Human health data comprehensive analysis system

Abstract

The invention relates to the field of health data analysis, in particular to a human body health data comprehensive analysis system which comprises a video frame type determination module, a frame extraction mode determination module, a data processing module and a data processing module, and is used for determining a video frame type according to an action domain variation degree and a same-frame reference value and determining a frame extraction mode according to frame extraction conditions, the feature acquisition unit is used for acquiring data features of the monitoring data acquired by the changed limb points in the extracted video frames; the mode judging unit is used for determining a monitoring motion state according to the same feature difference reference value and the feature dimension reference value and determining a feature processing mode according to the monitoring motion state; the first processing unit is used for determining a first feature selection mode according to a set condition; the second processing unit is used for determining a feature category according to the individual influence value and the behavior correlation value and determining a second feature selection mode according to a judgment condition; a motion recognition unit; according to the invention, effective features can be screened to improve the behavior recognition precision, so that the motion risk is reduced.

Description

Technical Field

[0001] The present invention relates to the field of health data analysis, and in particular to a comprehensive analysis system for human health data. Background Art

[0002] Wearable devices can only collect human health data, but cannot accurately identify the wearer's exercise type, and thus cannot warn of various risk factors that may exist in the wearer's health during exercise. Therefore, behavioral information is often identified through a human behavior recognition model. However, due to the complexity of human movement, the feature dimension collected based on the motion data is large, which increases the computational complexity of the human behavior recognition model, which not only leads to a long recognition time, but also may affect the accuracy of behavior recognition. Therefore, how to select effective features to conduct targeted training on the human behavior recognition model to improve the accuracy of behavior recognition and thereby reduce exercise risks is a technical problem that needs to be urgently solved by technical personnel in this field.

[0003] Chinese patent publication number CN114863566A discloses a method for identifying human motion behavior, including the following steps: Step 1, respectively use a motion capture system to collect data on multiple human actions, and obtain raw data including three-axis accelerometer data, three-axis gyroscope data, three-axis magnetometer data, and three-axis attitude angle data; Step 2, construct a CRF conditional random field model applied to text sequence data; Step 3, convert the time series of raw data of human motion containing all joints into feature vectors, and mark the corresponding human motion behavior types, and use the feature vectors and human motion behavior types as corpus to train the CRF model; Step 4, given the time series feature vectors of human motion data to be classified, human motion behavior recognition is performed to obtain the corresponding human motion behavior type. It can be seen that the above technical solution has the following problems: the data collection points are not selected in a targeted manner, resulting in a high redundancy of the collected raw data, and the feature vectors are not screened, resulting in a high feature dimension, which in turn leads to a long recognition time, and may also affect the accuracy of behavior recognition, and cannot accurately provide health reminders. Summary of the invention

[0004] To this end, the present invention provides a comprehensive analysis system for human health data to overcome the problems in the prior art that there is no targeted selection of data collection points, resulting in high redundancy of the collected original data, and no screening of feature vectors, resulting in high feature dimensions, which in turn leads to too long recognition time and may also affect the accuracy of behavior recognition and fail to provide accurate health reminders.

[0005] To achieve the above object, the present invention provides a human health data comprehensive analysis system, comprising:

[0006] A feature acquisition unit is used to determine the type of video frame according to the action domain variation and the same-frame reference value, and determine the frame extraction method according to the frame extraction condition to uniformly extract video frames or extract video frames according to a combination of similar two-class video frames and each class of video frames, and obtain data features of the monitoring data collected from the changed limb points in the extracted video frames;

[0007] A mode determination unit, which is connected to the feature acquisition unit, is used to determine the monitored motion state according to the same feature difference reference value and the feature dimension reference value, and determine the feature processing mode according to the monitored motion state to obtain the mapping value corresponding to each data feature and determine the first feature selection mode according to the setting conditions or determine the second feature selection mode according to the determination conditions;

[0008] A first processing unit, connected to the mode determination unit, is used to determine, according to the setting conditions, that the first feature selection mode is to select data features according to a difference threshold or a ratio correlation;

[0009] A second processing unit, connected to the mode determination unit, is used to determine the data feature category according to the individual influence value and the behavior-related value, and to determine the second feature selection mode according to the determination condition to select the data feature according to the redundant feature combination or the associated threshold;

[0010] A motion recognition unit is connected to the first processing unit and the second processing unit respectively, and is used to perform model training according to the selected features, and to provide health reminders according to the user's basic information when the user exercises.

[0011] Furthermore, the feature acquisition unit determines the video frame type according to the motion domain variation and the same frame reference value, and the video frame type includes:

[0012] A type of video frame whose motion domain variation is greater than or equal to a preset motion domain variation and whose same-frame reference value is less than a preset same-frame reference value;

[0013] The second type of video frame has a motion domain variation less than a preset motion domain variation or a same-frame reference value greater than or equal to a preset same-frame reference value.

