually collected, and all reference gait data segments in the reference gait data set are grouped based on the influencing parameters to obtain multiple reference gait data groups. The reference gait data set includes multiple reference gait data segments and gait labels corresponding to each reference gait data segment. It should be noted that there are obvious differences in walking habits between individual pedestrians. When affecting the walking of different individuals, the influencing factors of the foot movement pattern mainly include the pedestrian's gender, height and walking speed. Therefore, the present invention sets gender, height and walking speed as influencing parameters. After grouping the data according to gender, height and walking speed, what is obtained is the pace and gait data of pedestrians within a certain height range of a certain gender walking at a certain speed level. It should be noted that the present invention requires that the range of the influencing parameters be as wide as possible to cover reference gait data segments of various heights and various walking speeds as much as possible.

[0119] For example, in an embodiment of the present invention, the physiological characteristics of gait (gender, height) and the walking speed categories (fast, slow, and medium) provided by the reference gait dataset can be grouped layer by layer. Specifically, all reference gait data segments in the reference gait dataset are first divided into two major categories based on gender. Then, the reference gait data segments of each gender are divided into multiple height segments based on the height information of the gait. For example, the six height segments can be divided into: {140cm, 160cm}, {160cm, 165cm}, {165cm, 170cm}, {170cm, 175cm}, {175cm, 180cm}, and {180cm, 200cm}. These six height segments cover the height range of 140cm to 200cm.

[0120] Generally speaking, compared to healthy women, men have significantly greater lower limb strength. Therefore, the amplitude of the acceleration signal from the foot movement detected by sensors when men walk is larger than that of women. For pedestrians of different heights, stride length is generally positively correlated with height, with the foot swinging phase lasting longer in shorter individuals. In daily life, pedestrians' walking speeds often vary significantly depending on the context. When walking while thinking, they adopt a slow walking speed, even slower than normal. If pedestrians are in a hurry to reach a destination, such as catching the last bus, they often use a very rapid walking speed. Generally, pedestrians use three walking speeds: fast, normal, and slow, depending on the context. Specifically, for each gender and height range, the dataset subsets of stride data are divided into {slow}, {medium}, and {fast} according to the speed type. Among them, slow walking is common when pedestrians are thinking seriously while walking, so the step cycle is long and the foot movement is relatively gentle; fast walking is common when pedestrians are trying to catch the upcoming last bus, the step cycle is very short, and the foot movement is intense; medium-speed walking is a walking state between the first two. It should be noted that the above speed distinction method is only for reference, and the three speeds do not have a specific numerical range. Specifically, when constructing the data set, they can be distinguished according to the three walking speeds used by different pedestrians in life. Due to different usage scenarios, there may be obvious differences. Based on the above grouping method, the above reference gait data set can be divided into 36 gait data groups according to 2 genders, 6 height groups and 3 walking speeds.

[0121] It should be noted that the process of grouping the reference gait dataset based on the influencing parameters is not limited to the above form, and can also be set to other reasonable situations.

[0122] After the reference gait data set is grouped, all reference sliding window sample data segments of each reference gait data segment need to be obtained through the preset sliding window acquisition method, and the variance of each reference sliding window sample data segment needs to be calculated. Finally, the reference data segment of each reference gait data segment needs to be obtained according to the preset data segment acquisition method.

[0123] The process of obtaining all reference sliding window sample data segments of a single reference gait data segment by using the preset sliding window acquisition method is almost the same as the process of obtaining the latest sliding window sample data segment by using the preset sliding window acquisition method in step S101. The only difference is that the data targeted is the reference gait data segment, and the sampling point data in the reference gait data segment needs to be used as the target sampling point data in sequence to obtain the sliding window sample data segment. The obtained sliding window sample data segment is also the reference sliding window sample data segment.

[0124] At the same time, the process of obtaining the reference data segment of a single reference gait data segment according to the preset data segment acquisition method is almost the same as the process of obtaining the latest data segment to be identified according to the preset data segment acquisition method in step S101. The only difference is that the targeted data is the reference gait data segment, and the target variance therein should be the variance of the reference sliding window sample data segment, the target variance queue should be the reference variance queue, and the target data segment obtained is also the reference data segment.

