WO2023178594A1 - 动作的计数方法、装置、设备及存储介质 - Google Patents

动作的计数方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2023178594A1
WO2023178594A1 PCT/CN2022/082721 CN2022082721W WO2023178594A1 WO 2023178594 A1 WO2023178594 A1 WO 2023178594A1 CN 2022082721 W CN2022082721 W CN 2022082721W WO 2023178594 A1 WO2023178594 A1 WO 2023178594A1
Authority
WO
WIPO (PCT)
Prior art keywords
motion feature
target
current
acceleration signal
feature information
Prior art date
Application number
PCT/CN2022/082721
Other languages
English (en)
French (fr)
Inventor
才正国
Original Assignee
广东高驰运动科技股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 广东高驰运动科技股份有限公司 filed Critical 广东高驰运动科技股份有限公司
Priority to PCT/CN2022/082721 priority Critical patent/WO2023178594A1/zh
Publication of WO2023178594A1 publication Critical patent/WO2023178594A1/zh

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63B—APPARATUS FOR PHYSICAL TRAINING, GYMNASTICS, SWIMMING, CLIMBING, OR FENCING; BALL GAMES; TRAINING EQUIPMENT
    • A63B71/00—Games or sports accessories not covered in groups A63B1/00 - A63B69/00
    • A63B71/06—Indicating or scoring devices for games or players, or for other sports activities