[0014] Further, the feature acquisition unit responds to different frame extraction conditions to determine the frame extraction mode;

[0015] The frame extraction condition responded by the feature acquisition unit is that the proportion of one type of video frames is greater than or equal to the preset proportion of one type of video frames and the distribution coefficient of one type of video frames is greater than or equal to the preset distribution coefficient of one type of video frames, and the frame extraction method is to uniformly extract video frames, wherein,

[0016] Divide each video frame into N equal parts according to the time sequence, and record the video frame at each equal division point as an extracted frame;

[0017] The frame extraction condition responded by the feature acquisition unit is that the proportion of the first class of video frames is less than the preset proportion of the first class of video frames or the distribution coefficient of the first class of video frames is less than the preset distribution coefficient of the first class of video frames, and the frame extraction method is to extract video frames according to the combination of similar second class video frames and each first class of video frames, wherein,

[0018] Determine similar second-category video frame combinations according to the limb point similarity coefficients, randomly extract corresponding single second-category video frames from each similar second-category video frame combination, and record them as extracted second-category video frames. Sort each first-category video frame according to chronological order to obtain a reference sequence, divide the reference sequence into M equal parts, and record the first-category video frames at each equal division point and the extracted second-category video frames as extracted frames.

[0019] Further, the state analysis unit determines the monitoring motion state according to the same feature difference reference value and the feature dimension reference value, and the monitoring motion state includes:

[0020] A first monitored motion state in which the same feature difference reference value is greater than or equal to a preset same feature difference reference value or the feature dimension reference value is greater than or equal to a preset feature dimension reference value;

[0021] A second monitored motion state in which the same feature difference reference value is less than a preset same feature difference reference value and the feature dimension reference value is less than a preset feature dimension reference value.

[0022] Further, the mode determination unit responds to different monitored motion states to determine the feature processing mode;

[0023] The mode determination unit responds to the first monitored motion state, the feature processing mode is to obtain the mapping value corresponding to each data feature, and determine the first feature selection mode according to the set conditions;

[0024] The mode determination unit responds to the second monitored motion state, and the feature processing mode is to determine the second feature selection mode according to the determination condition.

[0025] Further, the mapping value corresponding to each data feature is determined according to the feature ratio, and the feature ratio is determined according to the ratio of the first difference to the second difference;

[0026] There is a positive correlation between the mapping value and the feature ratio.

[0027] Furthermore, the setting condition of the first processing unit response is that the mapping value distribution difference is greater than or equal to the preset mapping value distribution difference, and the first feature selection method is to select data features according to the difference threshold, wherein,

[0028] A similar set is determined according to a difference threshold, and a number of data features in each similar set are randomly extracted; the difference threshold is the absolute value of the difference between the mapping values ​​corresponding to the two features.

[0029] Furthermore, the setting condition of the first processing unit response is that the mapping value distribution difference is less than the preset mapping value distribution difference, and the first feature selection method is to select data features according to the ratio correlation, wherein,

[0030] Similar sets are determined according to ratio correlation, and several data features in each similar set are randomly extracted; the ratio correlation is determined according to the absolute value of the difference between the ratio correlations corresponding to the two data features, and the difference ratio is determined according to the ratio of the inter-class difference to the intra-class difference.

[0031] Furthermore, the second processing unit determines a data feature category according to the individual influence value and the behavior-related value, and the data feature category includes:

[0032] A type of data feature whose individual influence value is greater than or equal to a preset individual influence value or whose behavior-related value is less than a preset behavior-related value;

[0033] The second type of data features are that the individual influence value is less than the preset individual influence value and the behavior correlation value is greater than or equal to the preset behavior correlation value.

[0034] Further, the second processing unit determines a second feature selection method according to the determination condition;

[0035] The judgment condition of the second processing unit response is that the proportion of a type of data features is greater than or equal to the preset proportion of a type of data features and the feature richness is greater than or equal to the preset feature richness. The second feature selection method is to select data features according to the redundant feature combination, wherein,

[0036] Determine redundant feature combinations according to feature correlations, and randomly select a first preset number of data features from each redundant feature combination;

[0037] The judgment condition of the second processing unit response is that the proportion of a type of data features is less than the preset proportion of a type of data features or the feature richness is less than the preset feature richness, and the second feature selection method is to select data features according to the correlation threshold, wherein,

[0038] A second preset number of data features are selected according to the order of the associated thresholds from large to small.

[0039] Compared with the prior art, the beneficial effect of the present invention lies in that, in the technical scheme of the present invention, the video frame type is determined according to the action domain variability and the same-frame reference value, and the complexity of the human body motion in the video frame and the similarity between the human body motion and the human body motion in other video frames are effectively reflected through the action domain variability and the same-frame reference value, and then the frame extraction condition is effectively determined according to the video frame type, and the number and density of a type of video frames are effectively reflected through the frame extraction condition, and then different frame extraction methods are adaptively selected, so that the selection of the frame extraction method is more in line with the actual application scenario, and then the changed limb points in the extracted frame extracted according to the frame extraction method are used as data monitoring points, avoiding the defect of the prior art that the data collection points are not selected in a targeted manner, resulting in a high redundancy of the collected original data, thereby effectively reducing the cost of data collection and processing, and at the same time ensuring that the monitoring data of the data monitoring point is highly representative of key actions or posture changes, which is conducive to improving the representativeness of several data features obtained by feature extraction.