[0125] Finally, after obtaining the reference data segment of each reference gait data segment, the trend intensity feature of each reference data segment is obtained based on the preset trend intensity acquisition method. The process of obtaining the trend intensity feature of each reference data segment based on the preset trend intensity acquisition method is almost the same as the process of obtaining the trend intensity feature of the latest data segment to be identified based on the preset trend intensity acquisition method in step S102. The difference is that the reference data segment of each reference gait data segment is divided into target data segments.

[0126] After obtaining the trend strength features of each reference data segment, since the reference data segment is part of the reference gait data segment and each reference gait data segment has a corresponding gait label, we use the gait label corresponding to the reference gait data segment as the gait label for all reference data segments obtained from that reference gait data segment. It should be noted that in general, a single reference data segment is obtained from a single reference gait data segment, but in some cases, multiple reference data segments are obtained from a single reference gait data segment. In this case, we use all the obtained reference data segments as the reference data segments of the reference gait data segment.

[0127] We then aggregate the trend strength features and gait labels corresponding to each reference gait data segment into a trend strength feature set. It should be noted that if a reference gait data segment corresponds to multiple reference data segments, the trend strength features and gait labels corresponding to each reference data segment must be aggregated into a trend strength feature set.

[0128] After obtaining the trend intensity features corresponding to all reference gait data segments and their corresponding gait label sets as a trend intensity feature set, all trend intensity feature sets corresponding to each group of reference gait data groups are respectively grouped into a trend intensity feature set group, and all trend intensity feature set groups form a trend intensity feature library.

[0129] The trend strength feature set can be expressed as:

[0130] (11), where

[0131] (12);

[0132] Wherein, O represents the total number of groups obtained by grouping all reference gait data segments in the reference gait data set based on the influencing parameters, t represents the t-th reference gait data segment in the q-th reference gait data group, represents the trend strength feature of the t-th reference gait data segment in the q-th reference gait data group, They represent the trend intensity features corresponding to the t-th reference gait data segment in the q-th reference gait data group, represents the gait label of the t-th reference gait data segment in the q-th reference gait data group, and T represents the total number of reference gait data segments contained in the q-th reference gait data group. It should be noted that the total number of reference gait data segments in different reference gait data groups may be different.

[0133] After obtaining the trend strength feature library, the correlation calculation between the target trend strength feature and each trend strength feature set group in the trend strength feature library can be completed. That is, the membership degree between the target trend strength feature and each trend strength feature set group in the trend strength feature library is calculated.

[0134] Furthermore, the fuzzy clustering data association algorithm is to achieve association by dividing the data segments to be identified into each gait data group by minimizing the objective function.

[0135] First, the cost function of fuzzy clustering association is established as:

[0136] (13);

[0137] in, is the target trend strength characteristic, express and cluster centers The Euclidean distance between Indicates the membership of the target trend intensity feature and the qth trend intensity feature set, cluster center It will Temporarily classified as The cluster center is recalculated after the collection, and the cluster center algorithm is as follows:

[0138] ;

[0139] q=1,2,…,O; t=1,2…,T (14)

[0140] in, represents the cluster center of the qth reference gait data group, O represents the total number of groups obtained by grouping all reference gait data segments in the reference gait data set based on the influencing parameters, t represents the tth reference gait data segment in the qth reference gait data group, T represents the total number of segments containing reference gait data segments in the qth reference gait data group, Standard trend strength characteristics, They respectively represent the trend intensity features corresponding to the t-th reference gait data segment in the q-th reference gait data group.

[0141] and The following constraints apply:

[0142] (15);

[0143] Secondly, in order to describe the target trend strength characteristics in a minimal and unbiased manner And the membership of each trend strength feature set group, this scheme uses the maximum Shannon entropy principle, the expression is:

[0144] (16);

[0145] Furthermore, under the constraints of equations (14) and (15), we maximize equation (16) and apply the Lagrange multiplier method to define the objective function as:

[0146] (17);

[0147] Among them and is the Lagrange multiplier, maximizing formula (17), and the trend strength characteristic can be obtained and trend strength feature set The degree of membership between them is:

[0148] (18);

[0149] in, Indicates trend strength characteristics and trend strength feature set The degree of membership between them is:

[0150] Actually, it is to show the trend strength characteristics Grouped separately with all trend strength feature sets The sum of the membership degrees between them. In order to facilitate distinction, we use W instead of q in this formula to represent each trend strength feature set group. Similarly, the maximum value of W is 0.

[0151] The above formula can be used to calculate the membership between the target trend strength feature and each trend strength feature set group in the trend strength feature library.