Definitions

  • Embodiments of the present application relate to the field of smart terminal applications, for example, to an action counting method, device, equipment and storage medium.
  • counting is only for established limited types of actions. , it is impossible to count new action types or even user-created actions; secondly, the counting method strictly relies on certain characteristics of the acceleration sensor signal. Most of these characteristics are artificially set on a limited data set and are related to the execution of the action. Whether standards are relevant or not, different user behavior habits can easily introduce interference and cause the established rules to fail. Third, the counting method requires the participation of multiple sensors, which cannot meet the needs of devices with missing hardware devices.
  • Embodiments of the present application provide an action counting method, device, equipment and storage medium to solve the problem of how to count actions without established action rules.
  • embodiments of the present application provide an action counting method, applied to a wearable device, where the wearable device includes an acceleration sensor.
  • the method includes:
  • a cumulative value of the number of actions of the user is determined.
  • embodiments of the present application also provide an action counting device, including:
  • the target information acquisition module is configured to obtain target motion characteristic information
  • the current information determination module is configured to determine the user's current motion characteristic information based on the current acceleration signal
  • the number of action determination module is configured to determine the cumulative value of the number of actions of the user based on the current motion feature information and the target motion feature information.
  • embodiments of the present application further provide a wearable device, including: a memory and at least one processor;
  • the memory is configured to store at least one program
  • the at least one processor When the at least one program is executed by the at least one processor, the at least one processor is caused to implement the action counting method described in the first aspect.
  • embodiments of the present application further provide a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform the counting method of the actions described in the first aspect.
  • Figure 1 is a schematic flowchart of an action counting method provided in Embodiment 1 of the present application.
  • FIG. 2 is a schematic flowchart of another action counting method provided in Embodiment 2 of the present application.
  • Figure 3 is an example diagram of the approximate coding sequence determination process provided by the embodiment of the present application.
  • Figure 4 is an example diagram of sliding update provided by the embodiment of the present application.
  • Figure 5 is an example diagram for determining the overlap ratio provided by the embodiment of the present application.
  • Figure 6 is a structural block diagram of an action counting device provided in Embodiment 3 of the present application.
  • Figure 7 is a structural block diagram of a wearable device provided in Embodiment 4 of the present application.
  • FIG. 1 is a schematic flowchart of an action counting method provided in Embodiment 1 of the present application. This embodiment can be applied to situations where actions without preset action rules are counted.
  • the method may be performed by an action counting device, which may consist of hardware and/or software, and may generally be integrated in a wearable device including an acceleration sensor.
  • the following takes fitness actions as an example to illustrate the execution process of the action counting method.
  • the way to count fitness actions is usually by comparing whether the characteristics of fitness actions meet the set action rules and matching multiple repeated actions to realize the counting function of fitness actions.
  • the set action rules are often for action types in related technologies, are set manually and are known, and the actions to be matched are obtained by retrieving the set action rules in the rule library.
  • the feature information corresponding to the action to be matched is unknown, and it is impossible to match this type of action using existing technology. Therefore, counting cannot be achieved; in addition, whether the user actions are performed to a standard or not, different user behavior habits can easily introduce interference and cause the established action rules to fail.
  • the counting methods in the related art only target certain limited types of actions, it is necessary to find an action counting method that is more suitable for situations where there are no set action rules.
  • Embodiment 1 of the present application provides an action counting method, which includes the following operations:
  • the user's motion data needs to be compared with the set action rules. If the motion data meets the set action rules, the fitness actions can be counted and accumulated. If the data does not meet the set action rules, the fitness action will not be accumulated.
  • the set action rules are known. During comparison, the action to be matched can be obtained by calling the set action rules in the fitness action storage library. This technical solution is mainly aimed at the situation where there are no preset action rules. It can also be understood that the fitness action storage library does not store the action rules corresponding to the actions to be matched, so the characteristic information of the actions to be matched needs to be obtained in advance.
  • the target motion feature information can be understood as a set of feature information obtained through analysis and processing including the actions to be matched. After obtaining the target motion characteristic information, the user's motion data can be matched with the target motion characteristic information to determine the number of fitness actions.
  • the method is applied to a wearable device, and the wearable device includes an acceleration sensor.
  • Wearable devices can be any devices with corresponding functions, such as smart watches, smart bracelets, smart shoes, etc.
  • the wearable device can obtain the user's starting action information.
  • the starting action information can be the starting acceleration signal obtained through the acceleration sensor.
  • the initial acceleration signal can be transformed, the feature values therein can be extracted, and the set of feature values can be used as the target motion feature information.
  • the target motion feature information can be expressed in the form of a vector set.
  • the three-axis acceleration signal is used as the original data, multi-scale transformation is applied, and multiple wavelet transformed output signals are obtained in three axes and multiple scales, and a peak screening strategy is implemented.
  • a peak screening strategy is implemented.
  • the confidence degree of the fusion action duration stability and action form consistency is calculated for the adjacent interval signals, and the similarity and stability of the two action periods before and after (that is, the two adjacent interval signals) are quantified into a probability model.
  • the key information of the recent stable and repeated actions is automatically obtained and stored as target motion feature information for subsequent matching.
  • the target motion feature information is determined without using complex clustering operations. The distinction between different actions is preserved. It should be noted that fitness actions usually have significant features on a certain coordinate axis. Therefore, while determining the target motion feature information, it can also determine which coordinate axis the acceleration signal of the user can perform this fitness action by analyzing. Obtain motion characteristic information of fitness movements.
  • step S110 When the user turns on the wearable device and continues to perform fitness actions, motion data will continue to be generated and stored in the preset data cache area. Obtain the current acceleration signal within the sliding window from the data buffer area. In order to remove noise interference, the current acceleration signal can be filtered. Process the current acceleration signal to obtain current motion feature information that characterizes the current action. Considering that the counting method of fitness actions is to compare the current motion feature information with the target motion feature information, therefore, the representation form of the current motion feature information should be consistent with the representation form of the target motion feature information, and the current motion feature is determined based on the current acceleration signal. The steps for obtaining target motion feature information are the same.
  • step S110 while determining the target motion characteristic information, it is also possible to determine which coordinate axis acceleration signal of the user's current fitness action can be analyzed to obtain the motion characteristic information of the fitness action. Therefore, this step can extract the acceleration signal corresponding to the coordinate axis from the current acceleration signal, and only needs to process the acceleration signal corresponding to the single axis to obtain the corresponding current motion feature information.
  • the target motion feature information obtained in step S110 is used as the feature information corresponding to the action to be matched
  • the current motion feature information obtained in step S120 is matched with the target motion feature information
  • the cumulative number of user fitness actions is determined based on the matching results. value. If the current motion feature information matches the target motion feature information, it indicates that the current action meets the fitness action standards, and the number of user fitness actions is accumulated; if the current motion feature information does not match the target motion feature information, it indicates that the current action does not meet the fitness action standard. If the action standard is used, the number of user fitness actions will not be accumulated. In order to improve the accuracy of counting, this step adopts a hierarchical matching strategy to transform the counting problem into a pattern matching problem.
  • the current motion feature information and the target motion feature information are split into two parts of vectors.
  • the first part of the vector of the current motion feature information and the first part of the target motion feature information Perform a relatively simple overlapping area ratio calculation; then, calculate the cross-correlation coefficient based on the second part vector of the current motion feature information and the second part vector of the target motion feature information; and finally implement quantitative statistics on fitness action counts in the form of probability.
  • each level of matching in this hierarchical matching strategy is sequential. Only when the current level meets the matching requirements, the next level of matching will be performed. Otherwise, the number of actions will not be accumulated.
  • Embodiments of the present application disclose an action counting method, device, equipment and storage medium.
  • the method is applied to a wearable device.
  • the wearable device includes an acceleration sensor.
  • the method includes: obtaining target motion characteristic information; according to the current acceleration signal, Determine the user's current motion feature information; determine the user's cumulative number of actions based on the current motion feature information and the target motion feature information.
  • the counting of actions in related technologies is often based on established limited types of action types, it is impossible to count new action types or even user-created actions, and the counting method strictly relies on artificially set signal characteristics. Different users Whether the action is standard or not, it is easy to introduce interference and cause the established rules to fail. To address this problem, this technical solution does not make constraints and assumptions on the execution of the action.
  • the target motion characteristic information that characterizes the action is obtained. After the target motion feature information is determined, the current motion feature information is matched with the target motion feature information to determine the cumulative value of the number of actions, which improves the flexibility of action counting.
  • FIG. 2 is a schematic flowchart of another action counting method provided in Embodiment 2 of the present application. This embodiment can be applied to situations where actions without preset action rules are counted.
  • the method may be performed by an action counting device, which may consist of hardware and/or software, and may generally be integrated in a wearable device including an acceleration sensor.
  • FIG. 2 As shown in Figure 2, another action counting method provided by Embodiment 2 of the present application includes the following operations:
  • the wearable device includes a three-axis acceleration sensor.
  • motion data will be generated and stored in the data cache.
  • the starting point is extracted from the data cache in a sliding window.
  • Acceleration signal the initial acceleration signal can be understood as the first data generated by the user's movement, or the first data stored in the data buffer area. This signal can be used as the basic data for obtaining target motion characteristic information.
  • only the initial acceleration signals corresponding to the three axes are analyzed. In order to improve the counting detection rate of compound actions and complex actions, the repeatability of the action signals is analyzed at different granularities.
  • multi-scale transformation analysis is performed on the starting acceleration signal corresponding to the three axes. Under the three axes and multiple scales, multiple output signals after wavelet transformation are obtained.
  • the peak screening strategy effectively extracts the most critical and obvious motion interval segmentation points, that is, the target peak position sequence. This step only requires convolution calculations on a specified scale, and can complete scene coverage of common strength training action frequencies with a certain computational complexity and storage overhead.
  • the target peak position sequence high-quality interval signal division is obtained; considering the similarity and stability of adjacent interval signals, the confidence of adjacent interval signals can be calculated; motion features are extracted for continuous and stable high-confidence interval signals; When the confidence level of a set number of consecutive interval signals is higher than a certain set value, the current motion features can be clustered to form target motion feature information.
  • the target axis is the coordinate axis associated with the target motion feature information. It should be noted that after the target motion feature information is determined according to step S220, the target axis associated with the target motion feature information is also determined. After using the target axis as the acceleration coordinate axis to be tracked, the subsequent counting process of fitness actions only requires motion data analysis on the selected target axis.
  • the current acceleration signal is obtained from the data cache in a sliding window, and the acceleration signal corresponding to the target axis is extracted from the current acceleration signal as the target acceleration signal.
  • the steps for determining the user's current motion feature information are the same as the steps for determining the target motion feature information.
  • the target motion feature information is determined based on the three-axis acceleration signal, while the current motion feature information is based on A single target axis acceleration signal is determined.
  • the macroscopic form of the action and the microscopic details of the action are considered simultaneously, and the overlapping area of the feature information corresponding to the set score corresponding to the current motion feature information and the target motion feature information is calculated, and the current motion feature information and the target motion feature information are
  • the corresponding approximate coding sequence performs cross-correlation coefficient calculation on the corresponding feature information.
  • the current motion feature information includes feature information corresponding to the set score and feature information corresponding to the approximate coding sequence.
  • the current motion feature information is divided, the feature information corresponding to the set score is determined as the first current motion feature vector, and the feature information corresponding to the approximate coding sequence is determined as the second current motion feature vector.
  • the current motion feature information is expressed as: ⁇ frc 90 ,frc 10 ,s 1, s 2, ,s 3, ,s 4, ,s 5, ,s 6, ,s 7 , ,s 8 ⁇ , for After it is divided, the first current motion feature vector is determined to be ⁇ frc 90 , frc 10 ⁇ , and the second current motion feature vector is determined to be ⁇ s 1, ,s 2, , s 3, ,s 4, ,s 5, ,s 6, ,s 7, ,s 8 ⁇ .
  • the target motion feature information includes feature information corresponding to the set score and feature information corresponding to the approximate coding sequence.
  • the target motion feature information is divided, the feature information corresponding to the set score is determined as the first target motion feature vector, and the approximate The feature information corresponding to the coding sequence is determined as the second target motion feature vector.
  • the target motion feature information is expressed as: After dividing it, the first target motion feature vector is determined to be Determine the second target motion feature vector as
  • This step uses a hierarchical matching strategy to transform the counting problem into a pattern matching problem.
  • a relatively simple overlapping area ratio calculation is performed based on the first current motion feature vector and the first target motion feature vector; then, a cross-correlation coefficient calculation of the approximate coding sequence is performed based on the second current motion feature vector and the second target motion feature vector. ; Finally, based on the fusion confidence (that is, the confidence of the fusion of action duration stability and action form consistency), overlapping area ratio, and cross-correlation coefficient, the action matching probability is obtained, and quantitative statistics of fitness action counts are implemented in the form of probability.
  • the overlapping area ratio and cross-correlation coefficient are considered as macro-morphological and micro-detail factors of the action respectively, and the matching results are output in the form of probability, converting the counting problem into a pattern matching problem to quantify the probability.
  • the form realizes the statistics of action counting and improves the accuracy of counting.
  • multi-scale transformation analysis is performed on the initial acceleration signal, and the steps of determining the target peak position sequence are:
  • the multi-scale transformation can use the wavelet transform whose wavelet base is Harr wavelet.
  • the selection of the set scale will affect the frequency distribution range of daily strength training movements.
  • the determination of the set scale can be statistically obtained based on a certain scale of real user data sets.
  • the wavelet base is Harr wavelet
  • the scale frequencies are 0.4, 1.0, 1.5, 2.0, 2.5Hz
  • the sampling frequency is 25Hz
  • the lengths of the sequences are respectively 25, 40, 60, 120, 250.
  • the three-axis acceleration signal of acceleration is subjected to the multi-scale transformation of the above five scales.
  • the following uses the wavelet transformation of the x-axis acceleration on the fourth scale as an example to explain how to obtain a quantized response output.
  • the description continues with the example of S2102, and the median value is calculated for the peak height sequence ⁇ v 0 , v 1 , v 2 ,..., v P ⁇ as the axis input signal ⁇ x 0 , x 1 , x 2 , ...,x N ⁇ at this scale
  • the response output of ⁇ y 0 ,y 1 ,y 2 ,...,y N ⁇ after wavelet transform analysis under Wavelet transform analysis of all five scales is performed on the acceleration signal of the x-axis, and 5 response outputs are obtained. Find the peak position sequence corresponding to the largest response output Assume that the response output on the i-th scale maximum. Output the above 5 responses
  • the mean of is output as the overall response on the x-axis r x .
  • the algorithm should save at least the respective peak position sequences corresponding to the 3 axes and the overall response output corresponding to the 3 axes ⁇ r x , r y , r z ⁇ .
  • the target motion characteristic information is determined according to the target peak position sequence, including:
  • Each interval signal set is composed of two or more interval signals.
  • the peak position sequence corresponding to the scale with the maximum response output on the selected axis with the maximum overall response output occurs on the jth scale.
  • the merging operation must comply with the following rules: perform a "set merge" between the target peak position sequence to be merged and the current interval signal division position sequence. Operation, deduplication; the length of the merged interval signal division position sequence must be greater than or equal to 4; the merged interval signal division position sequence should be biased towards a signal that evenly divides the total length. That is, the lengths of the three adjacent interval signals are l 1 , l 2 and l 3 . It should be ensured that l 2 is not significantly smaller than l 1 and l 3 .
  • the rule selected in this step is that l 2 is greater than or equal to 1/2 of l 1 and greater than or equal to 1/2 of l 3 .
  • L x includes multiple By composed of and It is two adjacent elements in the peak position sequence that have been filtered to comply with the constraint rules.
  • S2202 Calculate the first confidence level of adjacent interval signals in each interval signal set and the second confidence level of a continuously set number of interval signals.
  • confidence is mainly used to measure the periodicity and stability of action execution. Considering that during the execution of daily strength training actions, two adjacent actions in a single group have consistency in movement speed, so it is used to evaluate adjacent intervals.
  • the first confidence level that the signal is stable is based on the time length of adjacent interval signals.
  • the first confidence level can be understood as characterizing the similarity and stability of adjacent interval signals.
  • the motion feature extraction calculation can be performed based on the interval signal.
  • calculate the first confidence level of adjacent interval signals in the interval signal set and the second confidence level of a set number of consecutive interval signals including:
  • the confidence used in this method to evaluate the stability of adjacent interval signals is based on the adjacent interval signals.
  • the length of time achieved That is, assuming that the durations in the adjacent interval signals ⁇ X a ,X b ⁇ are la and l b respectively, the ratio It should not be less than a certain threshold z 1 , and when this ratio is greater than a certain threshold z 2 , this ratio is forcibly modified to 1. This design ensures that when the execution cycles of two adjacent actions are not much different, the difference in time length does not affect the confidence level.
  • This step uses interpolation resampling of the sequence with a shorter signal length to extend its length to be consistent with the longer sequence. For example, the signal after X a interpolation and resampling
  • the durations in the adjacent interval signals ⁇ X a , X b ⁇ are determined to be la and lb respectively.
  • the interpolated resampled signal Perform cross-correlation calculation with another signal X b to obtain the cross-correlation coefficient R ab .
  • R ab is forcibly modified to 0.
  • the cross-correlation coefficient in this step is bounded by 0.5, which is obtained based on statistics of a large number of real user data. The data shows that if the cross-correlation coefficient is less than 0.5, it is considered not obviously relevant.
  • the first confidence level P conf that comprehensively considers the similarity and stability of adjacent interval signals is defined as: Since the division of interval signals is extracted after sliding windows, a certain interval signal may be in the middle of the current sliding window and at the tail of the previous sliding window. This makes the same interval signal possible in different sliding windows. Different confidence levels.