[0040] Furthermore, in the present invention, the monitored motion state is determined according to the same feature difference reference value and the feature dimension reference value, and the magnitude difference of the data features and the number of data features are effectively reflected through the same feature difference reference value and the feature dimension reference value, and then different feature processing methods are adaptively selected according to the monitored motion state, so that the selection of feature processing method is more in line with the actual application scenario, avoiding the defect of not screening the feature vector in the prior art, resulting in a high data feature dimension, and also avoiding the increase in computational complexity and the risk of overfitting due to excessively high data feature dimensions, thereby improving the recognition accuracy and generalization ability of the model, and then effectively warning of various risk factors that may exist in the wearer's health during exercise.

[0041] Furthermore, the present invention effectively reflects the distribution of mapping values ​​by setting conditions, and then adaptively selects different first feature selection methods according to the set conditions, so that the selection of the first feature selection method is more in line with the actual application scenario, avoiding the disadvantage of not being able to effectively select data features when the difference in the distribution of mapping values ​​is small, which is conducive to reducing redundant information and retaining key data features so that the model can more accurately capture the potential rules in the data, thereby improving the accuracy of behavior recognition.

[0042] Furthermore, in the present invention, data feature categories are determined according to individual influence values ​​and behavior-related values, and the degree of influence of individual differences on data features and the degree of correlation between data features and the data features in other behaviors are effectively reflected through individual influence values ​​and behavior-related values, so that the determination of data feature categories is more in line with actual application scenarios. The judgment conditions effectively reflect the number of data features of a type and the richness of non-zero data features, and then different second feature selection methods are adaptively selected according to the judgment conditions, which is conducive to removing redundant data features and reducing the overfitting of the model to specific data. It can also reduce the complexity and computational complexity of the model, thereby improving the training speed and behavior recognition accuracy of the model, and then providing health reminders based on human behavior and basic information. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A unit connection diagram of the human health data comprehensive analysis system of the present invention;

[0044] Figure 2 A flow chart of determining a frame extraction method according to a frame extraction condition of the present invention;

[0045] Figure 3 A flow chart of the present invention for determining a feature processing method according to a monitored motion state;

[0046] Figure 4 This is a flow chart of the present invention for determining the first feature selection method according to setting conditions. DETAILED DESCRIPTION

[0047] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0048] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0049] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is merely for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0050] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0051] See also Figures 1 to 4 As shown, the present invention provides a human health data comprehensive analysis system, comprising:

[0052] A feature acquisition unit is used to determine the type of video frame according to the action domain variation and the same-frame reference value, and determine the frame extraction method according to the frame extraction condition to uniformly extract video frames or extract video frames according to a combination of similar two-class video frames and each class of video frames, and obtain data features of the monitoring data collected from the changed limb points in the extracted video frames;

[0053] A mode determination unit, which is connected to the feature acquisition unit, is used to determine the monitored motion state according to the same feature difference reference value and the feature dimension reference value, and determine the feature processing mode according to the monitored motion state to obtain the mapping value corresponding to each data feature and determine the first feature selection mode according to the setting conditions or determine the second feature selection mode according to the determination conditions;

[0054] A first processing unit, connected to the mode determination unit, is used to determine, according to the setting conditions, that the first feature selection mode is to select data features according to a difference threshold or a ratio correlation;

[0055] A second processing unit, connected to the mode determination unit, is used to determine the data feature category according to the individual influence value and the behavior-related value, and to determine the second feature selection mode according to the determination condition to select the data feature according to the redundant feature combination or the associated threshold;

[0056] A motion recognition unit is connected to the first processing unit and the second processing unit respectively, and is used to perform model training according to the selected features, and to provide health reminders according to the user's basic information when the user exercises.

[0057] The present invention also includes a data acquisition unit for collecting health data information, which is connected to the feature acquisition unit and the motion recognition unit respectively. The health data information includes monitoring information and basic information. The monitoring information includes several monitoring videos. A single monitoring video is a monitoring video of a single user performing a motion behavior. The motion behavior includes running, walking, squatting and standing up, etc. The changed limb point is a limb point whose position has changed compared with the limb point position in the standard state. The standard state is a state in which the user maintains a military posture standing in the video frame; a point is randomly selected on each joint of the human body, and these points are limb points, including the head, shoulders, elbows, wrists, hips, knees and ankles, etc. The identification of the limb points can be determined by machine vision and deep learning networks. The basic information includes several health parameters, including but not limited to age, blood pressure, heart rate and blood oxygen saturation. This is content that is easy for technical personnel in this field to understand and will not be elaborated here. The monitoring data includes several data units. When a single data unit is a single user performing a motion behavior, the inertial sensor motion capture device is fixed at the corresponding data monitoring point to obtain the monitoring parameters corresponding to each time point monitored in real time as the duration of the motion behavior lasts. The monitoring parameters include motion acceleration, angular velocity, and direction, etc. Among them, the time point is set by the user. A method for setting the time point is provided, in which a time point is set every 1 second, that is, the monitoring parameters are recorded every 1 second.

[0058] When obtaining data features of monitoring data collected from changed limb points in extracted video frames, available acquisition methods include Fourier transform and wavelet transform, etc. The acquired data features include but are not limited to mean, variance, skewness, kurtosis, autocorrelation coefficient, interquartile range and root mean square. Users can obtain several data features according to their own needs. This is content that is easy for technical personnel in this field to understand and will not be elaborated on in detail.