[0152] After calculating the membership between the target trend strength feature and each trend strength feature set in the trend strength feature library, we select a preset number of memberships from all obtained memberships, from largest to smallest, as voting memberships, and use the trend strength feature set set corresponding to each voting membership as the voting set set. Preferably, the preset number can be 30% of the total number of groups obtained by grouping all reference gait data segments in the reference gait dataset based on the influence parameter.

[0153] Step S104: Based on the gait labels in each trend strength set, all trend strength feature sets in each voting set group are classified respectively to obtain the stationary phase subset and non-stationary phase subset of each voting set group, and the cluster center of each stationary phase subset and the cluster center of each non-stationary phase subset are calculated respectively.

[0154] Specifically, based on the gait labels in each trend strength set within the voting set, each voting set is classified into a stationary phase subset and a non-stationary phase subset. The cluster center of each stationary phase subset and the cluster center of each non-stationary phase subset are then calculated. The calculation method for the cluster centers here is similar to that of formula (12), except that the gait data set under consideration is changed to a stationary phase subset or a non-stationary phase subset. This will not be described in detail here.

[0155] Step S105, calculate the Euclidean distance between the target trend intensity feature and each stationary phase subset cluster center, and the Euclidean distance between the target trend intensity feature and each non-stationary phase subset cluster center, and obtain the stationary phase gait probability estimation value and the non-stationary phase gait probability estimation value of the corresponding voting set group based on the Euclidean distance corresponding to each voting set group.

[0156] Specifically, the Euclidean distance between the target trend strength feature and each stationary stage subset cluster center, and the Euclidean distance between the target trend strength feature and each non-stationary stage subset cluster center are calculated respectively. Thus, each voting set group corresponds to two Euclidean distances, where the Euclidean distance between the target trend strength feature and the stationary stage subset cluster center in the voting set group is expressed as The Euclidean distance between the target trend intensity feature and the cluster center of the non-stationary stage subset in the voting set is expressed as express.

[0157] Then, based on the Euclidean distance corresponding to each voting set group, a stationary phase gait probability estimation value and a non-stationary phase gait probability estimation value of the corresponding voting set group are obtained.

[0158] The estimated value of the gait probability in the stationary phase of the voting set is calculated as follows:

[0159] (19);

[0160] The estimated value of the non-stationary phase gait probability of the voting set is calculated as:

[0161] (20);

[0162] In step S106, all the estimated gait probabilities in the stationary phase are weighted and summed with the corresponding voting membership as weights to obtain the stationary phase decision value of the latest data segment to be identified, and the estimated gait probabilities in the non-stationary phase are weighted and summed to obtain the non-stationary phase decision value of the latest data segment to be identified.

[0163] Specifically, all stationary phase gait probability estimates are weighted using the voting membership corresponding to each voting set as a weight, and all weighted stationary phase gait probability estimates are summed to obtain the stationary phase decision value for the latest data segment to be identified. Similarly, all non-stationary phase gait probability estimates are weighted using the voting membership corresponding to each voting set as a weight, and all weighted non-stationary phase gait probability estimates are summed to obtain the non-stationary phase decision value for the latest data segment to be identified.

[0164] The calculation formula for the static phase decision value of the latest data segment to be identified is as follows:

[0165] (twenty one);

[0166] in, Represents the static phase decision value of the latest data segment to be identified, represents the estimated value of the stationary phase gait probability of the vth voting set group, represents the voting membership corresponding to the vth voting set group, and A represents the preset number.

[0167] The calculation formula of the non-stationary phase decision value of the latest data segment to be identified is as follows:

[0168] (twenty two);

[0169] in, represents the non-stationary phase decision value of the latest data segment to be identified, represents the estimated value of the non-stationary phase gait probability of the vth voting set, represents the voting membership corresponding to the vth voting set group, and A represents the preset number.

[0170] Step S107 , determining whether the static phase decision value is greater than the non-static phase decision value; if so, determining that the gait of the latest data segment to be identified is the static phase; otherwise, determining that the gait of the latest data segment to be identified is the non-static phase.

[0171] Specifically, when the static phase decision value is greater than the non-static phase decision value, the gait of the latest data segment to be identified is determined to be in the static phase, and the right boundary of the latest data segment to be identified is considered the last sample point of the current step, and the left boundary of the latest data segment to be identified is considered the right boundary of the previous step. When the static phase decision value is less than or equal to the non-static phase decision value, the gait of the latest data segment to be identified is determined to be in the non-static phase and does not contain any step boundary points.