  • the confidence level is mainly used to measure the periodicity and stability of action execution. Therefore, the highest confidence level from this interval signal to the current is always selected as the input of subsequent steps.
  • the set number can be set according to requirements, for example, the value can be 4, indicating that the fitness action has been repeated 4 times.
  • the first confidence level can be understood as characterizing the similarity and stability of adjacent interval signals.
  • the first confidence condition may be that the confidence is higher than a certain set threshold.
  • the initial motion feature information can be represented by the set quantile of each interval signal and the approximate coding sequence of the interval signal.
  • determine the starting motion characteristic information corresponding to the starting acceleration signal including:
  • the approximate coding sequence of the interval signal is implemented by performing equal division processing within the set quantile range of the interval signal. For example, the range corresponding to the 90th percentile and the 10th percentile of the interval signal is divided into six equal parts, and the 90th percentile, the 10th percentile and the 5 intermediate dividing points are used as the signal's fluctuation range dividing points.
  • the split point index of the split point matching sequence is saved as an approximate encoding sequence.
  • the split point matching sequence is ⁇ g 4 , g 2 , g 1 , g 4 , g 7 , g 6 , g 5 , g 5 ⁇
  • the split point will match the split point of the sequence.
  • the index ⁇ 4,2,1,4,7,6,5,5 ⁇ is saved as an approximate encoding sequence.
  • the approximate coding sequence of the interval signal in this step describes the morphological characteristics of the signal fluctuation with as few and simple features as possible without considering the absolute size and speed of the signal. From the perspective of engineering implementation, it also greatly reduces the number of Reduce storage overhead and reduce computational matching complexity.
  • the set quantile and approximate coding sequence are determined as the starting motion characteristic information corresponding to the starting acceleration signal. For example, the 90th percentile, the 10th percentile and the approximate coding sequence of the above steps are combined to determine the starting motion characteristic information corresponding to the starting acceleration signal.
  • the target motion feature information in this step is implemented based on motion feature clustering extracted from interval signals.
  • the key information of recent stable and repeated actions is automatically obtained and stored as target motion feature information for subsequent matching.
  • the target motion feature information is determined in The distinction between different actions is retained without using complex clustering operations.
  • the second confidence level condition can be set such that the second confidence level is higher than a certain set threshold.
  • the current motion feature vector is clustered to form target motion feature information for action matching in subsequent steps.
  • determine the target motion feature information and target axis based on the starting motion feature information including:
  • the starting motion feature parameters can be understood as the mean and variance corresponding to the starting motion feature information.
  • the corresponding motion feature vectors ⁇ frc 90 , frc 10 ,s 1, s 2, ,s 3, ,s 4, ,s 5, are calculated respectively ,s 6, ,s 7, ,s 8 ⁇ , that is, the 90th quantile, the 10th quantile and the approximate coding sequence (s i corresponds to one of ⁇ g 0 ,g 1 ,...,g 7 ⁇ ), calculate corresponding mean and variance.
  • the setting conditions can be set to whether the starting motion feature information falls within a 95% high probability with the mean as the center and 3 times the standard deviation as the reference. within the range.
  • the target motion characteristic information of this action will be formed, and the currently used coordinate axis will be marked as the target axis for subsequent motion matching tracking, that is, the subsequent motion matching and tracking will be determined.
  • Strength training counts track which of the x, y, and z axes.
  • the mean corresponding to the above 90th percentile, 10th percentile and approximate coding sequence is saved to form a 10-dimensional vector.
  • This vector is stored as the target motion feature vector corresponding to the current continuous, stable and repeated action at least 4 times.
  • fv fingerprint The target motion feature vector can be expressed as:
  • the starting acceleration signal within the starting sliding window can be obtained from the preset data buffer area, and the starting acceleration signal is filtered.
  • the preset cache area can be understood as being used to store motion data generated when the user performs exercise.
  • this method can be applied to wearable devices with limited storage and computing speed (such as smart watches/bracelets), because the sampling rate of the original acceleration signal will affect whether it can be realistic
  • this embodiment needs to set a reasonable sampling rate of the acceleration signal while taking into account both power consumption and performance.
  • the sampling rate of the acceleration signal can be 25Hz, that is, 25 acceleration x-axis sampling values, 25 acceleration y-axis sampling values, and 25 z-axis sampling values are obtained every second. axis sample value.
  • the starting acceleration signal can be obtained from the data buffer in the form of data windows. Considering the duration of a single set of single movements in daily strength training exercises and the number of times a certain action may be repeated in a single set of training, the size of the data window for the initial acceleration signal will affect whether enough and enough data can be observed within the window. Continuous action information, therefore, the size of the data window length is very important.
  • FIG. 4 is an example diagram of sliding update provided by the embodiment of the present application. As shown in Figure 4, the sliding windows are t 1 , t 2 and t 3 in sequence.
  • the 3-axis original acceleration signal needs to be filtered.
  • the selection of the low-pass filtering method will affect the speed of filtering convergence, the distortion of the signal waveform, and the storage and calculation overhead. Butterworth low-pass filtering with an order of 2 can be selected, and the first 8 filtered objects will be discarded during the filtering process. data to shield the miscount caused by the jitter of the signal waveform during the filtering convergence process.
  • determining the cumulative value of the number of actions of the user based on the first current motion feature vector, the second current motion feature vector, the first target motion feature vector, and the second target motion feature vector includes:
  • FIG. 5 is an example diagram for determining the overlap ratio provided by the embodiment of the present application.
  • step b) Determine whether the overlap ratio is less than the first set threshold. If the overlap ratio is greater than or equal to the first set threshold, perform step c); if the overlap ratio is less than the first set threshold, perform step h). .
  • the first set threshold may be obtained based on statistics of real user data sets of a certain scale, or may be set by device developers. Compare the overlap ratio with the first set threshold. If the overlap ratio is less than the first set threshold, it indicates that the current motion feature information does not match the target motion feature information, and the cumulative value of the number of actions used to control the user's fitness remains unchanged. If the overlap ratio is greater than or equal to the first set threshold, the correlation coefficient between the second current motion feature vector and the second target motion feature vector may be further determined.
  • This step is performed after determining that the overlap ratio is greater than or equal to the first set threshold, performing a cross-correlation operation on the second current motion feature vector and the second target motion feature vector, and calculating the cross-correlation coefficient R s .
  • Step d) Determine whether the cross-correlation coefficient is less than the second set threshold. If the cross-correlation coefficient is greater than or equal to the second set threshold, perform step e); if the cross-correlation coefficient is less than the second set threshold, perform step e). Step h).
  • the second set threshold may be obtained based on statistics of real user data sets of a certain scale, or may be set by device developers. Compare the cross-correlation coefficient with the second set threshold. If the cross-correlation coefficient is less than the second set threshold, it indicates that the current motion feature information does not match the target motion feature information, and the cumulative value of the number of actions to control the user's fitness remains unchanged; if the cross-correlation coefficient If the number is greater than or equal to the second set threshold, the estimated probability can be further determined based on the confidence level, overlap ratio and cross-correlation coefficient.
  • e Determine the estimated probability based on the confidence level, overlap ratio and cross-correlation coefficient.
  • the confidence level is determined by the adjacent interval signals.
  • This step is performed after it is determined that the overlap ratio is greater than or equal to the first set threshold, and the cross-correlation coefficient is greater than or equal to the second set threshold.
  • Interval signals that meet the above threshold judgment can be initially considered to be likely to match the target action feature information. . It should be noted that for interval signals that meet the above threshold judgment, it is also necessary to calculate the estimated probability.
  • the estimated probability is the product of the corresponding confidence level P conf in step d1), the overlap ratio P overlap and the cross-correlation coefficient R s calculated above.
  • the confidence in step e) has the same meaning as the fusion confidence and the first confidence mentioned above.
  • step f Determine whether the estimated probability is greater than the third set threshold. If the estimated probability is greater than the third set threshold, perform step g); if the estimated probability is less than or equal to the third set threshold, perform step h). .
  • the cumulative value of the number of actions that controls the user's fitness is added by 1 as the new cumulative value of the number of actions.
  • the wearable device Through the received user motion information, the technical means provided by this solution are used to form target motion feature information and count fitness actions. During this period, User A walked to the rest room to rest, then returned to the exercise area to continue exercising, and turned off the wearable device when finishing the exercise. From the process of turning on the wearable device to turning it off, the target motion feature information will only be formed once. The action of walking to the lounge does not match the target motion feature information. The number of fitness actions will not be accumulated for the walking action, and the number of fitness actions will not be accumulated for the walking action. Form new target action feature information based on walking actions.
  • the cumulative value of the user's number of actions after determining the cumulative value of the user's number of actions, it also includes: if the counting cycle end condition is not currently satisfied, using the acceleration signal of the next sliding window as the current acceleration signal, and returning to re-execute determining the user's number of actions based on the current acceleration signal. Operation of current motion feature information; when the counting cycle end condition is currently met, the cumulative value of the number of actions is fed back as the counting result of the action.
  • the end condition of the counting cycle is: the acceleration signal is not obtained within the set time; or the cumulative value of the number of actions is set times consecutively (that is, the continuous number of times of acquiring the acceleration signal of the next window in the form of a sliding window reaches the set number). (determined number of times) has not changed.
  • the set time can be set by the user. If the acceleration signal is not obtained within the set time, it can be understood that the user stops moving within the set time and no new acceleration signal is generated. It can also be understood that the acceleration signal is not generated within the set time. If there is no new motion data in the internal data cache for the wearable device to analyze, the count can be terminated.
  • the setting times can be set by the user.
  • the cumulative value of the number of actions does not change for the consecutive set times, it can be understood that the user no longer performs fitness exercises, but does other actions or stops actions, that is, the cumulative value of the number of actions has not changed for the consecutive set times. If a change occurs, the count can be terminated. After the counting is terminated, the cumulative value of the number of movements needs to be fed back as the counting result of the fitness movements. For example, it can be presented in the form of digital display or voice broadcast on a wearable device.
  • the counting cycle end condition For example, if the counting cycle end condition is not currently met, continue to obtain the data in the buffer area, obtain the acceleration signal of the next window in the form of a sliding window as the current acceleration signal, and return to re-execute to determine the user's current acceleration signal based on the current acceleration signal.
  • Motion feature information match the current motion feature information with the acquired target motion feature information, and perform counting statistics. What is clear is that if the counting loop end condition is not currently met, the acceleration signal of the next sliding window will continue to be obtained for subsequent steps, and so on.
  • the embodiment of this application refines the process of determining the target motion characteristic information, extracts the initial acceleration signal from the data buffer area in a sliding window and performs filtering processing; only performs multi-scale transformation analysis on the three-axis acceleration signal, effectively extracting the most optimal
  • the key and obvious motion interval dividing points that is, peak position information, are used to obtain high-quality interval signal divisions; the confidence of the fusion action duration stability and action form consistency is calculated for adjacent interval signals, and the two action periods before and after are calculated.
  • the similarity and stability are quantified into a probabilistic model to reduce the dependence on artificially set rules; when extracting motion features, both the absolute characteristics of the motion signal, that is, the quantile of the acceleration signal size, and the quantile of the motion signal are also considered.
  • the relative characteristics that is, the description of the fluctuation shape of the acceleration signal, without significantly increasing the computational complexity and storage overhead, are suitable for resource-constrained devices such as smart watches/bracelets; automatically obtain key information of recent stable and repeated actions, and store them as The target motion feature information is used for subsequent matching, and the distinction between different actions is retained without using complex clustering operations.
  • Figure 6 is a structural block diagram of an action counting device provided in Embodiment 3 of the present application.
  • the device can be applied to wearable devices.
  • the device includes: a target information acquisition module 31 and a current information determination module 32 and action number determination module 33.
  • the target information acquisition module 31 is configured to acquire target motion characteristic information
  • the current information determination module 32 is configured to determine the user's current motion characteristic information based on the current acceleration signal
  • the number of action determination module 33 is configured to determine the cumulative value of the number of actions of the user based on the current motion feature information and the target motion feature information.
  • the target information acquisition module 31 may include:
  • the target sequence determination unit is configured to perform multi-scale transformation analysis on the starting acceleration signal to determine the target peak position sequence
  • the target information determining unit is configured to determine the target motion characteristic information based on the target peak position sequence.
  • the target sequence determination unit is set to:
  • a target peak position sequence is determined.
  • the target information determination unit is set to:
  • Each interval signal set is composed of two or more interval signals
  • the target motion feature information and the target axis are determined based on the starting motion feature information.
  • the step of setting the target information determination unit to respectively calculate the first confidence level of each adjacent interval signal in the interval signal set and the second confidence level of a continuously set number of interval signals may include:
  • the shorter interval signal is interpolated and resampled to keep the length of the adjacent interval signals consistent
  • the first confidence level of the adjacent interval signals and the second confidence level of the consecutively set number of interval signals are determined.
  • the target information determination unit is configured to determine the starting motion characteristic information corresponding to the starting acceleration signal, which may include:
  • the initial motion characteristic information corresponding to the initial acceleration signal is determined.
  • the target information determination unit is configured to determine the target motion feature information and the target axis based on the starting motion feature information, and the step of determining the target motion feature information and the target axis may include:
  • the starting motion characteristic information determine the starting motion characteristic parameters of a consecutive set number of interval signals
  • the starting motion feature information is clustered to obtain the target motion feature information and the target axis.
  • the current information determination module 32 may include:
  • the target acceleration signal determination unit is configured to extract the target acceleration signal of the target axis from the current acceleration signal, where the target axis is the coordinate axis associated with the target motion characteristic information;
  • the current motion feature information determination unit is configured to determine the user's current motion feature information based on the target acceleration signal.
  • the action number determination module 33 is set to:
  • the cumulative value of the number of actions of the user is determined.
  • the number of action determination module 33 is configured to determine the cumulative value of the number of actions of the user based on the first current motion feature vector, the second current motion feature vector, the first target motion feature vector, and the second target motion feature vector, include:
  • step b) Determine whether the overlap ratio is less than the first set threshold, and in response to the overlap ratio being greater than or equal to the first set threshold, perform step c); in response to the overlap ratio being less than the first set threshold, perform step c) h).
  • step d Determine whether the cross-correlation coefficient is less than the second set threshold, and in response to the cross-correlation coefficient being greater than or equal to the second set threshold, perform step e); in response to the cross-correlation coefficient being less than the second set threshold , perform step h).
  • e Determine the estimated probability based on a confidence level determined by adjacent interval signals, the overlap ratio and the cross-correlation coefficient.
  • the device also includes a loop module:
  • the loop module is set to use the acceleration signal of the next sliding window as the current acceleration signal if the counting loop end condition is not currently met, and return to re-execute the operation of determining the user's current motion characteristic information based on the current acceleration signal; if the counting loop is currently met.
  • the end condition feeds back the cumulative value of the number of actions as the counting result of the action.
  • the counting cycle end condition is: the acceleration signal is not acquired within the set time; or, the cumulative value of the number of actions does not change for the set number of consecutive times.
  • the above-mentioned device can execute the action counting method provided by all the foregoing embodiments of this application, and has corresponding functional modules for executing the above-mentioned method.
  • the action counting method provided by all the foregoing embodiments of this application, and has corresponding functional modules for executing the above-mentioned method.
  • FIG 7 is a structural block diagram of a wearable device provided in Embodiment 4 of the present application.
  • the wearable device includes a processor 41, a memory 42, an input device 43 and an output device 44; the processor in the computer device
  • the number 41 may be at least one, and one processor 41 is taken as an example in Figure 7; the processor 41, memory 42, input device 43 and output device 44 in the wearable device may be connected through a bus or other means, and in Figure 7 For example, connect via a bus.
  • the memory 42 can be configured to store software programs, computer-executable programs and modules, such as modules corresponding to the action counting method in the embodiments of the present application (for example, target information in the action counting device Acquisition module 31, current information determination module 32 and action number determination module 33).
  • the processor 41 executes software programs, instructions and modules stored in the memory 42 to execute various functional applications and data processing of the computer device, that is, to implement the above action counting method.
  • the memory 42 may mainly include a stored program area and a stored data area, where the stored program area may store an operating system and at least one application program required for a function; the stored data area may store data created based on the use of the terminal, etc.
  • memory 42 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid-state storage device.
  • memory 42 may include memory located remotely from processor 41, and these remote memories may be connected to the computer device through a network. Examples of the above-mentioned networks include but are not limited to the Internet, intranets, local area networks, mobile communication networks and combinations thereof.
  • the input device 43 may be configured to receive input numeric or character information and to generate key signal inputs related to user settings and functional controls of the computer device.
  • the output device 44 may include a display device such as a display screen.
  • Embodiment 5 of the present application also provides a storage medium containing computer-executable instructions.
  • the computer-executable instructions are used to perform an action counting method when executed by a computer processor.
  • the method includes:
  • a cumulative value of the number of actions of the user is determined.
  • the embodiments of the present application provide a storage medium containing computer-executable instructions.
  • the computer-executable instructions are not limited to the method operations described above. They can also perform the counting method of actions provided by any embodiment of the application. related operations.
  • the computer-readable storage medium It can be a non-transitory storage medium, such as a computer's floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including several instructions It is used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments of this application.
  • a computer device which may be a personal computer, a server, or a network device, etc.