[0059] The present invention is provided with a health behavior library, which stores the warning parameter range of each health parameter corresponding to each exercise behavior, so as to provide a health reminder when the exercise recognition unit recognizes a certain exercise behavior of the human body and each health parameter in the user's basic information is not within the corresponding warning parameter range;

[0060] In the present invention, several historical records are correspondingly set up, and any historical record records the action domain variation, same-frame reference value, proportion of a type of video frame, distribution coefficient of a type of video frame and same feature difference reference value in the historical process of at least one data feature selection, and each historical record corresponds to a qualified mark, which records whether the recognition speed and recognition accuracy of the human behavior recognition model meet user requirements, and the qualified mark can be recorded manually.

[0061] Specifically, the feature acquisition unit determines the video frame type according to the action domain variation and the same frame reference value, and the video frame type includes:

[0062] A type of video frame whose motion domain variation is greater than or equal to a preset motion domain variation and whose same-frame reference value is less than a preset same-frame reference value;

[0063] The second type of video frame has a motion domain variation less than a preset motion domain variation or a same-frame reference value greater than or equal to a preset same-frame reference value.

[0064] Among them, action domain variability = number of changed limb points + reference value of change amplitude, the number of changed limb points is the total number of changed limb points, it can be understood that the video frame is an image with a rectangular shape, the lower left corner of the rectangle is taken as the coordinate origin, the straight line extending from the coordinate origin along the bottom edge of the rectangle to the right is taken as the x-axis, and the straight line extending from the coordinate origin along the left side of the rectangle upward is taken as the y-axis, a rectangular coordinate system is established, the limb point position is the coordinate of the limb point in the rectangular coordinate system, for a single limb point, the limb point is recorded as the target limb point, the shortest distance between the target limb point position in the video frame and the corresponding limb point position under the standard state is recorded as the change distance, and the reference value of the change amplitude is the sum of the change distances of each limb point;

[0065] The same-frame reference value confirmation method is that, for a surveillance video, a single video frame in the surveillance video is recorded as a target video frame, and other video frames excluding the target video frame are recorded as reference video frames. The reference video frame with the same change limb point as the target video frame and a change amplitude difference value less than a preset change amplitude difference value is recorded as the same frame. The same-frame reference value = (the number of same frames + 1) / the number of all video frames in the surveillance video. The change amplitude difference value is the absolute value of the difference between the change amplitude reference values ​​corresponding to the two video frames. The value of the preset change amplitude difference value can be determined by the user according to the actual application scenario. The greater the user's demand for the similarity of the action between the reference video frame and the target video frame, the smaller the value of the preset change amplitude difference value. A preset change amplitude difference value is provided, and the preset change amplitude difference value is 5% of the change amplitude reference value corresponding to the target video frame.

[0066] The values ​​of the preset action domain variation and the preset same-frame reference value can be determined by the user according to the actual application scenario. The higher the user's demand for the complexity of the action in a type of video frame, the larger the value of the preset action domain variation, and the smaller the value of the preset same-frame reference value. A value of the preset action domain variation and the preset same-frame reference value is provided, and the average value of the action domain variation corresponding to the historical records that can meet the user's needs is recorded as the preset action domain variation, and the average value of the same-frame reference values ​​corresponding to the historical records that can meet the user's needs is recorded as the preset same-frame reference value.

[0067] Specifically, the feature acquisition unit responds to different frame extraction conditions to determine the frame extraction mode;

[0068] The frame extraction condition responded by the feature acquisition unit is that the proportion of one type of video frames is greater than or equal to the preset proportion of one type of video frames and the distribution coefficient of one type of video frames is greater than or equal to the preset distribution coefficient of one type of video frames, and the frame extraction method is to uniformly extract video frames, wherein,

[0069] Divide each video frame into N equal parts according to the time sequence, and record the video frame at each equal division point as an extracted frame;

[0070] The frame extraction condition responded by the feature acquisition unit is that the proportion of the first class of video frames is less than the preset proportion of the first class of video frames or the distribution coefficient of the first class of video frames is less than the preset distribution coefficient of the first class of video frames, and the frame extraction method is to extract video frames according to the combination of similar second class video frames and each first class of video frames, wherein,

[0071] Determine similar second-category video frame combinations according to the limb point similarity coefficients, randomly extract corresponding single second-category video frames from each similar second-category video frame combination, and record them as extracted second-category video frames. Sort each first-category video frame according to chronological order to obtain a reference sequence, divide the reference sequence into M equal parts, and record the first-category video frames at each equal division point and the extracted second-category video frames as extracted frames.

[0072] The frame extraction condition includes a first frame extraction condition and a second frame extraction condition, the first frame extraction condition is that the proportion of one type of video frames is greater than or equal to the preset proportion of one type of video frames and the distribution coefficient of one type of video frames is greater than or equal to the preset distribution coefficient of one type of video frames, and the second frame extraction condition is that the proportion of one type of video frames is less than the preset proportion of one type of video frames or the distribution coefficient of one type of video frames is less than the preset distribution coefficient of one type of video frames;

[0073] The method for determining the proportion of video frames of a category is as follows: for a surveillance video, the proportion of video frames of a category = the number of video frames of a category in the surveillance video / the number of video frames in the surveillance video;

