Precise prediction method for individual heart rate in field domain scene

Through the speed estimation and heart rate prediction model, combined with the YOLOv5m model for contactless heart rate monitoring, the problem of heart rate data loss during exercise of wearable devices is solved, and more accurate and stable heart rate monitoring is achieved.

CN119055204AInactive Publication Date: 2024-12-03BEIJING UNIV OF POSTS & TELECOMM

Patent Information

Application Number
CN202411183319.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-27
Publication Date
2024-12-03
Estimated Expiration
Not applicable · inactive patent

Smart Images

  • Figure CN119055204A_ABST
    Figure CN119055204A_ABST
Patent Text Reader

Abstract

The invention relates to a method for accurately predicting the heart rate of an individual in a field domain scene, and the method comprises the steps: estimating the instantaneous speed of a to-be-detected target through a speed estimation model, obtaining the instantaneous speed of the to-be-detected target, and enabling the speed estimation model to be a relation model of a moment and the instantaneous speed; based on the instantaneous speed of the to-be-detected target, a heart rate prediction model is used for predicting the heart rate of the to-be-detected target, the heart rate of the to-be-detected target is obtained, and the heart rate prediction model is a relation model of the instantaneous speed and the heart rate. According to the method, the accurate exercise heart rate is deduced by utilizing the relevance between the exercise speed and the heart rate and combining an advanced algorithm, so that the common problems of a contact sensor are effectively avoided.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of heart rate measurement technology, and in particular to a method for accurately predicting an individual's heart rate in a localized scenario. Background Art

[0002] The measurement of exercise heart rate is an important means to evaluate cardiovascular adaptability, training effects and physical health. Exercise heart rate (EHR) refers to the heart rate level during exercise or physical activity, which reflects the heart's adaptation to exercise intensity and the individual's exercise tolerance. Entering the 21st century, photoplethysmography (PPG) has become a new development direction. This technology measures heart rate by detecting changes in blood flow on the surface of the skin and is widely used in various wearable smart watches and fitness trackers. However, PPG technology is greatly disturbed by the noise generated by exercise, and sweat during exercise will interfere with the measurement. Remote photoplethysmography (rPPG) is a non-contact method that uses a video camera to detect subtle changes in skin tone caused by blood flow, but it may be affected by environmental factors.

[0003] In the digital age, various big data models with data-driven as the core automatically process research objects by category. This categorized group has homogeneity in certain indicators. By collecting personalized and precise data and analyzing the characteristics of the homogeneity of certain groups, it is possible to intervene in certain groups in reverse and achieve the desired effect. After the above process, heart rate measurement has obviously greatly improved in measurement accuracy, convenience, convenience, and research. However, the most obvious problem is that the measurement method is usually applied to various scenarios, and the data support behind this measurement method cannot be adapted to various scenarios. Therefore, the measurement of exercise heart rate that is suitable for categorized and homogeneous groups is urgently needed.

[0004] Under normal circumstances, the heart rate and pulse of a normal person are basically the same. Most wearable devices on the market measure heart rate by estimating pulse values. Wearable devices are affected by changes in sensor fit and deformation of soft tissues and blood vessels during exercise heart rate detection. Therefore, motion interference needs to be suppressed during heart rate measurement. The specific method for motion interference suppression is to obtain the spectral amplitude of the acceleration signal after removing the mean value in the corresponding time window, and use 1 / 2 of the spectral peak as the threshold for judgment. Peak frequencies greater than the threshold are considered to be the main frequencies. However, during the transition between exercise states, motion frequency components may be lost. Summary of the invention

[0005] The purpose of the present invention is to provide a method for accurately predicting individual heart rate in a field scenario to solve the problem of heart rate data loss during exercise in current wearable devices, especially during high-intensity exercise.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A method for accurately predicting individual heart rate in a field scenario, comprising:

[0008] The instantaneous speed of the target to be detected is estimated by using a speed estimation model to obtain the instantaneous speed of the target to be detected, wherein the speed estimation model is a relationship model between time and instantaneous speed;

[0009] Based on the instantaneous speed of the target to be detected, the heart rate of the target to be detected is predicted using a heart rate prediction model to obtain the heart rate of the target to be detected, wherein the heart rate prediction model is a relationship model between instantaneous speed and heart rate.