Landscapes

  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Physical Education & Sports Medicine (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)

Abstract

本实施例公开一种动作的计数方法、装置、设备及存储介质，该方法应用于可穿戴设备，可穿戴设备包括加速度传感器，该方法包括：获取目标运动特征信息；根据当前加速度信号，确定用户的当前运动特征信息；基于当前运动特征信息和目标运动特征信息，确定用户的动作次数累计值。

Description

动作的计数方法、装置、设备及存储介质 技术领域
本申请实施例涉及智能终端应用领域，例如涉及一种动作的计数方法、装置、设备及存储介质。
背景技术
随着可穿戴设备的普及，智能手表/手环成为了众多健身和运动爱好者的必需品。利用智能手表/手环内丰富的传感器，对用户的多种行为和运动进行量化是主流可穿戴设备的基本功能。其中，尤其以健身、力量训练中的量化和运动统计分析更加被用户关注，用户除了需要实时监控运动期间的心率、卡路里等生理指标外，用户更需要一种可以辅助完成运动计数，动作计次等助手级别的功能。
相关技术中，有诸多不同的方法和系统在帮助智能手表/手环用户实现对健身、力量训练的动作计数，但是所述方法具有以下缺陷：第一、计数仅针对既定的有限类型的动作类型，无法对新的动作类型甚至用户自创动作做计数；第二、计数方法严格依赖于加速度传感器信号的某些特征，这些特征多数是在有限的数据集合上人为设定完成，与动作执行的标准与否相关，不同的用户行为习惯容易引入干扰而导致既定规则失效；第三、计数方法需要多种传感器的参与，无法满足在有部分硬件设备缺失的设备上的需求。
发明内容
本申请实施例提供一种动作的计数方法、装置、设备及存储介质，以解决对于无既定动作规则的动作如何进行计数的问题。
第一方面，本申请实施例提供一种动作的计数方法，应用于可穿戴设备，所述可穿戴设备包括加速度传感器，该方法包括：
获取目标运动特征信息；
根据当前加速度信号，确定用户的当前运动特征信息；
基于所述当前运动特征信息和所述目标运动特征信息，确定用户的动作次数累计值。
第二方面，本申请实施例还提供一种动作的计数装置，包括：
目标信息获取模块，设置为获取目标运动特征信息；
当前信息确定模块，设置为根据当前加速度信号，确定用户的当前运动特征信息；
动作次数确定模块，设置为基于所述当前运动特征信息和所述目标运动特征信息，确定用户的动作次数累计值。
第三方面，本申请实施例还提供一种可穿戴设备，包括：存储器以及至少一个处理器；
所述存储器，设置为存储至少一个程序；
当所述至少一个程序被所述至少一个处理器执行，使得所述至少一个处理器实现如上述第一方面所述的动作的计数方法。
第四方面，本申请实施例还提供一种包含计算机可执行指令的存储介质，所述计算机可执行指令在由计算机处理器执行时用于执行如第一方面所述的动作的计数方法。
附图说明
图1为本申请实施例一提供的一种动作的计数方法的流程示意图；
图2为本申请实施例二提供的另一种动作的计数方法的流程示意图；
图3为本申请实施例提供的近似编码序列确定过程的示例图；
图4为本申请实施例提供的滑动更新的示例图；
图5为本申请实施例提供的确定重叠比例的示例图；
图6为本申请实施例三提供的一种动作的计数装置的结构框图；
图7为本申请实施例四提供的一种可穿戴设备的结构框图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述。
需要说明的是，本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的本申请的实施例能够以除了在这里图示或描述的那些以外的顺序实施。此外，术语“包括”和“具有”以及他们的任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
实施例一
图1为本申请实施例一提供的一种动作的计数方法的流程示意图，本实施例可适用于对无既定设定动作规则的动作进行计数的情况。该方法可以由动作的计数装置来执行，该装置可由硬件和/或软件组成，并一般可集成在可穿戴设备中，可穿戴设备包括加速度传感器。下面以健身动作为例，说明动作的计数方法的执行流程。
需要说明的是，相关技术中，对健身动作的计数方式通常是通过比较健身动作的特征是否满足设定动作规则，并对多次重复的动作进行匹配，实现对健身动作的计数功能。其中，设定动作规则往往是针对相关技术中的动作类型，由人为设定且已知的，通过调取规则库中的设定动作规则得到待匹配动作。然而，对于新的动作类型甚至用户自创动作做计数时，由于规则库中无该动作对应的动作规则，待匹配动作对应的特征信息是未知的，利用现有技术无法对该类型动作进行匹配进而无法实现计数；另外，用户动作执行的标准与否，不同的用户行为习惯容易引入干扰而导致既定动作规则失效。考虑到相关技术中的 计数方法仅针对既定的有限类型的动作类型，因此需要找到一种更适合无既定设定动作规则情况下的动作计数方法。
如图1所示，本申请实施例一提供一种动作的计数方法，包括如下操作：
S110、获取目标运动特征信息。
需要知道的是，相关技术中在进行健身动作计数时，需要将用户的运动数据与设定动作规则进行比较，若运动数据满足设定动作规则，则可以对该健身动作进行计数累加，若运动数据不满足设定动作规则，则不累加该健身动作。设定动作规则是已知的，在进行比较时，可以通过调取健身动作存储库中的设定动作规则，来获取待匹配动作。而本技术方案主要针对无既定设定动作规则的情况，也可以理解为健身动作存储库中没有存储待匹配动作对应的动作规则，因此需要预先获取待匹配动作的特征信息。其中，目标运动特征信息可以理解为经过分析处理得到的包含待匹配动作的特征信息集合。在获取目标运动特征信息之后，可以将用户的运动数据与该目标运动特征信息进行匹配，用以确定健身动作次数。
本实施例中，该方法应用于可穿戴设备上，可穿戴设备包括加速度传感器。可穿戴设备可以是任一具有相应功能的设备，例如可以是智能手表、智能手环、智能鞋等。当用户想要通过可穿戴设备对健身动作进行计数时，可以穿戴好可穿戴设备，在开始运动前开启该可穿戴设备，并进行健身动作。用户进行健身动作后，可穿戴设备可以获得用户的起始动作信息，起始动作信息可以是通过加速度传感器获取的起始加速度信号。在获得起始加速度信号之后，可以对起始加速度信号进行变换，提取其中的特征值，将特征值的集合作为目标运动特征信息。示例性的，目标运动特征信息可以以向量集合的形式表示。
示例性的，首先，本实施例中以三轴加速度信号为原始数据，应用多尺度变换，在三个轴和多个尺度下，得到多个小波变换后的输出信号，并实施峰筛选策略，通过对输出信号做峰提取，按照设定规则选取某个尺度和某个坐标轴下的峰序列，有效提取最关键和明显的运动区间分割点，即峰值位置信息。将 经过筛选得到的峰值位置信息合并到当前的区间信号划分位置序列中，得到高质量的区间信号划分。
其次，对相邻区间信号做融合动作时长稳定性和动作形态一致性的置信度计算，将前后两个动作周期(即相邻的两个区间信号)的相似性和稳定性量化为概率模型，减少对人为设定的规则的依赖。对连续稳定的高置信度区间信号提取运动特征，既考虑运动信号的绝对特点，即加速度信号大小的分位数，也考虑运动信号的相对特点，即加速度信号的波动形态描述。
最后基于区间信号提取的运动特征，自动获取最近稳定且重复动作的关键信息，存储为目标运动特征信息用做后续的匹配，目标运动特征信息的确定在不采用复杂的聚类操作的前提下，保留了不同动作之间的区分度。需要说明的是，健身动作通常是在某一个坐标轴上有显著的特征，因此，在确定目标运动特征信息的同时，可以确定该用户此次健身动作通过对哪个坐标轴的加速度信号进行分析可以获得健身动作的运动特征信息。
S120、根据当前加速度信号，确定用户的当前运动特征信息。
本步骤是在步骤S110之后进行的，当用户开启可穿戴设备后，用户持续进行健身动作，则会一直产生运动数据并存储在预设的数据缓存区中。从数据缓存区中获取滑动窗口内的当前加速度信号，为了去除噪声干扰可以对当前加速度信号进行滤波处理。对当前加速度信号进行处理，获得表征当前动作的当前运动特征信息。考虑到健身动作的计数方式是通过将当前运动特征信息与目标运动特征信息进行比较，因此，当前运动特征信息的表征形式应与目标运动特征信息的表征形式一致，根据当前加速度信号确定当前运动特征信息与获取目标运动特征信息的步骤是一样的。需要说明的是，由步骤S110可知，在确定目标运动特征信息的同时，可以确定该用户此次健身动作通过对哪个坐标轴的加速度信号进行分析可以获得健身动作的运动特征信息。因此，本步骤可以从当前加速度信号中提取对应坐标轴的加速度信号，只需要对单轴对应的加速度信号进行处理，获得相应的当前运动特征信息。
S130、基于当前运动特征信息和目标运动特征信息，确定用户的动作次数累计值。
本实施例中，将步骤S110获得的目标运动特征信息作为待匹配动作对应的特征信息，将步骤S120获得的当前运动特征信息与目标运动特征信息进行匹配，根据匹配结果确定用户健身的动作次数累计值。若当前运动特征信息与目标运动特征信息相匹配，表明当前动作符合健身动作标准，则对用户健身的动作次数进行累加；若当前运动特征信息与目标运动特征信息不匹配，表明当前动作不符合健身动作标准，则不对用户健身的动作次数进行累加。为了提高计数的准确率，本步骤采用分级匹配的策略，将计数问题转化为模式匹配问题。首先，从动作宏观形态和动作微观细节因素进行考量，将当前运动特征信息和目标运动特征信息均拆分成两部分向量，根据当前运动特征信息的第一部分向量和目标运动特征信息的第一部分向量进行较为简单的重叠面积比例计算；然后，根据当前运动特征信息的第二部分向量和目标运动特征信息的第二部分向量进行互相关系数计算；最后以概率的形式实现对健身动作计数的量化统计。需要说明的是，该分级匹配的策略中的每一级匹配是有先后顺序的，只有当前一级满足匹配要求时，才会进行下一级的匹配，否则会直接对该动作次数不累加。
本申请实施例公开了一种动作的计数方法、装置、设备及存储介质，该方法应用于可穿戴设备，可穿戴设备包括加速度传感器，该方法包括：获取目标运动特征信息；根据当前加速度信号，确定用户的当前运动特征信息；基于当前运动特征信息和目标运动特征信息，确定用户的动作次数累计值。考虑到相关技术中对动作的计数往往是针对既定的有限类型的动作类型，无法对新的动作类型甚至用户自创动作做计数，且计数方法严格依赖于人为设定的信号特征，不同的用户动作标准与否，容易引入干扰而导致既定规则失效；针对此问题，本技术方案不对动作执行做约束和假设，通过对接收到的初始加速度信号进行处理，获得表征动作的目标运动特征信息，在目标运动特征信息确定后，将当前运动特征信息和目标运动特征信息匹配，确定动作次数累计值，提高了动作 计数的灵活性。
实施例二
图2为本申请实施例二提供的另一种动作的计数方法的流程示意图，本实施例可适用于对无既定设定动作规则的动作进行计数的情况。该方法可以由动作的计数装置来执行，该装置可由硬件和/或软件组成，并一般可集成在可穿戴设备中，可穿戴设备包括加速度传感器。
如图2所示，本申请实施例二提供的另一种动作的计数方法，包括如下操作：
S210、对起始加速度信号进行多尺度变换分析，确定目标峰值位置序列。
本实施例中，可穿戴设备包含三轴加速度传感器，当用户开启可穿戴设备并进行运动后，会产运动数据并存储在数据缓存区中，从数据缓存区中以滑动窗口的方式提取起始加速度信号，起始加速度信号可以理解为用户运动最先产生的数据，也可以理解为数据缓存区中最先存储的数据，该信号可以作为获取目标运动特征信息的基础数据。本实施例中仅对三轴对应的起始加速度信号进行分析，为了提高对复合动作和复杂动作的计数检出率，采用在不同粒度上对动作信号的重复性做分析。在仅获取三轴的起始加速度信号的条件下，应用多尺度变换，对不同尺度下的变换信号做峰特征检测，通过设定规则选取响应最大对应的某个尺度和某个坐标轴下的峰序列，作为目标峰值位置序列。
示例性的，对三轴对应的起始加速度信号做多尺度变换分析，在三个轴和多个尺度下，得到多个小波变换后的输出信号，通过对输出信号做峰提取，并实施提出的峰筛选策略，有效提取最关键和明显的运动区间分割点，即目标峰值位置序列。本步骤仅需要在指定的尺度上进行卷积计算，在一定的计算复杂度和存储开销的情况下，能够完成对常见力量训练动作频率的场景覆盖。
S220、根据目标峰值位置序列，确定目标运动特征信息。
根据目标峰值位置序列，得到高质量的区间信号划分；考虑相邻区间信号 的相似性和稳定性，可以计算相邻区间信号的置信度；对于连续稳定的高置信度区间信号提取运动特征；对连续设定个数的区间信号置信度高于某设定值时，可以聚类当前的运动特征，形成目标运动特征信息。
可以清楚的是，执行步骤S210-S220后，基于目标运动特征信息，可以将用户继续进行的健身动作与该目标运动特征信息进行匹配，以进行健身动作计数。
S230、从当前加速度信号中提取目标轴的目标加速度信号。
其中，目标轴为目标运动特征信息关联的坐标轴。需要说明的是，根据步骤S220确定目标运动特征信息后，同时也确定出目标运动特征信息关联的目标轴。将目标轴作为要跟踪的加速度坐标轴后，后续对健身动作的计数过程，只需要在已经选定的目标轴上做运动数据分析即可。
示例性的，在确定目标运动特征信息后，从数据缓存区中以滑动窗口的方式获取当前加速度信号，从当前加速度信号中提取目标轴对应的加速度信号作为目标加速度信号。
S240、基于目标加速度信号，确定用户的当前运动特征信息。
需要知道的是，确定用户的当前运动特征信息的步骤与确定目标运动特征信息的步骤是相同的，区别仅在于目标运动特征信息是基于三轴加速度信号进行确定的，而当前运动特征信息是基于单个目标轴加速度信号进行确定的。
S250、将当前运动特征信息进行划分，获得第一当前运动特征向量和第二当前运动特征向量。
本步骤中，从动作宏观形态和动作微观细节因素同时进行考量，将当前运动特征信息与目标运动特征信息对应的设定分数对应特征信息做重叠面积计算，将当前运动特征信息与目标运动特征信息对应的近似编码序列对应特征信息做互相关系数计算。
示例性的，当前运动特征信息包含设定分数对应的特征信息及近似编码序列对应的特征信息。对当前运动特征信息进行划分，将设定分数对应的特征信 息确定为第一当前运动特征向量，将近似编码序列对应的特征信息确定为第二当前运动特征向量。示例性的，当前运动特征信息表示为：{frc 90,frc 10,s 1,s 2,,s 3,,s 4,,s 5,,s 6,,s 7,,s 8}，对其进行划分后，则确定第一当前运动特征向量为{frc 90,frc 10}，确定第二当前运动特征向量为{s 1,,s 2,,s 3,,s 4,,s 5,,s 6,,s 7,,s 8}。
S260、将目标运动特征信息进行划分，获得第一目标运动特征向量和第二目标运动特征向量。
其中，目标运动特征信息包含设定分数对应的特征信息及近似编码序列对应的特征信息，对目标运动特征信息进行划分，将设定分数对应的特征信息确定为第一目标运动特征向量，将近似编码序列对应的特征信息确定为第二目标运动特征向量。示例性的，目标运动特征信息表示为：
Figure PCTCN2022082721-appb-000001
对其进行划分后，则确定第一目标运动特征向量为
Figure PCTCN2022082721-appb-000002
确定第二目标运动特征向量为
Figure PCTCN2022082721-appb-000003