[0074] The distribution coefficient of the first-class video frame = adjacent difference threshold + distance segment value, the adjacent difference threshold is the average value of the adjacent distance means corresponding to each first-class video frame in a single monitoring video; for a single first-class video frame, the first-class video frame is recorded as the target first-class video frame, the average value of the distance between the first-class video frames adjacent to the target first-class video frame and the target first-class video frame is recorded as the adjacent distance mean, and the average value of the adjacent distance means corresponding to each abnormal frequency is recorded as the adjacent difference threshold; the distance between two first-class video frames is the absolute value of the difference in monitoring time corresponding to the two first-class video frames; in the time direction, the distance segment value = the monitoring time corresponding to the last first-class video frame - the monitoring time corresponding to the first first-class video frame;

[0075] The values ​​of the preset proportion of a class of video frames and the preset distribution coefficient of a class of video frames can be determined by the user according to the actual application scenario. The greater the user's demand for the representativeness of the data monitoring point monitoring, the smaller the values ​​of the proportion of a class of video frames and the preset distribution coefficient of a class of video frames are. A preset proportion of a class of video frames and the preset distribution coefficient of a class of video frames are provided, and the average value of the proportion of a class of video frames corresponding to the historical records that can meet the user's needs is recorded as the preset proportion of a class of video frames, and the average value of the distribution coefficient of a class of video frames corresponding to the historical records that can meet the user's needs is recorded as the preset distribution coefficient of a class of video frames;

[0076] The values ​​of N and M can be determined by the user according to actual needs. The greater the user's demand for the representativeness of the data monitoring point monitoring, the larger the values ​​of N and M are. Provided is a value of N and M, N is 40% of the total number of video frames in a single monitoring video, and M is 40% of the total number of video frames in the reference sequence.

[0077] For two types of video frames, the number of changed limb points corresponding to one type of video frame is recorded as A, and the number of changed limb points corresponding to the other type of video frame is recorded as B. The limb point similarity coefficient = the number of the same changed limb points in the two video frames / the larger value of A and B;

[0078] Determining a similar second-category video frame combination according to a limb point similarity coefficient includes: performing a combination analysis on each second-category video frame in a surveillance video according to the order of monitoring time corresponding to the second-category video frames from early to late; when performing a combination analysis on a single second-category video frame, recording the second-category video frame as a target second-category video frame, recording other second-category video frames excluding the target second-category video frame as reference second-category video frames, recording a reference second-category video frame whose limb point similarity coefficient is greater than a preset limb point similarity coefficient as a similar second-category video frame combination, and performing a combination analysis on second-category video frames that are not recorded as a similar second-category video frame combination, until each second-category video frame is recorded as a similar second-category video frame combination, then stopping the combination analysis.

[0079] Specifically, the state analysis unit determines the monitoring motion state according to the same feature difference reference value and the feature dimension reference value, and the monitoring motion state includes:

[0080] A first monitored motion state in which the same feature difference reference value is greater than or equal to a preset same feature difference reference value or the feature dimension reference value is greater than or equal to a preset feature dimension reference value;

[0081] A second monitored motion state in which the same feature difference reference value is less than a preset same feature difference reference value and the feature dimension reference value is less than a preset feature dimension reference value.

[0082] Among them, each changed limb point extracted from each video frame is recorded as a data detection point, and for a movement behavior, a set of data features of monitoring data of each data detection point corresponding to a single monitoring video corresponding to the movement behavior is recorded as a feature set;

[0083] The reference value of the same feature difference is the maximum value of the inter-class difference corresponding to each data feature in a motion behavior. The confirmation method of the inter-class difference is to record each feature set corresponding to the motion behavior as a reference set, and for a single data feature, record the data feature as a target feature. The inter-class difference = the maximum value of the target feature in each reference set - the minimum value of the target feature in each reference set; the feature dimension reference value is the number of data features corresponding to a motion behavior;

[0084] The values ​​of the preset same feature difference reference value and the preset feature dimension reference value can be determined by the user according to the actual application scenario. It can be understood that the larger the values ​​of the preset same feature difference reference value and the preset feature dimension reference value are, the greater the user's need to select a feature processing method to determine a second feature selection method according to a judgment condition. A value of a preset same feature difference reference value and a preset feature dimension reference value is provided, and a feature processing method is detected to obtain a mapping value corresponding to each data feature and determine a historical record of the first feature selection method according to the set conditions. The average value of the same feature difference reference values ​​corresponding to the historical records that can meet the user's needs is recorded as the preset same feature difference reference value, and the average value of the feature dimension reference values ​​corresponding to the historical records that can meet the user's needs is recorded as the preset feature dimension reference value.

[0085] Specifically, the mode determination unit responds to different monitored motion states to determine the feature processing mode;

[0086] The mode determination unit responds to the first monitored motion state, the feature processing mode is to obtain the mapping value corresponding to each data feature, and determine the first feature selection mode according to the set conditions;

[0087] The mode determination unit responds to the second monitored motion state, and the feature processing mode is to determine the second feature selection mode according to the determination condition.

[0088] The setting condition includes a first setting condition and a second setting condition, the first setting condition is that the mapping value distribution difference is greater than or equal to a preset mapping value distribution difference, and the second setting condition is that the mapping value distribution difference is less than a preset mapping value distribution difference;

[0089] The judgment conditions include a first judgment condition and a second judgment condition. The first judgment condition is that the proportion of one type of data features is greater than or equal to the preset proportion of one type of data features and the feature richness is greater than or equal to the preset feature richness. The second judgment condition is that the proportion of one type of data features is less than the preset proportion of one type of data features or the feature richness is less than the preset feature richness.