[0010] Optionally, constructing the speed estimation model includes:

[0011] Acquire a motion video of the detection target, sample the video at a preset sampling interval, and obtain the actual coordinates of the detection target at a certain number of samplings;

[0012] Based on the actual coordinates, calculate the instantaneous speed and obtain a time-instantaneous speed sequence;

[0013] The speed estimation model is obtained by performing piecewise fitting based on the time-instantaneous speed sequence using the least square method.

[0014] Optionally, obtaining the actual coordinates of the detection target includes: using a YOLOv5m model to perform human target detection on the target to be detected, visually marking the target to be detected, and obtaining the actual coordinates of the target to be detected.

[0015] Optionally, obtaining the actual coordinates of the target to be detected includes: mapping the actual coordinates to a top view by performing a projection transformation on the actual coordinates;

[0016] The method of mapping the actual coordinates to the top view is:

[0017]

[0018] Among them, (x, y) is the pixel coordinates, (x real ,y real ) is the actual coordinate, M is the number of pixels in the horizontal direction of the top view, N is the number of pixels in the vertical direction of the top view, m is the length of the target scene, and n is the width of the target scene.

[0019] Optionally, the heart rate prediction model is:

[0020]

[0021] in, is the heart rate at time t, C is a constant, indicating that the heart rate at time t is related to the previous tC, t-C+1, …, t-1 for a total of C times, f(v t-i ) is the long time at speed v t-i The stable heart rate achieved by running at a constant speed is min{C,t-1}, which is the minimum value between the moment C and the moment t-1.

[0022] Optionally, before using the heart rate prediction model to predict the heart rate of the target to be detected, the process includes: obtaining a number of discrete speed-heart rate point pairs, performing least squares fitting, obtaining a speed-heart rate function, comparing the speed-heart rate function with a true speed-heart rate function, and verifying the heart rate prediction model.

[0023] The beneficial effects of the present invention are as follows: The present invention aims to solve the problem of heart rate data loss during exercise in current wearable devices, especially during high-intensity exercise, through a novel non-contact heart rate monitoring technology. Traditional heart rate monitoring devices often require direct contact with the skin and are susceptible to frequent vibrations and sweat caused by exercise, resulting in unstable or lost heart rate data. The present invention utilizes the correlation between exercise speed and heart rate, combined with advanced algorithms to derive accurate exercise heart rate, thereby effectively avoiding the common problems of contact sensors.

[0024] The present invention adopts a method combining movement speed with a heart rate algorithm model. It overcomes the various inconveniences of previous heart rate measurement, thereby significantly improving the accuracy of heart rate monitoring. By introducing movement speed as a key parameter for calculating heart rate, the present invention has significantly improved the accuracy and stability of providing heart rate data. Compared with traditional technologies, it can provide heart rate data more stably and accurately at various exercise intensities, especially when the external environmental factors have a greater impact, such as outdoor running, cycling, etc.

[0025] The present invention not only enables users to obtain heart rate data more conveniently and quickly when exercising through this non-contact heart rate monitoring technology, but also encourages more people to participate in exercise. Because it is easy to use, more people may start to pay attention to and monitor their heart rate, thereby managing their health more effectively. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0027] Figure 1 This is a comparison chart of the accuracy and inference speed of each YOLOv5 model in an embodiment of the present invention;

[0028] Figure 2 Schematic diagram of projection transformation according to an embodiment of the present invention;

[0029] Figure 3 The actual speed, instantaneous speed, and estimated speed curves of the experimenter in the embodiment of the present invention;

[0030] Figure 4 The speed and heart rate curves of 10 randomly selected experimenters according to the embodiment of the present invention, wherein (a) is the first experimenter, (b) is the second experimenter, (c) is the third experimenter, (d) is the fourth experimenter, (e) is the fifth experimenter, (f) is the sixth experimenter, (g) is the seventh experimenter, (h) is the eighth experimenter, (i) is the ninth experimenter, and (j) is the tenth experimenter;

[0031] Figure 5 The point pairs and function images of the embodiment of the present invention;

[0032] Figure 6 Actual heart rate curves and predicted heart rate curves of 10 randomly selected subjects according to an embodiment of the present invention, wherein (a) is the first subject, (b) is the second subject, (c) is the third subject, (d) is the fourth subject, (e) is the fifth subject, (f) is the sixth subject, (g) is the seventh subject, (h) is the eighth subject, (i) is the ninth subject, and (j) is the tenth subject;