S270、根据第一当前运动特征向量、第二当前运动特征向量、第一目标运动特征向量以及第二目标运动特征向量，确定用户健身的动作次数累计值。
本步骤采用分级匹配的策略，将计数问题转化为模式匹配问题。首先，根据第一当前运动特征向量和第一目标运动特征向量进行较为简单的重叠面积比例计算；然后，根据第二当前运动特征向量和第二目标运动特征向量进行近似编码序列的互相关系数计算；最后根据融合置信度(即融合动作时长稳定性和动作形态一致性的置信度)、重叠面积比例、互相关系数，得到动作匹配概率，以概率的形式实现对健身动作计数的量化统计。可以知道的是，重叠面积比例和互相关系数，分别作为动作宏观形态和动作微观细节因素进行考量，并将匹配结果以概率输出的形式，将计数问题，转化为模式匹配问题，以量化概率的形式实现对动作计数的统计，提高了计数的准确率。
作为本申请实施例的一个可选实施例，在上述实施例的基础上，对起始加 速度信号进行多尺度变换分析，确定目标峰值位置序列的步骤为：
S2101、基于起始加速度信号，确定起始加速度信号对应多个设定尺度的小波序列。
对滤波后的起始加速度信号中每一个轴对应的加速度信号，进行多尺度变换。其中多尺度变换可以采用小波基为哈尔(Harr)小波的小波变换。设定尺度的选取将影响日常力量训练动作的频率分布区间，为了最大限度地在日常健身动作频率上获得明显的响应输出，设定尺度的确定可以依据一定规模的真实用户数据集进行统计获得。可选的，分别选定对应0.4、1.0、1.5、2.0、2.5Hz的5个尺度，可以最大限度地在日常健身动作频率上获得明显的响应输出。示例性地，确定小波基为Harr小波，尺度频率为0.4、1.0、1.5、2.0、2.5Hz，采样频率为25Hz，即可以得到相应的小波基序列，序列的长度分别为25、40、60、120、250。对加速度的三轴加速度信号分别做上述5个尺度的多尺度变换，以下将以x轴加速度在第4个尺度上的小波变换为实例，说明如何得到量化的响应输出。
示例性的，在x轴加速度信号序列{x 0,x 1,x 2,…,x N}和第4个尺度上的小波基序列
Figure PCTCN2022082721-appb-000004
做卷积，得到长度为max{N,M}的小波变换后的序列。因为采样率为25Hz且窗口长度为15秒，N＝375,M＝120，因此，最终得到的变换后小波序列为{y 0,y 1,y 2,…,y N}。
S2102、根据小波序列，确定起始加速度信号对应的多个峰值位置序列以及多个峰值高度序列。
本步骤中，接S2101的示例进行描述，对{y 0,y 1,y 2,…,y N}做局部极大值提取，即得到了序列信号的峰提取。峰提取的实施需要确保峰值左侧和右侧的信号均小于峰值点的信号大小，并且峰与峰之间满足某种约束关系，这种约束关系包括相邻峰的间距不得大于某个较大值，不得小于某个较小值，每个候选的峰值点的信号大小不应明显小于所有候选峰值点信号大小的中值的预设比例。在执行上述对峰提取的约束规则后，得到了序列信号{y 0,y 1,y 2,…,y N}的峰值位置序列{l 0,l 1,l 2,…,l P}和峰值高度序列{v 0,v 1,v 2,…,v P}。
S2103、基于多个峰值高度序列，确定起始加速度对应的多个总体响应输出值。
本步骤中，继续接S2102的示例进行描述，对峰值高度序列{v 0,v 1,v 2,…,v P}计算中值，作为这个轴输入信号{x 0,x 1,x 2,…,x N}在这个尺度
Figure PCTCN2022082721-appb-000005
下的小波变换分析后{y 0,y 1,y 2,…,y N}的响应输出
Figure PCTCN2022082721-appb-000006
对x轴的加速度信号都执行全部5个尺度的小波变换分析，得到5个响应输出
Figure PCTCN2022082721-appb-000007
找出其中最大的响应输出对应的峰值位置序列
Figure PCTCN2022082721-appb-000008
假设第i个尺度上的响应输出
Figure PCTCN2022082721-appb-000009
最大。将上述5个响应输出
Figure PCTCN2022082721-appb-000010
的均值作为x轴的总体响应输出r x。
依照上述对x轴加速度信号的处理，完成对y轴和z轴的处理，此时算法中应至少保存有3个轴对应的各自峰值位置序列和3个轴对应的总体响应输出{r x,r y,r z}。
S2104、根据多个总体响应输出值，确定目标峰值位置序列。
示例性的，选取{r x,r y,r z}中最大值对应的轴指向的峰值位置序列
Figure PCTCN2022082721-appb-000011
作为下一步骤区间信号划分的基础，假设在选定总体响应输出最大的轴上具备最大响应输出的尺度对应的峰值位置序列(即目标峰值位置序列)发生在第j个尺度上。
作为本申请实施例的一个可选实施例，在上述实施例的基础上，根据目标峰值位置序列，确定目标运动特征信息，包括：
S2201、根据目标峰值位置序列，结合预设合并规则，生成起始加速度信号的多个区间信号集，每个区间信号集由两个或两个以上区间信号组成。
示例性的，继续接上述示例进行描述，假设在选定总体响应输出最大的轴上具备最大响应输出的尺度对应的峰值位置序列发生在第j个尺度上。将上述步骤S2104经过筛选得到的目标峰值位置序列合并到当前的区间信号划分位置序列中，合并操作须符合如下规则：待合并的目标峰值位置序列与当前的区间信号划分位置序列执行“集合并”操作，去重；合并后的区间信号划分位置序列 的长度必须大于或等于4；合并后的区间信号划分位置序列应偏向均匀地划分总长度的信号。即相邻的3个区间信号，其各自的长度为l 1,l 2和l 3，应确保l 2不明显小于l 1和l 3，可选的，该步骤中选用的规则为l 2大于或等于1/2的l 1且大于或等于1/2的l 3。
经过上述对信号的区间划分，可以确定三个轴加速度信号的区间信号集合L x,＝L y和L z。以L x为例，L x包括多个
Figure PCTCN2022082721-appb-000012
是由
Figure PCTCN2022082721-appb-000013
组成，其中
Figure PCTCN2022082721-appb-000014
和
Figure PCTCN2022082721-appb-000015
为相邻的2个经过筛选符合约束规则的峰值位置序列中的元素。
S2202、分别计算每个区间信号集中相邻区间信号的第一置信度以及连续设定个数区间信号的第二置信度。
其中，置信度主要用于衡量动作执行的周期性和稳定性，考虑到日常力量训练动作的执行过程中，单组的相邻两次动作具有运动快慢的一致性，因此用于评估相邻区间信号稳定的第一置信度是基于相邻区间信号的时间长度实现的。第一置信度可以理解为表征相邻区间信号的相似性和稳定性。
当若干相邻区间信号具有较高的第一置信度时，即说明这些相邻的区间信号对应了某个动作的多次重复。因此，可以计算连续设定个数区间信号的第一置信度之和，作为第二置信度。当第二置信度高于设定阈值时，则可以基于该区间信号做运动特征的提取计算。
可选的，分别计算区间信号集中相邻区间信号的第一置信度以及连续设定个数区间信号的第二置信度，包括：
a1)若区间信号集中相邻区间信号的长度不一致，则对其中较短的区间信号进行插值重采样，使相邻区间信号长度保持一致。
考虑到日常力量训练动作的执行过程中，单组的相邻两次动作具有运动快慢的一致性，因此该方法中采用的用于评估相邻区间信号稳定的置信度是基于相邻区间信号的时间长度实现的。即假设相邻区间信号{X a,X b}中持续时间分别为l a和l b，比值
Figure PCTCN2022082721-appb-000016
不应小于某个阈值z 1，并且当这个比值大于某个阈值z 2时， 这个比值被强制修改为1。这样设计确保在相邻两个动作执行周期相差不大的情况下，时间长度的差异不影响置信度的大小。对于2个相邻区间信号{X a,X b}(假设持续时间l a＜l b)，一般存在信号长度差异。本步骤采用对信号长度较短的序列做插值重采样，拓展其长度与长度较长的序列一致。例如，对X a插值重采样后的信号
Figure PCTCN2022082721-appb-000017
b1)分别确定每个区间信号集中相邻区间信号的持续时长。
示例性的，假设确定相邻区间信号{X a,X b}中持续时间分别为l a和l b。
c1)确定每个区间信号集中相邻区间信号的互相关系数。
在长度一致的情况下，对插值重采样后的信号
Figure PCTCN2022082721-appb-000018
与另一个信号X b做互相关计算，得到互相关系数R ab。在互相关系数小于0.5的情况下，R ab被强制修改为0，本步骤互相关系数以0.5为界限是基于大量真实用户数据统计得到的，数据表明，互相关系数小于0.5认为不明显相关。
d1)基于相邻区间信号持续时长及互相关系数，确定相邻区间信号的第一置信度以及连续设定个数区间信号的第二置信度。
示例性的，综合考虑相邻区间信号的相似性和稳定性的第一置信度P conf被定义为：
Figure PCTCN2022082721-appb-000019
由于区间信号的划分是在滑动分窗后提取的，因此某个区间信号可能在当前滑动窗口的中部，且在上一个滑动窗口的尾部，这使得同一个区间信号在不同的滑动窗口内可能有不同的置信度，本实施例中置信度主要用于衡量动作执行的周期性和稳定性，因此始终选择这个区间信号截至到当前最大的置信度作为后续步骤的输入。其中，设定个数可以是按需求设定，例如可以取值为4个，表征健身动作重复了4次。
S2203、当第一置信度满足第一置信度条件时，确定起始加速度信号对应的起始运动特征信息。
其中，第一置信度可以理解为表征相邻区间信号的相似性和稳定性。第一置信度条件可以是置信度高于某设定阈值。当第一置信度高于某设定阈值时，则可以每段区间信号的设定分位数及区间信号的近似编码序列表征起始运动特 征信息。
可选的，确定起始加速度信号对应的起始运动特征信息，包括：
a2)对区间信号的设定分位数范围内进行等分处理，获取分割点匹配序列。
可以清楚的是，当若干相邻区间信号具有较高的第一置信度时，即说明这些相邻的区间信号对应了某个动作的多次重复。则可以对这些相邻区间信号做运动特征的提取计算。可选的，计算每段区间信号的10分位数，90分位数和区间信号的近似编码序列。
区间信号的近似编码序列的实现方式是对区间信号的设定分位数范围内进行等分处理。示例性的，对区间信号的90分位数和10分位数对应的范围等分为6份，以90分位数，10分位数和5个中间分割点作为信号的波动范围分割点，构成长度为8的区间序列{g 0,g 1,…,g 7}，在时间维度上将该区间信号等分为8段，以每段信号的均值作为基准查找最近的可以落在区间{g 0,g 1,…,g 7}中的点，如此完成全部8段信号的分割点匹配。示例性的，错误！未找到引用源。为本申请实施例提供的分割点匹配序列确定过程的示例图。分割点匹配序列为{g 4,g 2,g 1,g 4,g 7,g 6,g 5,g 5}。
b2)基于分割点匹配序列，确定近似编码序列。
示例性的，将分割点匹配序列的分割点索引作为近似编码序列保存。示例性的，继续接上述示例，假设分割点匹配序列为{g 4,g 2,g 1,g 4,g 7,g 6,g 5,g 5}，则将分割点匹配序列的分割点索引{4,2,1,4,7,6,5,5}作为近似编码序列保存。
本步骤中区间信号的近似编码序列在不考虑信号的绝对大小和快慢的前提下，用尽量少和简单的特征描述了信号波动在形态上的特点，从工程实现的角度，也极大减少了对存储的开销和降低计算匹配的复杂度。
c2)根据设定分位数与近似编码序列，确定起始加速度信号对应的起始运动特征信息。
本步骤中，设定分位数与近似编码序列，可以从区间信号的绝对大小和波动形态上对动作进行特异性描述。因此将设定分位数与近似编码序列，确定为 起始加速度信号对应的起始运动特征信息。示例性的，综合上述步骤的90分位数，10分位数和近似编码序列，确定为起始加速度信号对应的起始运动特征信息。
S2204、当第二置信度满足第二置信度条件时，根据起始运动特征信息，确定目标运动特征信息及目标轴。
本步骤中目标运动特征信息，是基于区间信号提取的运动特征聚类实现的，自动获取最近稳定且重复动作的关键信息，存储为目标运动特征信息做后续的匹配，目标运动特征信息的确定在不采用复杂的聚类操作的前提下，保留了不同动作之间的区分度。其中，第二置信度条件可以设置为第二置信度高于某个设定阈值。
示例性的，假设设定连续设定个数取值为4，第二置信度条件为大于3.8，,当且仅当连续的4个区间信号的置信度
Figure PCTCN2022082721-appb-000020
总和大于3.8的条件下，则聚类当前的运动特征向量，形成目标运动特征信息，用于后续步骤的动作匹配。
可选的，根据起始运动特征信息，确定目标运动特征信息及目标轴，包括：
a3)根据起始运动特征信息，确定连续设定个数区间信号的起始运动特征参数。
其中，起始运动特征参数可以理解为起始运动特征信息对应的均值和方差。继续接步骤S2204示例进行描述，对上述符合要求的4个区间信号，分别计算对应的运动特征向量{frc 90,frc 10,s 1,s 2,,s 3,,s 4,,s 5,,s 6,,s 7,,s 8}，即90分位数，10分位数和近似编码序列(s i对应{g 0,g 1,…,g 7}中的某一个)，计算各自对应的均值和方差。
b3)若起始运动特征参数满足设定条件，则聚类起始运动特征信息，获得目标运动特征信息及目标轴。
示例性的，判断起始运动特征参数是否满足设定条件，可选的，设定条件可以设置为起始运动特征信息是否落在以均值为中心，3倍标准差为参考的95％ 高概率范围内。当4个区间信号的运动特征均落在上述高概率范围内时，才会形成这个动作的目标运动特征信息，并标记当前所采用的坐标轴作为后续运动匹配跟踪的目标轴，即确定此后的力量训练计数跟踪x，y和z轴中的哪一个。
此时，保存上述90分位数，10分位数和近似编码序列对应的均值，构成10维度的向量，这个向量被存储为当前这个连续稳定且重复至少4次的动作对应的目标运动特征向量fv fingerprint。目标运动特征向量可以表示为：
Figure PCTCN2022082721-appb-000021
可选的，需要说明，在对起始加速度信号进行多尺度变换分析之前，可以从预设的数据缓存区中获取起始滑动窗口内的起始加速度信号，起始加速度信号经过滤波处理。
其中，预设的缓存区可以理解为用于存储用户进行运动时产生的运动数据。考虑到本实施例提供的技术方案应用的广泛性，对于存储和计算速度有限的可穿戴设备(如智能手表/手环)中能够应用该方法，由于原始加速度信号的采样率将影响是否能真实还原常见快速/慢速的力量训练动作的关键信息，本实施例在兼顾功耗和性能的前提下，需设置合理的加速度信号的采样率。可选的，基于在一定规模的真实用户数据集上的统计，加速度信号的采样率可以为25Hz，即每秒钟获得25个加速度x轴采样值，25个加速度y轴采样值，25个z轴采样值。
需要知道是的，可以以数据分窗的形式从数据缓存区中获取起始加速度信号。考虑到日常从事力量训练动作中单组单个动作的持续时间和单组训练中某个动作可能重复的次数，对起始加速度信号的数据分窗大小将影响是否能够在窗口内观察到足够多且连续的动作信息，因此，数据分窗长度的大小是很重要的。可选的，可以基于在一定规模的真实用户数据集上的统计，确定数据分窗长度为15秒，即对三轴起始加速度信号做缓存，缓存长度对应连续15秒的动作时长。另外，由于数据分窗的引入，为了确保在单个窗口内尽可能观察到完整的动作信号，需要采用合理的滑动更新的规则。可选的，可以采用1/2滑动更 新的规则。图4为本申请实施例提供的滑动更新的示例图，如图4所示，滑动分窗依次为t 1、t 2和t 3。