[0090] Specifically, the mapping value corresponding to each data feature is determined according to the feature ratio, and the feature ratio is determined according to the ratio of the first difference to the second difference;

[0091] There is a positive correlation between the mapping value and the feature ratio.

[0092] Among them, for a single data feature, the numerical value corresponding to the feature is recorded as the feature value, the mapping value = 2×feature ratio - 1, the feature ratio = first difference / second difference, the first difference = feature value - minimum eigenvalue, the second difference = maximum eigenvalue - minimum eigenvalue, the maximum eigenvalue and the minimum eigenvalue are respectively the maximum and minimum values ​​of the numerical values ​​corresponding to each data feature in a single feature set.

[0093] Specifically, the setting condition of the first processing unit response is that the mapping value distribution difference is greater than or equal to the preset mapping value distribution difference, and the first feature selection method is to select data features according to the difference threshold, wherein,

[0094] A similar set is determined according to a difference threshold, and a number of data features in each similar set are randomly extracted; the difference threshold is the absolute value of the difference between the mapping values ​​corresponding to the two features.

[0095] The mapping value distribution difference is the average value of the minimum difference corresponding to each mapping value. The minimum difference is confirmed by recording a mapping value as a target mapping value, recording other mapping values ​​excluding the target mapping value as reference mapping values, detecting the absolute value of the difference between each reference mapping value and the target mapping value as a reference value, and recording the minimum value of each reference value as the minimum difference.

[0096] The value of the preset mapping value distribution difference can be determined by the user according to the actual application scenario. It can be understood that the larger the value of the preset mapping value distribution difference is, the greater the user's need to determine the similarity set according to the difference threshold is. A value of the preset mapping value distribution difference is provided, and the historical records corresponding to the similarity set determined according to the difference threshold are detected, and the average value of the mapping value distribution difference corresponding to the historical records that can meet the user's needs is recorded as the preset mapping value distribution difference;

[0097] Determining a similarity set according to a difference threshold includes: performing similarity analysis on each data feature in a feature set according to the order of mapping values ​​from small to large, when performing similarity analysis on a single data feature, recording the data feature as a first target feature, recording other features other than the first target feature as a first reference feature, recording the first reference feature and the first target feature whose difference threshold with the first target feature is less than a preset difference threshold as a similarity set, and performing similarity analysis on data features that are not recorded as similarity sets, until all data features are recorded as similarity sets, then stopping the similarity analysis.

[0098] The value of the preset difference threshold can be determined by the user according to the actual application scenario. The greater the user's demand for the degree of correlation of data features in the similarity set, the smaller the value of the preset difference threshold. A value of the preset difference threshold is provided, and the average value of the difference threshold corresponding to the historical records that can meet the user's needs is recorded as the preset difference threshold.

[0099] Specifically, the setting condition of the first processing unit response is that the mapping value distribution difference is less than the preset mapping value distribution difference, and the first feature selection method is to select data features according to the ratio correlation, wherein,

[0100] Similar sets are determined according to ratio correlation, and several data features in each similar set are randomly extracted; the ratio correlation is determined according to the absolute value of the difference between the ratio correlations corresponding to the two data features, and the difference ratio is determined according to the ratio of the inter-class difference to the intra-class difference.

[0101] Among them, the difference ratio = inter-class difference / intra-class difference, the intra-class difference is confirmed by recording a data feature in the feature set as the target reference feature, and recording the absolute value of the difference between other data features in the detection feature set that do not include the target reference feature and the target reference feature as the reference absolute value, and the intra-class difference = the maximum value of each reference absolute value - the minimum value of each reference absolute value;

[0102] Determining similar sets according to ratio correlations includes: performing set analysis on each data feature in a feature set according to the order of ratio correlations from small to large, when performing set analysis on a single data feature, recording the data feature as a second target feature, recording other data features other than the second target feature as second reference features, recording the second reference feature and the second target feature whose ratio correlation with the second target feature is less than a preset ratio correlation as a similar set, and performing set analysis on data features that are not recorded as similar sets, until all data features are recorded as similar sets, then stopping the set analysis.

[0103] The value of the preset ratio correlation can be determined by the user according to the actual application scenario. The greater the user's demand for the correlation degree of data features in the similarity set, the smaller the value of the preset ratio correlation will be. A value of the preset ratio correlation is provided, and the average value of the ratio correlation corresponding to the historical records that can meet the user's needs is recorded as the preset ratio correlation.

[0104] Specifically, the second processing unit determines the data feature category according to the individual influence value and the behavior-related value, and the data feature category includes:

[0105] A type of data feature whose individual influence value is greater than or equal to a preset individual influence value or whose behavior-related value is less than a preset behavior-related value;

[0106] The second type of data features are that the individual influence value is less than the preset individual influence value and the behavior correlation value is greater than or equal to the preset behavior correlation value.

[0107] Among them, for each feature set corresponding to a single motion behavior, the age of the user corresponding to each feature set is detected. For a single data feature, each feature value corresponding to the data feature is recorded as the target feature value, the average value of each target feature value corresponding to the maximum age is recorded as K1, and the average value of each target feature value corresponding to the minimum age is recorded as K2. The individual influence value = |K1-K2| / (maximum age-minimum age).