[0172] Repeating the above steps S101-S107 can divide all gaits of the sensor data to be divided.

[0173] The adaptive gait segmentation method provided by an embodiment of the present invention primarily extracts trend strength features from data segments detected by sensor data with fluctuations below a certain threshold. Based on the established correlation between feature sets in an existing feature library, the method then performs fuzzy classification on the data segments to be identified based on this correlation information and the gait category of the feature set. The fuzzy classification results are then weighted and summed based on the correlation with the feature set to determine the segment to be identified. Finally, the last sampling point of the data segment to be identified, determined to be stationary, is used as the boundary of the gait signal for that segment. This method maintains good gait recognition accuracy and stability in complex situations involving a wide range of individual pedestrian speeds.

[0174] Example 2

[0175] In order to solve the technical problems existing in the prior art, an embodiment of the present invention provides a gait adaptive recognition device.

[0176] Figure 6 FIG2 shows a schematic diagram of the structure of the adaptive gait division device according to the second embodiment of the present invention; Figure 6 As shown, the adaptive gait segmentation device of an embodiment of the present invention includes a latest data segment acquisition module to be identified, a target trend intensity feature acquisition module, a voting set group acquisition module, a cluster center acquisition module, a gait probability estimation value acquisition module, a decision value acquisition module and a stage judgment module.

[0177] The latest data segment to be identified acquisition module is used to obtain the sampling point data of the sensor data to be divided in real time, and obtain the latest sliding window sample data segment according to the preset sliding window acquisition method, calculate the variance of the latest sliding window sample data segment, and obtain the latest data segment to be identified according to the preset data segment acquisition method;

[0178] The target trend strength feature acquisition module is used to acquire the trend strength feature of the latest data segment to be identified based on a preset trend strength acquisition method, and use the trend strength feature of the latest data segment to be identified as the target trend strength feature;

[0179] The voting set group acquisition module is used to calculate the membership between the target trend strength feature and each trend strength feature set group in the trend strength feature library, and select a preset number of memberships from all memberships from large to small as voting memberships, and use the trend strength feature set group corresponding to each voting membership as the voting set group;

[0180] The cluster center acquisition module is used to classify all trend intensity feature sets in each voting set group based on the gait labels in each trend intensity set to obtain a stationary phase subset and a non-stationary phase subset of each voting set group, and calculate the cluster center of each stationary phase subset and the cluster center of each non-stationary phase subset respectively;

[0181] The gait probability estimation value acquisition module is used to calculate the Euclidean distance between the target trend intensity feature and the cluster center of each stationary phase subset, as well as the Euclidean distance between the target trend intensity feature and the cluster center of each non-stationary phase subset, and obtain the stationary phase gait probability estimation value and the non-stationary phase gait probability estimation value of the corresponding voting set group based on the Euclidean distance corresponding to each voting set group;

[0182] The decision value acquisition module is used to weight and sum all the estimated gait probabilities of the stationary phase using the corresponding voting membership as weights to obtain the decision value of the stationary phase of the latest data segment to be identified, and to weight and sum all the estimated gait probabilities of the non-stationary phase to obtain the decision value of the non-stationary phase of the latest data segment to be identified;

[0183] The stage judgment module is used to judge whether the static stage decision value is greater than the non-static stage decision value. If so, the gait of the latest data segment to be identified is judged to be the static stage; otherwise, the gait of the latest data segment to be identified is judged to be the non-static stage.

[0184] The adaptive gait segmentation device provided by an embodiment of the present invention is based on a large number of reference gait data sets containing a wide range of influencing parameter data, obtains a trend intensity feature library with gait labels, and then classifies the sensor data to be segmented based on the trend intensity feature, that is, obtains the degree of correlation between the sensor data to be segmented and the different feature groups in the trend intensity feature library. Finally, the correlation degree is used as a weight parameter to make a centralized decision on the results of the fuzzy classification, so that the step segmentation algorithm can meet the requirements of cross-individual and wide-speed domains in actual application scenarios. The present invention improves the accuracy of step recognition for different pedestrian individuals and at different walking speeds, and helps pedestrian indoor positioning systems meet the requirements for positioning system robustness in actual applications.