在获取到足够填充满数据分窗大小的加速度原始信号的条件下，为了避免信号的测量噪声和动作抖动引起干扰，需要对3轴原始加速度信号做滤波处理。可选的，分别对3轴原始加速度信号做截止频率大于或等于2.5Hz的低通滤波，以去除信号的测量噪声和动作抖动引起的干扰。低通滤波方法的选取将影响到滤波收敛的速度、信号波形的畸变和存储计算开销，可以选用阶数为2的巴特沃滋低通滤波，并在滤波执行过程中舍去前8个滤波后的数据，以屏蔽滤波收敛过程中信号波形的抖动引起的误计数。
可选的，根据第一当前运动特征向量、第二当前运动特征向量、第一目标运动特征向量以及第二目标运动特征向量，确定用户的动作次数累计值，包括：
a)基于第一当前运动特征向量与第一目标运动特征向量，确定重叠比例。
示例性的，将第一当前运动特征向量与第一目标运动特征向量进行重叠面积检查，确定两者的重叠比例P overlap。图5为本申请实施例提供的确定重叠比例的示例图。
b)判定重叠比例是否小于第一设定阈值，在重叠比例大于或等于第一设定阈值的情况下，执行步骤c)；在重叠比例小于第一设定阈值的情况下，执行步骤h)。
其中，第一设定阈值可以是基于在一定规模的真实用户数据集统计获得的，也可以是设备开发人员设定的。比较重叠比例与第一设定阈值，若重叠比例小于第一设定阈值，表征当前运动特征信息与目标运动特征信息不匹配，则控制用户健身的动作次数累计值不变。若重叠比例大于或等于第一设定阈值，则可以进一步确定第二当前运动特征向量与第二目标运动特征向量的互相关系数。
c)基于第二当前运动特征向量与第二目标运动特征向量，确定互相关系数。
本步骤是在确定重叠比例大于或等于第一设定阈值之后进行的，对第二当前运动特征向量与第二目标运动特征向量做互相关操作，计算互相关系数R s。
d)判定互相关系数是否小于第二设定阈值，在互相关系数大于或等于第二设定阈值的情况下，执行步骤e)；在互相关系数小于第二设定阈值的情况下，执行步骤h)。
其中，第二设定阈值可以是基于在一定规模的真实用户数据集统计获得的，也可以是设备开发人员设定的。比较互相关系数与第二设定阈值，若互相关系数小于第二设定阈值，表征当前运动特征信息与目标运动特征信息不匹配，则控制用户健身的动作次数累计值不变；若互相关系数大于或等于第二设定阈值，则可以进一步根据置信度、重叠比例以及互相关系数，确定估计概率。
e)根据置信度、重叠比例以及互相关系数，确定估计概率，置信度由相邻区间信号确定。
本步骤是在确定重叠比例大于或等于第一设定阈值，且互相关系数大于或等于第二设定阈值之后进行的，对于符合上述阈值判断的区间信号可以初步认为与目标动作特征信息可能匹配。需要说明的是，对于符合上述阈值判断的区间信号还需要计算估计概率，估计概率是d1)步骤中对应的置信度P conf与上述计算的重叠比例P overlap和互相关系数R s的乘积。估计概率可以表示为：P estimate＝P conf·P overlap·R s。需要说明的是步骤e)中的置信度和上文提到的融合置信度、第一置信度含义相同。
f)判定估计概率是否大于第三设定阈值，在估计概率大于第三设定阈值的情况下，执行步骤g)；在估计概率小于或等于第三设定阈值的情况下，执行步骤h)。
当且仅当这个估计概率大于第三设定阈值时，才判定这个区间信号对应的动作与最近保存的目标运动特征信息匹配。
g)控制用户的动作次数累计值加1作为新的动作次数累计值。
当区间信号对应的动作与最近保存的目标运动特征信息相匹配时，控制用户健身的动作次数累计值加1作为新的动作次数累计值。
h)控制用户的动作次数累计值不变。
当区间信号对应的动作与最近保存的目标运动特征信息不匹配时，控制用户健身的动作次数累计值不变，且不对不匹配动作形成新的目标运动特征信息。需要清楚的是，当用户每次开启可穿戴设备后，利用本技术方案可以形成对应当次目标运动特征信息，也可以理解为每次开启可穿戴设备后，只形成一种目标运动特征信息。为了更清楚的表述，以某一应用场景为例进行说明，示例性的，用户甲到达健身房后欲进行自创动作的运动，开启自身佩戴的可穿戴设备后，用户甲开始运动，可穿戴设备通过接收到的用户运动信息，利用本方案提供的技术手段，形成目标运动特征信息，并进行健身动作计数。期间一段时间，用户甲走步去休息室休息，后返回运动区继续进行运动，结束运动时关闭可穿戴设备。从可穿戴设备开启至关闭过程，只会形成一次目标运动特征信息，其中走步去休息室的动作不与目标运动特征信息匹配，不会对走步动作进行健身动作次数累加，并且也不会根据走步动作形成新的目标动作特征信息。
可选地，在确定用户的动作次数累计值之后，还包括：如果当前不满足计数循环结束条件，则将下一滑动窗口的加速度信号作为当前加速度信号，返回重新执行根据当前加速度信号确定用户的当前运动特征信息的操作；在当前满足计数循环结束条件的情况下，将动作次数累计值作为动作的计数结果进行反馈。
其中，计数循环结束条件为：加速度信号在设定时间内未被获取到；或者，动作次数累计值连续设定次(即指以滑动窗口的形式获取下一窗口的加速度信号的连续次数达到设定次数)未发生变化。本实施例中，设定时间可由用户设定，加速度信号在设定时间内未被获取到可以理解为用户在设定时间内停止运动未产生新的加速度信号，也可以理解为在设定时间内数据缓存区中无新的运动数据供可穿戴设备分析，则可以终止计数。设定次可由用户设定，动作次数累计值连续设定次未发生变化可以理解为用户不再进行健身动作的运动，而是做其他动作或者停止动作，即动作次数累计值连续设定次未发生变化，则可以终止计数。在终止计数后，需将动作次数累计值作为健身动作的计数结果进行 反馈。示例性的，可以是在可穿戴设备上以数字显示或语音播报的方式进行呈现。
示例性的，如果当前不满足计数循环结束条件，则继续获取缓存区中的数据，以滑动窗口的形式获取下一窗口的加速度信号作为当前加速度信号，返回重新执行根据当前加速度信号确定用户的当前运动特征信息，将当前运动特征信息与已获取的目标运动特征信息进行匹配，进行计数统计。可以清楚的是，如果当前不满足计数循环结束条件，则继续获取再下一滑动窗口的加速度信号进行后续步骤，如此循环执行。
本申请实施例细化了目标运动特征信息的确定过程，从数据缓存区中以滑动窗口的方式提取起始加速度信号并进行滤波处理；仅对三轴加速度信号做多尺度变换分析，有效提取最关键和明显的运动区间分割点，即峰值位置信息，得到高质量的区间信号划分；对相邻区间信号做融合动作时长稳定性和动作形态一致性的置信度计算，将前后两个动作周期的相似性和稳定性量化为概率模型，减少对人为设定的规则的依赖；在进行运动特征提取时既考虑了运动信号的绝对特点，即加速度信号大小的分位数，也考虑了运动信号的相对特点，即加速度信号的波动形态描述，在不明显增加计算复杂度和存储开销的情况下，适合智能手表/手环等资源受限设备；自动获取最近稳定且重复动作的关键信息，存储为目标运动特征信息做后续的匹配，不采用复杂的聚类操作的前提下，保留了不同动作之间的区分度。
另外，在确定目标运动特征信息后，进行动作匹配时只考虑目标轴对应特征信息是否匹配，提高了计算效率，节省了空间。同时，采用分级匹配的策略，由计算较为简单的重叠面积比例计算，到近似编码序列的互相关系数计算，融合置信度、动作宏观形态和动作微观细节因素，得到了动作匹配概率(即估计概率)，以概率的形式实现对力量训练计数的量化统计。
实施例三
图6为本申请实施例三提供的一种动作的计数装置的结构框图，该装置可应用于可穿戴设备，如图6所示，该装置包括：目标信息获取模块31、当前信息确定模块32和动作次数确定模块33。
目标信息获取模块31，设置为获取目标运动特征信息；
当前信息确定模块32，设置为根据当前加速度信号，确定用户的当前运动特征信息；
动作次数确定模块33，设置为基于当前运动特征信息和目标运动特征信息，确定用户的动作次数累计值。
可选地，目标信息获取模块31可以包括：
目标序列确定单元，设置为对起始加速度信号进行多尺度变换分析，确定目标峰值位置序列；
目标信息确定单元，设置为根据目标峰值位置序列，确定目标运动特征信息。
可选地，目标序列确定单元设置为：
基于起始加速度信号，确定起始加速度信号对应多个设定尺度的小波序列；
根据小波序列，确定起始加速度信号对应的多个峰值位置序列以及多个峰值高度序列；
基于多个峰值高度序列，确定起始加速度对应的多个总体响应输出值；
根据多个总体响应输出值，确定目标峰值位置序列。
可选地，目标信息确定单元设置为：
根据目标峰值位置序列，结合预设合并规则，生成起始加速度信号的多个区间信号集，每个区间信号集由两个或两个以上区间信号组成；
分别计算每个区间信号集中相邻区间信号的第一置信度以及连续设定个数区间信号的第二置信度；
当第一置信度满足第一置信度条件时，确定起始加速度信号对应的起始运动特征信息；
当第二置信度满足第二置信度条件时，根据起始运动特征信息，确定目标运动特征信息及目标轴。
可选地，目标信息确定单元设置为分别计算区间信号集中各相邻区间信号的第一置信度以及连续设定个数区间信号的第二置信度的步骤可以包括：
若区间信号集中相邻区间信号的长度不一致，则对其中较短的区间信号进行插值重采样，使相邻区间信号长度保持一致；
分别确定每个区间信号集中相邻区间信号的持续时长；
确定每个区间信号集中相邻区间信号的自相关系数；
基于相邻区间信号持续时长及自相关系数，确定相邻区间信号的第一置信度以及连续设定个数区间信号的第二置信度。
可选地，目标信息确定单元设置为确定起始加速度信号对应的起始运动特征信息的步骤可以包括：
对区间信号的设定分位数范围内进行等分处理，获取分割点匹配序列；
基于分割点匹配序列，确定近似编码序列；
根据设定分位数与近似编码序列，确定起始加速度信号对应的起始运动特征信息。
可选地，目标信息确定单元设置为根据起始运动特征信息，确定目标运动特征信息及目标轴的步骤可以包括：
根据起始运动特征信息，确定连续设定个数区间信号的起始运动特征参数；
若起始运动特征参数满足设定条件，则聚类起始运动特征信息，获得目标运动特征信息及目标轴。
可选的，当前信息确定模块32，可以包括：
目标加速度信号确定单元，设置为从当前加速度信号中提取目标轴的目标加速度信号，其中，目标轴为目标运动特征信息关联的坐标轴；
当前运动特征信息确定单元，设置为基于目标加速度信号，确定用户的当前运动特征信息。
可选的，动作次数确定模块33设置为：
将当前运动特征信息进行划分，获得第一当前运动特征向量和第二当前运动特征向量；
将目标运动特征信息进行划分，获得第一目标运动特征向量和第二目标运动特征向量；
根据第一当前运动特征向量、第二当前运动特征向量、第一目标运动特征向量以及第二目标运动特征向量，确定用户的动作次数累计值。
可选地，动作次数确定模块33设置为根据第一当前运动特征向量、第二当前运动特征向量、第一目标运动特征向量以及第二目标运动特征向量，确定用户的动作次数累计值的步骤，包括：
a)基于所述第一当前运动特征向量与所述第一目标运动特征向量，确定重叠比例。
b)判定所述重叠比例是否小于第一设定阈值，响应于所述重叠比例大于或等于第一设定阈值，执行步骤c)；响应于所述重叠比例小于第一设定阈值，执行步骤h)。
c)基于所述第二当前运动特征向量与所述第二目标运动特征向量，确定互相关系数。
d)判定所述互相关系数是否小于第二设定阈值，响应于所述互相关系数大于或等于第二设定阈值，执行步骤e)；响应于所述互相关系数小于第二设定阈值，执行步骤h)。
e)根据置信度、所述重叠比例以及所述互相关系数，确定估计概率，所述置信度由相邻区间信号确定。
f)判定所述估计概率是否大于第三设定阈值，响应于所述估计概率大于第三设定阈值，执行步骤g)；响应于所述估计概率小于或等于第三设定阈值，则执行步骤h)。
g)控制用户健身的动作次数累计值加1作为新的动作次数累计值。
h)控制用户健身的动作次数累计值不变。
可选地，该装置还包括循环模块：
循环模块，设置为如果当前不满足计数循环结束条件，则将下一滑动窗口的加速度信号作为当前加速度信号，返回重新执行根据当前加速度信号确定用户的当前运动特征信息的操作；如果当前满足计数循环结束条件，将动作次数累计值作为动作的计数结果进行反馈。
可选地，计数循环结束条件为：加速度信号在设定时间内未被获取到；或者，动作次数累计值连续设定次未发生变化。
上述装置可执行本申请前述所有实施例所提供的动作的计数方法，具备执行上述方法相应的功能模块。未在本实施例中详尽描述的技术细节，可参见本申请前述所有实施例所提供的方法。
实施例四
图7为本申请实施例四提供的一种可穿戴设备的结构框图，如图7所示，该可穿戴设备包括处理器41、存储器42、输入装置43和输出装置44；计算机设备中处理器41的数量可以是至少一个，图7中以一个处理器41为例；可穿戴设备中的处理器41、存储器42、输入装置43和输出装置44可以通过总线或其他方式连接，图7中以通过总线连接为例。
存储器42作为一种计算机可读存储介质，可设置为存储软件程序、计算机可执行程序以及模块，如本申请实施例中的动作的计数方法对应的模块(例如，动作的计数装置中的目标信息获取模块31、当前信息确定模块32和动作次数确定模块33)。处理器41通过运行存储在存储器42中的软件程序、指令以及模块，从而执行计算机设备的各种功能应用以及数据处理，即实现上述的动作的计数方法。
存储器42可主要包括存储程序区和存储数据区，其中，存储程序区可存储操作系统、至少一个功能所需的应用程序；存储数据区可存储根据终端的使用 所创建的数据等。此外，存储器42可以包括高速随机存取存储器，还可以包括非易失性存储器，例如至少一个磁盘存储器件、闪存器件、或其他非易失性固态存储器件。在一些实例中，存储器42可包括相对于处理器41远程设置的存储器，这些远程存储器可以通过网络连接至计算机设备。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
输入装置43可设置为接收输入的数字或字符信息，以及产生与计算机设备的用户设置以及功能控制有关的键信号输入。输出装置44可包括显示屏等显示设备。
实施例五
本申请实施例五还提供一种包含计算机可执行指令的存储介质，所述计算机可执行指令在由计算机处理器执行时用于执行一种动作的计数方法，该方法包括：
获取目标运动特征信息；
根据当前加速度信号，确定用户的当前运动特征信息；
基于所述当前运动特征信息和所述目标运动特征信息，确定用户的动作次数累计值。
当然，本申请实施例所提供的一种包含计算机可执行指令的存储介质，其计算机可执行指令不限于如上所述的方法操作，还可以执行本申请任意实施例所提供的动作的计数方法中的相关操作。
通过以上关于实施方式的描述，所属领域的技术人员可以清楚地了解到，本申请可借助软件及必需的通用硬件来实现，当然也可以通过硬件实现。基于这样的理解，本申请的技术方案本质上或者说对相关技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品可以存储在计算机可读存储介质中，计算机可读存储介质可以为非暂态存储介质，如计算机的软盘、只读存储器(Read-Only Memory,ROM)、随机存取存储器(Random Access Memory, RAM)、闪存(FLASH)、硬盘或光盘等，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施例所述的方法。
值得注意的是，上述动作的计数装置的实施例中，所包括的各个单元和模块只是按照功能逻辑进行划分的，但并不局限于上述的划分，只要能够实现相应的功能即可；另外，各功能单元的名称也只是为了便于相互区分，并不用于限制本申请的保护范围。

