TWI626072B - Image-based heart rate feedback control fitness device - Google Patents

Abstract

The invention comprises a motion unit, an image capturing unit, a temperature sensing unit and a control unit. The image capturing unit captures an instant image of the user moving in the motion unit, and obtains an instant heart rhythm signal through operation. The temperature sensing unit instantly captures the temperature signal; the heart rate signal and the temperature signal are used to convert the amount of motion of the user. The control unit controls the motion unit to match the amount of motion of the user with the preset amount of target motion. When the control unit compares and obtains at least one of the heart rate signal greater than a heart rate threshold and the temperature signal greater than a temperature threshold, the control motion unit reduces the amount of motion of the user to reduce the damage caused by excessive exercise. Therefore, this case can improve the exercise effect and actively improve the safety of sports, and has the advantages of safety protection mechanism to reduce secondary injury.

Description

Imaging rhythm control exercise training machine

The invention relates to an image rhythm control exercise training machine, in particular to an image rhythm control exercise training machine which can improve the exercise effect and has an active safety protection mechanism to reduce secondary injury.

In 2010, MIT pioneered the development of non-contact heart rate detection technology to capture the specific area of the face through the camera to find out the heart rate of the subject in the "non-sports" state. However, the study can only be detected under static conditions. The main cause is the influence of light, physiological condition, vibration caused by exercise, etc., which makes the physiological signal in the image more complicated and difficult.

At present, the rhythm measurement technique used on the physical exercise machine utilizes the transmission or reflection of infrared rays to sense the heart rhythm, and there are roughly two types of contact measurement tools. One is a wristband or a heartbeat belt worn by the user, and the other is a grip with a metal electrode attached to the physical training machine. While running, the hand is held on the grip and the electrode conducts current to measure Heart rate. Both of the above devices cause physical burden and inconvenience to the user, and the flexion is limited to the heart rate measuring device, which increases the restriction on the posture and rhythm of the user, and does not conform to the ergonomics during actual exercise. In addition, the safety protection mechanism of the traditional physical exercise machine is passive, lacking the function of predicting the physiological condition of the user and real-time active safety protection.

In view of this, it is necessary to develop a technique that can solve the above disadvantages.

The object of the present invention is to provide an image rhythm control exercise training machine, which can improve the safety of exercise, meet the ergonomics during actual exercise, and have the advantages of safety protection prediction mechanism to reduce secondary injury. In particular, the problem to be solved by the present invention is that the conventional treadmill cannot pass through Dynamic control to match the user's own set of exercise intensity, unable to detect that the user's physical fitness is about to run out and automatically reduce the exercise intensity, and the safety mechanism of the traditional treadmill when the user leaves or the body is uncomfortable and falls down. Lack of automatic detection and active protection, causing problems such as the user falling down and being injured by the running treadmill.

The technical means for solving the above problems is to provide an image-based rhythm-controlled exercise training machine, comprising: a motion unit for a user to exercise, the user has a face; an image capture unit is disposed in the The motion unit is configured to capture a facial image of the face; the facial image has a red pixel signal, a green pixel signal, and a blue pixel signal; and a temperature sensing unit is disposed in the motion The unit is configured to perform surface temperature sensing on the face and obtain a temperature signal; a control unit is disposed on the motion unit, and is coupled to the image capturing unit and the temperature sensing unit, the control The unit has a heart rate threshold and a temperature threshold; when the red pixel signal, the green pixel signal and the blue pixel signal are received, the Fourier transform is used to obtain the largest energy spectrum among the signals. And its harmonics, which are heart rhythm signals of the user, the heart rate signal and the temperature signal are used to convert the amount of exercise of the user, and the control unit is used to control the exercise ticket And the amount of movement of the user is matched with the preset target exercise amount; and when the control unit compares and obtains that the heart rhythm signal is greater than the heart rate threshold, and the temperature signal is greater than the temperature threshold, the system automatically controls The motion unit reduces the amount of exercise of the user and reduces the damage caused by excessive movement of the user.

The above objects and advantages of the present invention will be readily understood from the following detailed description of the preferred embodiments illustrated herein.