[0185] Example 3

[0186] In order to solve the above technical problems existing in the prior art, an embodiment of the present invention further provides a storage medium storing a computer program, which, when executed by a processor, can implement all the steps in the adaptive gait segmentation method in embodiment 1.

[0187] The specific steps of the adaptive gait division method and the beneficial effects obtained by applying the readable storage medium provided in the embodiment of the present invention are the same as those in the first embodiment and will not be described in detail here.

[0188] It should be noted that the storage medium includes various media that can store program codes, such as ROM, RAM, magnetic disk or optical disk.

[0189] Example 4

[0190] In order to solve the above technical problems existing in the prior art, an embodiment of the present invention further provides a terminal.

[0191] Figure 7 The schematic diagram of the structure of the fourth terminal of the embodiment of the present invention is shown. Figure 7 In this embodiment, the terminal includes a processor and a memory connected to each other; the memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory, so that when the terminal executes, all steps in the adaptive gait segmentation method of embodiment 1 can be implemented.

[0192] The specific steps of the adaptive gait division method and the beneficial effects obtained by applying the terminal provided by the embodiment of the present invention are the same as those in the first embodiment, and are not described in detail here.

[0193] It should be noted that the memory may include random access memory (RAM) and non-volatile memory (non-volatile memory), such as at least one disk storage device. Similarly, the processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0194] Although the embodiments disclosed herein are as described above, the contents described herein are merely embodiments for facilitating understanding of the present invention and are not intended to limit the present invention. Any person skilled in the art to which the present invention belongs may make any modifications and variations in the form and details of the embodiments without departing from the spirit and scope of the present invention.

Claims

1. An adaptive gait segmentation method, comprising: Acquire sampling point data of the sensor data to be divided in real time, obtain the latest sliding window sample data segment according to a preset sliding window acquisition method, calculate the variance of the latest sliding window sample data segment, and obtain the latest data segment to be identified according to a preset data segment acquisition method; The trend strength feature of the latest data segment to be identified is obtained based on a preset trend strength acquisition method, and the trend strength feature of the latest data segment to be identified is used as the target trend strength feature; wherein, multiple time scales are set as sliding window widths, and the target data segment is slidingly sampled using the preset sliding window acquisition method to obtain all sliding window data segments of each time scale, and a moving average sequence of each time scale is obtained based on all sliding window data segments of each time scale, and the most obvious trend data segment is obtained from the target data segment using the moving average sequence of each time scale. After the most obvious trend data segment is obtained, the trend strength feature of the most obvious trend data segment is calculated; Calculating the membership between the target trend strength feature and each trend strength feature set group in the trend strength feature library, and selecting a preset number of memberships from all the memberships from largest to smallest as voting memberships, and taking the trend strength feature set group corresponding to each voting membership as the voting set group; Based on the gait labels in each trend strength set, classify all trend strength feature sets in each voting set group to obtain a stationary phase subset and a non-stationary phase subset of each voting set group, and calculate the cluster center of each stationary phase subset and the cluster center of each non-stationary phase subset; Calculating the Euclidean distance between the target trend intensity feature and each of the stationary phase subset cluster centers, and the Euclidean distance between the target trend intensity feature and each of the non-stationary phase subset cluster centers, and obtaining a stationary phase gait probability estimate value and a non-stationary phase gait probability estimate value corresponding to each of the voting set groups based on the Euclidean distance corresponding to each of the voting set groups; Using the corresponding voting membership as weight, weighting and summing all the static phase gait probability estimation values ​​respectively to obtain the static phase decision value of the latest data segment to be identified, and weighting and summing the non-static phase gait probability estimation values ​​respectively to obtain the non-static phase decision value of the latest data segment to be identified; Determine whether the static phase decision value is greater than the non-static phase decision value, and if so, determine that the gait of the latest data segment to be identified is the static phase; otherwise, determine that the gait of the latest data segment to be identified is the non-static phase; The process of acquiring the trend strength feature library is as follows: Acquire a reference gait dataset, and group all reference gait data segments in the reference gait dataset based on an influencing parameter to obtain a plurality of reference gait data groups, wherein the reference gait dataset includes a plurality of reference gait data segments and a gait label corresponding to each reference gait data segment; Acquire all reference sliding window sample data segments of each reference gait data segment by a preset sliding window acquisition method, calculate the variance of each reference sliding window sample data segment, and acquire a reference data segment of each reference gait data segment by a preset data segment acquisition method; Acquire the trend strength feature of each reference data segment based on a preset trend strength acquisition method, and group the trend strength feature corresponding to each reference gait data segment and its corresponding gait label into a trend strength feature set, and group all trend strength feature sets corresponding to each reference gait data group into a trend strength feature set group, and all trend strength feature set groups form a trend strength feature library; Wherein, the influencing parameters include gender, height and walking speed.