[0108] The behavior correlation value is confirmed in the following manner: for a movement behavior, the movement behavior is recorded as the target movement behavior, and other movement behaviors excluding the target movement behavior are recorded as reference movement behaviors. For a single data feature in the target movement behavior, the data feature is recorded as the reference feature, and the feature mean corresponding to each movement behavior is detected. The feature mean is the average value of the feature value corresponding to the reference feature in the single movement behavior. The reference movement behavior whose feature mean difference with the target movement behavior is less than the preset feature mean difference is recorded as a similar movement behavior. The behavior correlation value = the number of similar movement behaviors / the number of all movement behaviors. The feature mean difference is the absolute value of the difference between the feature means corresponding to the two movement behaviors.

[0109] The values ​​of the preset individual impact value and the preset behavior-related value can be determined by the user according to the actual application scenario. It can be understood that the smaller the value of the preset individual impact value and the larger the value of the preset behavior-related value, the greater the user's need to determine that the data feature is a Class I data feature. A preset individual impact value and a preset behavior-related value are provided, and the historical records corresponding to each Class II data feature are detected. The average value of the individual impact values ​​corresponding to the historical records that can meet the user's needs is recorded as the preset individual impact value, and the average value of the behavior-related values ​​that can meet the user's needs is recorded as the preset behavior-related value.

[0110] Specifically, the second processing unit determines the second feature selection method according to the determination condition;

[0111] The judgment condition of the second processing unit response is that the proportion of a type of data features is greater than or equal to the preset proportion of a type of data features and the feature richness is greater than or equal to the preset feature richness. The second feature selection method is to select data features according to the redundant feature combination, wherein,

[0112] Determine redundant feature combinations according to feature correlations, and randomly select a first preset number of data features from each redundant feature combination;

[0113] The judgment condition of the second processing unit response is that the proportion of a type of data features is less than the preset proportion of a type of data features or the feature richness is less than the preset feature richness, and the second feature selection method is to select data features according to the correlation threshold, wherein,

[0114] A second preset number of data features are selected according to the order of the associated thresholds from large to small.

[0115] Among them, for a feature set, the proportion of a type of data features = the number of a type of data features in the feature set / the total number of data features in the feature set, feature richness = the number of non-zero data features in the feature set / the total number of data features in the feature set;

[0116] The values ​​of the proportion of a preset type of data features and the richness of the preset features can be determined by the user according to the actual application scenario. It can be understood that the smaller the values ​​of the proportion of a preset type of data features and the richness of the preset features are, the greater the user's need to select the second feature selection method of selecting data features according to the combination of redundant features. A value of the proportion of a preset type of data features and the richness of the preset features is provided, and the historical records of the second feature selection method of selecting data features according to the associated threshold are detected, and the average value of the proportion of a type of data features corresponding to the historical records that can meet the user's needs is recorded as the proportion of the preset type of data features, and the average value of the feature richness corresponding to the historical records that can meet the user's needs is recorded as the preset feature richness;

[0117] For any two data features, the calculation formula for the feature correlation σ corresponding to the two data features is:

[0118]

[0119] m is the number of feature sets corresponding to a single motion behavior; x k and k are the values ​​of the two data features in the kth feature set in the feature set corresponding to a single motion behavior, For x k The average value of the corresponding data feature, for yk The average value of the corresponding data feature, k = 1, 2, 3, ..., m;

[0120] Determining a redundant feature combination according to feature correlation includes: performing redundant analysis on each data feature in a feature set, when performing redundant analysis on a single data feature, recording the data feature as a third target feature, recording other data features other than the third target feature as third reference features, recording the third reference feature whose feature correlation with the third target feature is less than a preset feature correlation and the third target feature as a redundant feature combination, and continuing to perform redundant analysis on data features that are not recorded as redundant feature combinations until all data features are recorded as redundant feature combinations, then stopping the redundant analysis.

[0121] Association threshold = individual influence value - behavior correlation value. The values ​​of the first preset number and the second preset number can be determined by the user according to his own needs. A first preset number and a second preset number are provided. The first preset number is 20% of the number of data features in a single redundant feature combination, and the preset number is 20% of the number of data features in a single feature set.

[0122] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0123] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A human health data comprehensive analysis system, characterized in that: include: A feature acquisition unit is used to determine the type of video frame according to the action domain variation and the same-frame reference value, and determine the frame extraction method according to the frame extraction condition to uniformly extract video frames or extract video frames according to a combination of similar two-class video frames and each class of video frames, and obtain data features of the monitoring data collected from the changed limb points in the extracted video frames; A mode determination unit, which is connected to the feature acquisition unit, is used to determine the monitored motion state according to the same feature difference reference value and the feature dimension reference value, and determine the feature processing mode according to the monitored motion state to obtain the mapping value corresponding to each data feature and determine the first feature selection mode according to the setting conditions or determine the second feature selection mode according to the determination conditions; A first processing unit, connected to the mode determination unit, is used to determine, according to the setting conditions, that the first feature selection mode is to select data features according to a difference threshold or a ratio correlation; A second processing unit, connected to the mode determination unit, is used to determine the data feature category according to the individual influence value and the behavior-related value, and to determine the second feature selection mode according to the determination condition to select the data feature according to the redundant feature combination or the associated threshold; A motion recognition unit is connected to the first processing unit and the second processing unit respectively, and is used to perform model training according to the selected features, and to provide health reminders according to the user's basic information when the user exercises.