[0033] Figure 7 This is a flow chart of a method for accurately predicting individual heart rate in a localized scenario according to an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0036] like Figure 7 As shown, this embodiment provides a method for accurately predicting individual heart rate in a localized scenario, including:

[0037] The instantaneous speed of the target to be detected is estimated by using a speed estimation model to obtain the instantaneous speed of the target to be detected, wherein the speed estimation model is a relationship model between the time and the instantaneous speed;

[0038] Based on the instantaneous speed of the target to be detected, the heart rate of the target to be detected is predicted using a heart rate prediction model to obtain the heart rate of the target to be detected, wherein the heart rate prediction model is a relationship model between the instantaneous speed and the heart rate.

[0039] This embodiment adopts a method of combining movement speed with a heart rate algorithm model to overcome various inconveniences in previous heart rate measurement, thereby significantly improving the accuracy of heart rate monitoring.

[0040] Furthermore, constructing the speed estimation model includes: obtaining a motion video of the detection target, sampling the video at a preset sampling interval, and obtaining the actual coordinates of the detection target at a plurality of sampling times;

[0041] Based on the actual coordinates, the instantaneous velocity is calculated to obtain the time-instantaneous velocity sequence;

[0042] Based on the time-instantaneous speed series, the least square method is used to perform piecewise fitting to obtain the speed estimation model.

[0043] Furthermore, obtaining the actual coordinates of the detection target includes: using the YOLOv5m model to perform human target detection on the target to be detected, visually marking the target to be detected, and obtaining the actual coordinates of the target to be detected.

[0044] Specifically, the current YOLOv5 (You Only Look Once version 5) is a popular target detection model. It is one of the latest versions in the YOLO series and is widely used in the field of real-time target detection. The YOLOv5 official code provides a total of five pre-trained models of different sizes based on the COCO dataset, namely: YOLOv5n, YOLOv5s, YOLOv5m, YOLOv5l, and YOLOv5x. The number of parameters increases successively, so the recognition accuracy increases successively, and the reasoning speed decreases successively. This embodiment uses the YOLOv5m model with a relatively balanced accuracy and reasoning speed for testing. The specific relationship is as follows: Figure 1 shown.

[0045] The image captured by the camera is processed by the YOLOv5m model, and all detected human bodies are visually marked. The main targets in the image are selected for detection, and the midpoint P at the bottom of the marked box is taken to obtain the coordinates of the target to be detected, and the final visualization result is obtained.

[0046] Further, obtaining the actual coordinates of the target to be detected includes: mapping the actual coordinates to a top view by performing a projection transformation on the actual coordinates;

[0047] The method of mapping the actual coordinates to the top view is:

[0048]

[0049] Among them, (x, y) is the pixel coordinates, (x real ,y real ) is the actual coordinate, M is the number of pixels in the horizontal direction of the top view, N is the number of pixels in the vertical direction of the top view, m is the length of the target scene, and n is the width of the target scene.

[0050] In general, the projection transformation can be expressed as follows:

[0051]

[0052] Among them, (x, y, z) are the homogeneous coordinates of the original image pixel points, is the homogeneous coordinate of the image pixel after transformation.

[0053] To determine the values ​​of each coefficient in the projection matrix, it is necessary to give four pairs of vertices (A, A′), (B, B′), (C, C′), (D, D′) in the original plane and the projection plane, which represent the mapping relationship between the original plane and the projection plane, such as Figure 2 shown.

[0054] In this embodiment, the resolution of the top view is 750px×700px, and the size of a half-court basketball court is 15m×14m, so the following conversion is performed:

[0055]

[0056] Among them, (x, y) is the pixel coordinates, (x real ,y real ) are the actual coordinates.

[0057] In this embodiment, sampling is performed every Δt=1s. The pixel coordinates of the target at the kth sampling of the video are (x k ,y k ), then at this time t k = k·Δt, at this time the instantaneous velocity v K It can be approximately expressed as:

[0058] v1=0

[0059]

[0060] The idea of ​​least squares method is used for fitting, and the speed estimation model, namely model a, is obtained. The idea of ​​least squares method is that for an unknown continuous function f(x), according to the given n+1 time-instantaneous speed discrete points Find a low-degree polynomial function P(x) such that:

[0061]

[0062] It can be considered that P(x) can fit f(x) within a certain range, and P(x k ) is P(x) at x=x k The value of f(x k ) is f(x) at x=x k The verification of the fitting effect includes:

[0063] Data was collected from 30 subjects. A video of a subject running around a basketball court at a constant speed of 3 m / s for a period of time was randomly selected for display, and the real speed curve was recorded by a sports watch. Through target detection, data processing and other steps, the subject's coordinate-time series {(x1, y1), (x2, y2), ..., (x t ,y t )}. According to the instantaneous speed v K The calculation formula can be used to obtain the estimated time-instantaneous speed sequence And perform piecewise least squares fitting to fit into a continuous time-velocity function v * (t), draw the image as Figure 3 shown.