2. The method according to claim 1, characterized in that The preset sliding window acquisition method is: Calculating the acceleration modulus value of the target sampling point data as the target acceleration modulus value, and adding the target acceleration modulus value to the end of the sliding window queue, and sequentially selecting w acceleration modulus values ​​from the sliding window queue starting from the target acceleration modulus value as the sliding window sample data segment; Wherein, when the targeted data is sensor data to be divided, the target sampling point data is the latest sampling point data of the sensor data to be divided, and the obtained sliding window sample data segment is the latest sliding window sample data segment; When the targeted data is a reference gait data segment, the sampling point data in the reference gait data segment is sequentially used as the target sampling point data, and the obtained sliding window sample data segment is the reference sliding window sample data segment; When the targeted data is a target data segment, the sampling point data in the target data segment is sequentially used as the target sampling point data, and the obtained sliding window sample data segment is a sliding window data segment.

3. The method according to claim 2, characterized in that The preset data segment is obtained in the following manner: The target variance is placed at the end of the target variance queue. If the target variance is less than a preset variance threshold, and the variance of the previous bit of the target variance in the current target variance queue is greater than the preset variance threshold, the sampling point data corresponding to the acceleration modulus value at the end point of the sliding window sample data segment corresponding to the target variance is determined to be the starting point of the target data segment. If the target variance is greater than a preset variance threshold, and the variance of the previous target variance in the current target variance queue is less than the preset variance threshold, then the sampling point data corresponding to the acceleration modulus value of the end point of the sliding window sample data segment corresponding to the target variance is determined to be the end point of the target data segment; Whenever an end point of the target data segment is determined, the sampling point data from the most recently determined start point of the target data segment to the most recently determined end point of the target data segment in the currently targeted data is used as the target data segment; Wherein, when the targeted data is sensor data to be divided, the target variance is the variance of the latest sliding window sample data segment, the target variance queue is the latest variance queue, and the target data segment is the latest data segment to be identified; When the targeted data is a reference gait data segment, the target variance is the variance of the reference sliding window sample data segment, the target variance queue is a reference variance queue, and the target data segment is a reference data segment.

4. The method according to claim 3, characterized in that Acquiring the trend strength feature of the target data segment based on a preset trend strength acquisition method includes: Using multiple time scales as sliding window widths, the target data segments are respectively subjected to sliding sampling by a preset sliding window acquisition method to obtain sliding window data segments of each time scale; Based on all sliding window data segments of each time scale, obtain the moving average sequence of each time scale; All trend data segments are intercepted from the target data segment based on the signal trend condition, and the one with the longest data length is selected from all the trend data segments as the most obvious trend data segment, wherein all the sampling point data in the trend data segment must satisfy: all the moving average values ​​corresponding to a single sampling point data must satisfy the signal trend condition, and all the moving average values ​​corresponding to the sampling point data include: when the sampling point data is used as the target sampling point data, the moving average values ​​are obtained based on the sliding window data segments of each time scale; Calculating a trend strength feature of the most obvious trend data segment, the trend strength feature including a ratio of the time length of the most obvious trend data segment to the target data segment, a signal energy of the most obvious trend data segment, an average energy value of the most obvious trend data segment, and a slope of the most obvious trend data segment; The signal trend condition is as follows: all the moving average sequences increase or decrease in sequence according to the order of increasing or decreasing time scales; and the target data segment is the latest data segment to be identified or the reference data segment.

5. The method according to claim 4, characterized in that The moving average sequence of a single time scale obtained based on all sliding window data segments of the time scale is: Wherein, represents the kth moving average value in the moving average sequence with a time scale of w, and represents the acceleration modulus value of the kth sampling point data in the target data segment.