Claims (21)

  1. 一种动作的计数方法，应用于可穿戴设备，所述可穿戴设备包括加速度传感器，包括：
    获取目标运动特征信息；
    根据当前加速度信号，确定用户的当前运动特征信息；
    基于所述当前运动特征信息和所述目标运动特征信息，确定用户的动作次数累计值。
  2. 根据权利要求1所述的方法，其中，所述获取目标运动特征信息，包括：
    对起始加速度信号进行多尺度变换分析，确定目标峰值位置序列；
    根据所述目标峰值位置序列，确定目标运动特征信息。
  3. 根据权利要求2所述的方法，其中，所述对起始加速度信号进行多尺度变换分析，确定目标峰值位置序列，包括：
    基于所述起始加速度信号，确定所述起始加速度信号对应多个设定尺度的小波序列；
    根据所述小波序列，确定所述起始加速度信号对应的多个峰值位置序列以及多个峰值高度序列；
    基于所述多个峰值高度序列，确定起始加速度对应的多个总体响应输出值；
    根据所述多个总体响应输出值，确定目标峰值位置序列。
  4. 根据权利要求2所述的方法，其中，所述根据所述目标峰值位置序列，确定目标运动特征信息，包括：
    根据所述目标峰值位置序列，结合预设合并规则，生成所述起始加速度信号的多个区间信号集，每个所述区间信号集由两个或两个以上区间信号组成；
    分别计算每个所述区间信号集中相邻所述区间信号的第一置信度以及连续设定个数所述区间信号的第二置信度；
    响应于所述第一置信度满足第一置信度条件，确定所述起始加速度信号对应的起始运动特征信息；
    响应于所述第二置信度满足第二置信度条件，根据所述起始运动特征信息， 确定所述目标运动特征信息及目标轴。
  5. 根据权利要求4所述的方法，其中，所述分别计算每个所述区间信号集中相邻所述区间信号的第一置信度以及连续设定个数所述区间信号的第二置信度，包括：
    响应于所述区间信号集中相邻所述区间信号的长度不一致，对其中较短的区间信号进行插值重采样，使相邻所述区间信号长度保持一致；
    分别确定每个所述区间信号集中相邻所述区间信号的持续时长；
    确定每个所述区间信号集中相邻所述区间信号的互相关系数；
    基于相邻所述区间信号持续时长及所述互相关系数，确定相邻所述区间信号的第一置信度以及连续设定个数所述区间信号的第二置信度。
  6. 根据权利要求4所述的方法，其中，所述确定所述起始加速度信号对应的起始运动特征信息，包括：
    对所述区间信号的设定分位数范围内进行等分处理，获取分割点匹配序列；
    基于所述分割点匹配序列，确定近似编码序列；
    根据所述设定分位数与所述近似编码序列，确定所述起始加速度信号对应的起始运动特征信息。
  7. 根据权利要求4所述的方法，其中，所述根据所述起始运动特征信息，确定所述目标运动特征信息及目标轴，包括：
    根据所述起始运动特征信息，确定连续设定个数所述区间信号的起始运动特征参数；
    响应于所述起始运动特征参数满足设定条件，聚类所述起始运动特征信息，获得目标运动特征信息及目标轴。
  8. 根据权利要求1所述的方法，其中，所述根据所述当前加速度信号，确定所述用户的当前运动特征信息，包括：
    从所述当前加速度信号中提取目标轴的目标加速度信号，其中，所述目标轴为所述目标运动特征信息关联的坐标轴；
    基于所述目标加速度信号，确定所述用户的当前运动特征信息。
  9. 根据权利要求1所述的方法，其中，所述基于所述当前运动特征信息和所述目标运动特征信息，确定用户的动作次数累计值，包括：
    将所述当前运动特征信息进行划分，获得第一当前运动特征向量和第二当前运动特征向量；
    将所述目标运动特征信息进行划分，获得第一目标运动特征向量和第二目标运动特征向量；
    根据所述第一当前运动特征向量、第二当前运动特征向量、第一目标运动特征向量以及第二目标运动特征向量，确定用户的动作次数累计值。
  10. 根据权利要求9所述的方法，其中，所述根据所述第一当前运动特征向量、第二当前运动特征向量、第一目标运动特征向量以及第二目标运动特征向量，确定用户的动作次数累计值，包括：
    基于所述第一当前运动特征向量与所述第一目标运动特征向量，确定重叠比例；
    判定所述重叠比例是否小于第一设定阈值，响应于所述重叠比例大于或等于第一设定阈值，基于所述第二当前运动特征向量与所述第二目标运动特征向量，确定互相关系数；响应于所述重叠比例小于第一设定阈值，控制用户的动作次数累计值不变；
    判定所述互相关系数是否小于第二设定阈值，响应于所述互相关系数大于或等于第二设定阈值，根据置信度、所述重叠比例以及所述互相关系数，确定估计概率，所述置信度由相邻区间信号确定；响应于所述互相关系数小于第二设定阈值，控制用户的动作次数累计值不变；
    判定所述估计概率是否大于第三设定阈值，响应于所述估计概率大于第三设定阈值，控制用户的动作次数累计值加1作为新的动作次数累计值；响应于所述估计概率小于或等于第三设定阈值，控制用户的动作次数累计值不变。
  11. 根据权利要求1-10任一项所述的方法，在确定所述用户的动作次数累 计值之后，还包括：
    响应于当前不满足计数循环结束条件，将下一滑动窗口的加速度信号作为当前加速度信号，返回重新执行根据当前加速度信号确定用户的当前运动特征信息的操作；响应于当前满足计数循环结束条件，将所述动作次数累计值作为动作的计数结果进行反馈。
  12. 根据权利要求11所述的方法，其中，所述计数循环结束条件为：所述加速度信号在设定时间内未被获取到；或者，所述动作次数累计值连续设定次未发生变化。
  13. 一种动作的计数装置，包括：
    目标信息获取模块，设置为获取目标运动特征信息；
    当前信息确定模块，设置为根据当前加速度信号，确定用户的当前运动特征信息；
    动作次数确定模块，设置为基于所述当前运动特征信息和所述目标运动特征信息，确定用户的动作次数累计值。
  14. 根据权利要求13所述的装置，其中，所述目标信息获取模块包括：
    目标序列确定单元，设置为对起始加速度信号进行多尺度变换分析，确定目标峰值位置序列；
    目标信息确定单元，设置为根据所述目标峰值位置序列，确定目标运动特征信息。
  15. 根据权利要求14所述的装置，其中，所述目标序列确定单元设置为：
    基于所述起始加速度信号，确定所述起始加速度信号对应多个设定尺度的小波序列；
    根据所述小波序列，确定所述起始加速度信号对应的多个峰值位置序列以及多个峰值高度序列；
    基于所述多个峰值高度序列，确定起始加速度对应的多个总体响应输出值；
    根据所述多个总体响应输出值，确定目标峰值位置序列。
  16. 根据权利要求14所述的装置，其中，所述目标信息确定单元设置为：
    根据所述目标峰值位置序列，结合预设合并规则，生成所述起始加速度信号的多个区间信号集，每个所述区间信号集由两个或两个以上区间信号组成；
    分别计算每个所述区间信号集中相邻所述区间信号的第一置信度以及连续设定个数所述区间信号的第二置信度；
    响应于所述第一置信度满足第一置信度条件，确定所述起始加速度信号对应的起始运动特征信息；
    响应于所述第二置信度满足第二置信度条件，根据所述起始运动特征信息，确定所述目标运动特征信息及目标轴。
  17. 根据权利要求13所述的装置，其中，所述当前信息确定模块，包括：
    目标加速度信号确定单元，设置为从所述当前加速度信号中提取目标轴的目标加速度信号，其中，所述目标轴为所述目标运动特征信息关联的坐标轴；
    当前运动特征信息确定单元，设置为基于所述目标加速度信号，确定所述用户的当前运动特征信息。
  18. 根据权利要求13所述的装置，其中，所述动作次数确定模块设置为：
    将所述当前运动特征信息进行划分，获得第一当前运动特征向量和第二当前运动特征向量；
    将所述目标运动特征信息进行划分，获得第一目标运动特征向量和第二目标运动特征向量；
    根据所述第一当前运动特征向量、第二当前运动特征向量、第一目标运动特征向量以及第二目标运动特征向量，确定用户的动作次数累计值。
  19. 根据权利要求13-18任一项所述的装置，还包括：
    循环模块，设置为响应于当前不满足计数循环结束条件，将下一滑动窗口的加速度信号作为当前加速度信号，返回重新执行根据当前加速度信号确定用户的当前运动特征信息的操作；响应于当前满足计数循环结束条件，将所述动作次数累计值作为动作的计数结果进行反馈。
  20. 一种可穿戴设备，包括：存储器以及至少一个处理器；
    所述存储器，设置为存储至少一个程序；
    当所述至少一个程序被所述至少一个处理器执行，使得所述至少一个处理器实现如权利要求1-12任一所述的动作的计数方法。
  21. 一种包含计算机可执行指令的存储介质，所述计算机可执行指令在由计算机处理器执行时用于执行如权利要求1-12任一所述的动作的计数方法。
PCT/CN2022/082721 2022-03-24 2022-03-24 动作的计数方法、装置、设备及存储介质 WO2023178594A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/082721 WO2023178594A1 (zh) 2022-03-24 2022-03-24 动作的计数方法、装置、设备及存储介质