[0064] By reading the data from the sports watch, we can get the actual running speed function v(t). To evaluate the estimated speed function v * The accuracy of (t) is defined as the average error and the maximum error:

[0065] 1. Average error:

[0066]

[0067] The average error represents the estimated speed function v *The average deviation of the estimated velocity function from the actual velocity function v(t) over the entire time interval. The calculation results show that the average error is about 0.079. This means that the average deviation between the estimated velocity function and the actual velocity is small over the entire observation time, indicating that the method has a high accuracy overall.

[0068] 2. Maximum error:

[0069]

[0070] The maximum error represents the maximum deviation between the estimated speed function and the actual speed function at all observation time points. The calculation results show that the maximum error is about 0.657. This means that at certain specific moments, there is a large instantaneous deviation between the estimated speed and the actual speed, but this deviation does not affect the overall estimation accuracy.

[0071] Furthermore, the heart rate prediction model, model b, is:

[0072]

[0073] in, is the heart rate at time t, C is a constant, indicating that the heart rate at time t is related to the previous tC, t-C+1, ..., t-1, a total of C times, f(v t-i ) is the long time at speed v t-i Constant speed running is the stable heart rate reached, i is the time before time t. t-i ) is obtained through the mapping function f(v) from instantaneous speed to heart rate. The mapping function from instantaneous speed to heart rate is fitted by the least squares method, and min{C, t-1} is the minimum value between moment C and moment t-1.

[0074] Specifically, the verification of the heart rate prediction model includes:

[0075] Data were collected from 30 college students, and the data of 10 of them were randomly displayed. They were asked to run around the playground at a constant speed for a period of time, and then walk for a period of time, simulating daily running training and rest time. Their running speed and heart rate change curves were recorded through sports watches, heart rate belts and other equipment. Figure 4 As shown in (a)-(j).

[0076] By observing the speed and heart rate curve, we can find that the heart rate and speed change are generally positively correlated, that is, the heart rate increases with speed increase, and the heart rate decreases with speed slowdown. In addition, it is noted that the change of heart rate is a relatively slow process and will not change instantly with a sudden change in speed, so the influence of time factors still needs to be considered.

[0077] Assume the subject's resting heart rate is B restThe highest heart rate that can be achieved during aerobic exercise is B max , the speed and heart rate at the test time t are recorded as v t , The following modeling is available:

[0078]

[0079] Where f(v) is a mapping function from speed to heart rate, which represents the stable heart rate reached when running at a constant speed v for a long time, and min{C, t-1}+1 represents the length of the historical time period used to calculate the current heart rate. It satisfies:

[0080]

[0081] C is a constant, indicating that the heart rate at time t (t>C) is related to the previous tC, t-C+1, ..., t-1, a total of C times. This model indicates that the heart rate at time t can be estimated as the average of the stable heart rate corresponding to the speed from the min{C, t-1} time before the current time to the current time.

[0082] Furthermore, before using the heart rate prediction model to predict the heart rate of the target to be detected, the steps include: obtaining a number of discrete speed-heart rate point pairs, performing least squares fitting, obtaining a speed-heart rate function, comparing the speed-heart rate function with the actual speed-heart rate function, and verifying the heart rate prediction model.

[0083] Specifically, in this embodiment, sampling is selected every Δt=1s, C=100, f(v) is a univariate fourth-order polynomial function, and considering that the speed of the 10 subjects in actual running does not exceed 5m / s, it can be estimated that f(5)=B max .