2. The human health data comprehensive analysis system according to claim 1, characterized in that: The feature acquisition unit determines the video frame type according to the action domain variation and the same frame reference value, and the video frame type includes: A type of video frame whose motion domain variation is greater than or equal to a preset motion domain variation and whose same-frame reference value is less than a preset same-frame reference value; The second type of video frame has a motion domain variation less than a preset motion domain variation or a same-frame reference value greater than or equal to a preset same-frame reference value.

3. The human health data comprehensive analysis system according to claim 2 is characterized in that: The feature acquisition unit responds to different frame extraction conditions to determine the frame extraction mode; The frame extraction condition responded by the feature acquisition unit is that the proportion of one type of video frames is greater than or equal to the preset proportion of one type of video frames and the distribution coefficient of one type of video frames is greater than or equal to the preset distribution coefficient of one type of video frames, and the frame extraction method is to uniformly extract video frames, wherein, Divide each video frame into N equal parts according to the time sequence, and record the video frame at each equal division point as an extracted frame; The frame extraction condition responded by the feature acquisition unit is that the proportion of the first class of video frames is less than the preset proportion of the first class of video frames or the distribution coefficient of the first class of video frames is less than the preset distribution coefficient of the first class of video frames, and the frame extraction method is to extract video frames according to the combination of similar second class video frames and each first class of video frames, wherein, Determine similar second-category video frame combinations according to the limb point similarity coefficients, randomly extract corresponding single second-category video frames from each similar second-category video frame combination, and record them as extracted second-category video frames. Sort each first-category video frame according to chronological order to obtain a reference sequence, divide the reference sequence into M equal parts, and record the first-category video frames at each equal division point and the extracted second-category video frames as extracted frames.

4. The human health data comprehensive analysis system according to claim 3 is characterized in that: The state analysis unit determines the monitoring motion state according to the same feature difference reference value and the feature dimension reference value, and the monitoring motion state includes: A first monitored motion state in which the same feature difference reference value is greater than or equal to a preset same feature difference reference value or the feature dimension reference value is greater than or equal to a preset feature dimension reference value; A second monitored motion state in which the same feature difference reference value is less than a preset same feature difference reference value and the feature dimension reference value is less than a preset feature dimension reference value.

5. The human health data comprehensive analysis system according to claim 4 is characterized in that: The mode determination unit responds to different monitored motion states to determine the feature processing mode; The mode determination unit responds to the first monitored motion state, the feature processing mode is to obtain the mapping value corresponding to each data feature, and determine the first feature selection mode according to the set conditions; The mode determination unit responds to the second monitored motion state, and the feature processing mode is to determine the second feature selection mode according to the determination condition.

6. The human health data comprehensive analysis system according to claim 5, characterized in that: The mapping value corresponding to each data feature is determined according to the feature ratio, and the feature ratio is determined according to the ratio of the first difference to the second difference; There is a positive correlation between the mapping value and the feature ratio.

7. The human health data comprehensive analysis system according to claim 6, characterized in that: The setting condition of the first processing unit response is that the mapping value distribution difference is greater than or equal to the preset mapping value distribution difference, and the first feature selection method is to select data features according to the difference threshold, wherein, A similar set is determined according to a difference threshold, and a number of data features in each similar set are randomly extracted; the difference threshold is the absolute value of the difference between the mapping values ​​corresponding to the two features.

8. The human health data comprehensive analysis system according to claim 7, characterized in that: The setting condition of the first processing unit response is that the mapping value distribution difference is less than the preset mapping value distribution difference, and the first feature selection method is to select data features according to the ratio correlation, wherein, Similar sets are determined according to ratio correlation, and several data features in each similar set are randomly extracted; the ratio correlation is determined according to the absolute value of the difference between the ratio correlations corresponding to the two data features, and the difference ratio is determined according to the ratio of the inter-class difference to the intra-class difference.

9. The human health data comprehensive analysis system according to claim 5, characterized in that: The second processing unit determines the data feature category according to the individual influence value and the behavior-related value, and the data feature category includes: A type of data feature whose individual influence value is greater than or equal to a preset individual influence value or whose behavior-related value is less than a preset behavior-related value; The second type of data features are that the individual influence value is less than the preset individual influence value and the behavior correlation value is greater than or equal to the preset behavior correlation value.

10. The human health data comprehensive analysis system according to claim 9, characterized in that: The second processing unit determines a second feature selection method according to the determination condition; The judgment condition of the second processing unit response is that the proportion of a type of data features is greater than or equal to the preset proportion of a type of data features and the feature richness is greater than or equal to the preset feature richness. The second feature selection method is to select data features according to the redundant feature combination, wherein, Determine redundant feature combinations according to feature correlations, and randomly select a first preset number of data features from each redundant feature combination; The judgment condition of the second processing unit response is that the proportion of a type of data features is less than the preset proportion of a type of data features or the feature richness is less than the preset feature richness, and the second feature selection method is to select data features according to the correlation threshold, wherein, A second preset number of data features are selected according to the order of the associated thresholds from large to small.

Images