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/082721 WO2023178594A1 (zh) 2022-03-24 2022-03-24 动作的计数方法、装置、设备及存储介质

Publications (1)

Publication Number Publication Date
WO2023178594A1 true WO2023178594A1 (zh) 2023-09-28

Family

ID=88099436

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/082721 WO2023178594A1 (zh) 2022-03-24 2022-03-24 动作的计数方法、装置、设备及存储介质

Country Status (1)

Country Link
WO (1) WO2023178594A1 (zh)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106406516A (zh) * 2016-08-26 2017-02-15 兰州理工大学 一种智能手机本地实时的运动轨迹特征提取与识别方法
CN109603142A (zh) * 2018-11-27 2019-04-12 电子科技大学 一种哑铃运动的识别计数方法及其装置
CN109663323A (zh) * 2018-12-20 2019-04-23 出门问问信息科技有限公司 一种基于可穿戴设备的运动状态监测方法及可穿戴设备
CN111460238A (zh) * 2020-04-03 2020-07-28 广东高驰运动科技有限公司 利用智能手表记录路亚钓鱼数据的方法及智能手表
CN111569397A (zh) * 2020-04-30 2020-08-25 东莞全创光电实业有限公司 手柄类运动计数方法及终端

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106406516A (zh) * 2016-08-26 2017-02-15 兰州理工大学 一种智能手机本地实时的运动轨迹特征提取与识别方法
CN109603142A (zh) * 2018-11-27 2019-04-12 电子科技大学 一种哑铃运动的识别计数方法及其装置
CN109663323A (zh) * 2018-12-20 2019-04-23 出门问问信息科技有限公司 一种基于可穿戴设备的运动状态监测方法及可穿戴设备
CN111460238A (zh) * 2020-04-03 2020-07-28 广东高驰运动科技有限公司 利用智能手表记录路亚钓鱼数据的方法及智能手表
CN111569397A (zh) * 2020-04-30 2020-08-25 东莞全创光电实业有限公司 手柄类运动计数方法及终端

Similar Documents

Publication Publication Date Title
CN109189991A (zh) 重复视频识别方法、装置、终端及计算机可读存储介质
CN112381011B (zh) 基于人脸图像的非接触式心率测量方法、系统及装置
Xi et al. Deep dilation on multimodality time series for human activity recognition
WO2010083562A1 (en) Activity detection
WO2020088491A1 (zh) 一种运动行为模式分类方法、系统以及装置
CN110956139B (zh) 一种基于时间序列回归预测的人体运动动作分析方法
CN110136144B (zh) 一种图像分割方法、装置及终端设备
CN110059770B (zh) 基于位置预测的自适应任务分发方法、装置及相关组件
CN113536862A (zh) 一种聚类方法、装置、设备及存储介质
CN110458235B (zh) 一种视频中运动姿势相似度比对方法
CN109858454B (zh) 一种基于双模型自适应核相关滤波追踪方法
WO2023178594A1 (zh) 动作的计数方法、装置、设备及存储介质
Huang et al. A novel method for video moving object detection using improved independent component analysis
CN113297883A (zh) 信息处理方法、分析模型的获得方法、装置及电子设备
Zahan et al. Learning sparse temporal video mapping for action quality assessment in floor gymnastics
CN111982149B (zh) 一种计步识别方法、装置、设备及可读存储介质
CN114333039B (zh) 一种人像聚类的方法、装置及介质
US11426627B2 (en) Method and apparatus for counting foot step based on acceleration information, and device
CN109582713A (zh) 一种运动状态的识别方法、装置及终端
CN113283465B (zh) 一种弥散张量成像数据分析方法及装置
CN115240269A (zh) 基于体型变换的步态识别方法、设备及存储介质
Chen et al. A multi-person pose estimation with LSTM for video stream
CN116943130A (zh) 一种计数方法、装置、设备及存储介质
CN113850160A (zh) 重复动作的计数方法及装置
CN114140816A (zh) 帕金森病起立任务的评估方法及系统、存储介质及终端

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22932659

Country of ref document: EP

Kind code of ref document: A1