6. The method according to claim 1, characterized in that Calculating the membership between the target trend strength feature and each of the individual trend strength feature sets in the trend strength feature library includes: ； in, Indicates the target trend strength characteristic, Represents the cluster center of the current trend strength feature set, represents the Euclidean distance between the target trend intensity feature and the cluster center of the current trend intensity feature set, Indicates the trend strength characteristics, Grouped separately with all trend strength feature sets The sum of the membership degrees between represents the cluster center of the wth group of trend strength feature sets in the trend strength feature library, O is the number of trend strength feature sets in the trend strength feature library, represents the Lagrange multiplier.

7. An adaptive gait classification device, characterized in that: It includes the latest data segment acquisition module to be identified, the target trend strength feature acquisition module, the voting set group acquisition module, the cluster center acquisition module, the gait probability estimation value acquisition module, the decision value acquisition module and the stage judgment module; The latest data segment to be identified acquisition module is used to acquire the sampling point data of the sensor data to be divided in real time, and obtain the latest sliding window sample data segment according to a preset sliding window acquisition method, calculate the variance of the latest sliding window sample data segment, and obtain the latest data segment to be identified according to the preset data segment acquisition method; The target trend strength feature acquisition module is used to acquire the trend strength feature of the latest data segment to be identified based on a preset trend strength acquisition method, and use the trend strength feature of the latest data segment to be identified as the target trend strength feature; wherein, multiple time scales are set as sliding window widths, and sliding sampling is performed on the target data segment using the preset sliding window acquisition method to obtain all sliding window data segments of each time scale, and a moving average sequence of each time scale is obtained based on all sliding window data segments of each time scale. The most obvious trend data segment is obtained from the target data segment using the moving average sequence of each time scale. After the most obvious trend data segment is obtained, the trend strength feature of the most obvious trend data segment is calculated; The voting set group acquisition module is used to calculate the membership between the target trend strength feature and each trend strength feature set group in the trend strength feature library, and select a preset number of memberships from all the memberships from large to small as voting memberships, and use the trend strength feature set group corresponding to each voting membership as the voting set group; The cluster center acquisition module is used to classify all trend strength feature sets in each voting set group based on the gait labels in each trend strength set, so as to obtain a stationary phase subset and a non-stationary phase subset of each voting set group, and respectively calculate the cluster center of each stationary phase subset and the cluster center of each non-stationary phase subset; The gait probability estimation value acquisition module is used to calculate the Euclidean distance between the target trend intensity feature and each of the stationary phase subset cluster centers, and the Euclidean distance between the target trend intensity feature and each of the non-stationary phase subset cluster centers, and obtain the stationary phase gait probability estimation value and the non-stationary phase gait probability estimation value corresponding to each of the voting set groups based on the Euclidean distance corresponding to the voting set group; The decision value acquisition module is configured to weight and sum all the estimated gait probabilities of the stationary phase using the corresponding voting membership as weights to obtain the stationary phase decision value of the latest data segment to be identified, and to weight and sum all the estimated gait probabilities of the non-stationary phase to obtain the non-stationary phase decision value of the latest data segment to be identified; The stage judgment module is used to judge whether the static stage decision value is greater than the non-static stage decision value, and if so, judge the gait of the latest data segment to be identified as the static stage; otherwise, judge the gait of the latest data segment to be identified as the non-static stage; The process of acquiring the trend strength feature library is as follows: Acquire a reference gait dataset, and group all reference gait data segments in the reference gait dataset based on an influencing parameter to obtain a plurality of reference gait data groups, wherein the reference gait dataset includes a plurality of reference gait data segments and a gait label corresponding to each reference gait data segment; Acquire all reference sliding window sample data segments of each reference gait data segment by a preset sliding window acquisition method, calculate the variance of each reference sliding window sample data segment, and acquire a reference data segment of each reference gait data segment by a preset data segment acquisition method; Acquire the trend strength feature of each reference data segment based on a preset trend strength acquisition method, and group the trend strength feature corresponding to each reference gait data segment and its corresponding gait label into a trend strength feature set, and group all trend strength feature sets corresponding to each reference gait data group into a trend strength feature set group, and all trend strength feature set groups form a trend strength feature library; Wherein, the influencing parameters include gender, height and walking speed.

8. A storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the adaptive gait segmentation method according to any one of claims 1 to 6 is implemented.

9. A terminal, characterized in that: include: a processor and a memory, wherein the memory is communicatively connected to the processor; The memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory, so that the terminal performs the adaptive gait segmentation method according to any one of claims 1 to 6.

Images