[0084] The resting heart rates of 30 subjects were recorded, as well as the stable heart rates reached after running at a constant speed of 1.5 m / s, 2 m / s, 3 m / s, 4 m / s, and 5 m / s for 5 minutes. Finally, the average values ​​of the 10 subjects were calculated to obtain five discrete speed-heart rate point pairs: Substituting it into f(v) for least squares fitting, we can get:

[0085] f(v)=0.394v 4 -5.435v 3 +22.058v 2 -3.760v+90.122, the point pair and function graph are as follows Figure 5 As shown, the black points are speed and heart rate point pairs, and the curve is the fitting function.

[0086] Next, 30 subjects were asked to perform a running test with a time-speed sequence {v1, v2, …, v t Substitute it into the heart rate prediction model for calculation, and you can get the predicted time-heart rate sequence Draw it as an image, where the dotted line represents the interval where the heart rate tends to be stable after running for a period of time, such as Figure 6 As shown in (a)-(j).

[0087] The time-heart rate series predicted by the model Compared with the time-heart rate sequence measured by the device t} to verify the accuracy of the model estimation.

[0088] The average error and maximum error are defined as:

[0089]

[0090] For the overall data, the average error and maximum error of 10 experimenters were calculated, and the results are shown in Table 1:

[0091] Table 1

[0092]

[0093] For the data within the stable heart rate range, the average error and maximum error of the 10 subjects were calculated. The results are shown in Table 2:

[0094] Table 2

[0095]

[0096] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for accurately predicting individual heart rate in a localized scenario, characterized in that: include: The instantaneous speed of the target to be detected is estimated by using a speed estimation model to obtain the instantaneous speed of the target to be detected, wherein the speed estimation model is a relationship model between time and instantaneous speed; Based on the instantaneous speed of the target to be detected, the heart rate of the target to be detected is predicted using a heart rate prediction model to obtain the heart rate of the target to be detected, wherein the heart rate prediction model is a relationship model between instantaneous speed and heart rate.

2. The accurate prediction method of individual heart rate in a localized scenario according to claim 1, characterized in that: Constructing the speed estimation model includes: Acquire a motion video of the detection target, sample the video at a preset sampling interval, and obtain the actual coordinates of the detection target at a certain number of samplings; Based on the actual coordinates, calculate the instantaneous speed and obtain a time-instantaneous speed sequence; The speed estimation model is obtained by performing piecewise fitting based on the time-instantaneous speed sequence using the least square method.

3. The accurate prediction method of individual heart rate in a localized scenario according to claim 2, characterized in that: Acquiring the actual coordinates of the detection target includes: using the YOLOv5m model to perform human target detection on the target to be detected, visually marking the target to be detected, and acquiring the actual coordinates of the target to be detected.

4. The accurate prediction method of individual heart rate in a localized scenario according to claim 2, characterized in that: After obtaining the actual coordinates of the target to be detected, the method includes: mapping the actual coordinates to a top view by performing a projection transformation on the actual coordinates; The method of mapping the actual coordinates to the top view is: Among them, (x, y) is the pixel coordinates, (x real ,y real ) is the actual coordinate, M is the number of pixels in the horizontal direction of the top view, N is the number of pixels in the vertical direction of the top view, m is the length of the target scene, and n is the width of the target scene.

5. The accurate prediction method of individual heart rate in a localized scenario according to claim 1, characterized in that: The heart rate prediction model is: in, is the heart rate at time t, C is a constant, indicating that the heart rate at time t is related to the previous tC, t-C+1, …, t-1 for a total of C times, f(v t-i ) is the long time at speed v t-i The stable heart rate achieved by running at a constant speed is min{C,t-1}, which is the minimum value between the moment C and the moment t-1.

6. The accurate prediction method of individual heart rate in a localized scenario according to claim 1, characterized in that: Before using the heart rate prediction model to predict the heart rate of the target to be detected, the process includes: obtaining a number of discrete speed-heart rate point pairs, performing least squares fitting, obtaining a speed-heart rate function, comparing the speed-heart rate function with the actual speed-heart rate function, and verifying the heart rate prediction model.

Citation Information

Patent Citations

  • Forward-looking sonar image target detection method and system based on YOLOv5

    CN114677602A

  • Exercise heart rate prediction method based on time convolution network

    CN116167412A

  • Moving target speed estimation method

    CN116298372A

  • Method and system for determining the fitness index of a person

    EP2696766A1

  • Method for determining an instant velocity of a user and for improving estimation of heart rate

    US20150238146A1

Cited By

  • Running machine heart rate self-adaptive control method and system

    CN122018585A

  • Treadmill heart rate adaptive control method and system

    CN122018585B