The invention will be described in detail in the following examples in conjunction with the drawings:

10‧‧‧ sports unit

20‧‧‧Image capture unit

30‧‧‧Temperature sensing unit

40‧‧‧Control unit

40A‧‧‧heart rate threshold

40B‧‧‧temperature threshold

41‧‧‧Touch display structure

71‧‧‧Sports machine parameter control unit

72‧‧‧sports machine

73‧‧ ‧ athletes

74‧‧‧ Physiological signal sensor

81‧‧‧ image data

82‧‧‧Face area detection and tracking

83‧‧‧RGB three-channel color value average

84‧‧‧Color space projection

85‧‧‧Bandpass filtering

86‧‧‧ Fourier transform

87‧‧‧Extracting the peak frequency and the second harmonic frequency

88‧‧‧Reconstruction of heart rate signals

90‧‧‧Face

90A‧‧‧ heart rate signal

90B‧‧‧temperature signal

90C‧‧‧ energy consumption value

91‧‧‧Face images

911‧‧‧Face area

912‧‧‧Feature points

91A‧‧‧Red pixel signal

91B‧‧‧Green pixel signal

91C‧‧‧Blue pixel signal

92A‧‧‧peak frequency

92B‧‧‧second harmonic frequency

L1‧‧‧ initial training curve

L2‧‧‧ training late curve

Figure 1 is a schematic view of an embodiment of the present invention

2 is a schematic view of a touch display structure of the present invention

Figure 3A is a physical photograph of the facial image of the present invention.

Figure 3B is a schematic diagram of facial image changes in the motion of the present invention.

Figure 4 is a reference flow chart of the heart rate calculation process of the present invention

Figure 5 is a waveform diagram of RGB three-channel color value averaging according to the present invention.

Figure 6 is a schematic diagram of the heart rhythm signal (Fourier transform) of the present invention

Figure 7 is a block diagram of the closed loop control architecture of the present invention.

Figure 8A is a graph of the initial training of the present invention

Figure 8B is a graph of the training later in the present invention.

Referring to Figures 1 and 2, the present invention is an image-based rhythm-controlled exercise training machine comprising: a motion unit 10 for exercising by a user having a face 90.

An image capturing unit 20 is disposed on the moving unit 10 for capturing a facial image 91 of the face 90 (see FIGS. 3A and 3B); the facial image 91 has a red pixel Signal 91A, a green pixel signal 91B and a blue pixel signal 91C (as shown in Figure 5).

A temperature sensing unit 30 is disposed on the motion unit 10 for performing body surface temperature sensing on the face 90 and obtaining a temperature signal 90B.

A control unit 40 is disposed on the motion unit 10 and coupled to the image capturing unit 20 and the temperature sensing unit 30. The control unit 40 has a heart rate threshold 40A and a temperature threshold 40B. When the red pixel signal 91A, the green pixel signal 91B, and the blue pixel signal 91C are received, the Fourier transform is used to obtain the largest energy spectrum and its harmonics in the signal, which is the user's heart rate signal 90A. The heart rate signal 90A and the temperature signal 90B are used to convert the amount of motion of the user, and the control unit 40 is used to control the motion unit 10 to make the movement amount of the user and its preset purpose. The target motion amount is consistent; and when the control unit 40 compares the heart rate signal 90A to be greater than the heart rate threshold 40A and the temperature signal 90B is greater than the temperature threshold 40B, the motion unit 10 is automatically controlled to reduce the usage. The amount of exercise, which can reduce the damage caused by excessive exercise of the user.

The exercise unit 10 can be at least one of a treadmill, a fitness device, or other exercise trainer.

The image capturing unit 20 can be at least one of a photosensitive coupled device (CCD) and a complementary metal-oxide semiconductor (CMOS).

The temperature sensing unit 30 can be an infrared temperature sensing device (known in the art).

The control unit 40 can further include a touch display structure 41 for touch-operating the image rhythm-controlled exercise training machine, and for displaying the facial image 91, the heart rhythm signal 90A, the temperature signal 90B and an energy The consumption value is 90C.

The touch display structure 41 further displays a running image of a virtual reality (for example, a plurality of people performing a running competition at the same time), and can improve the variability of the exercise.

The focus of the present invention is on image heart rate signal acquisition and body temperature detection to convert user energy consumption.

The image capturing unit 20 and the temperature sensing unit 30 are disposed on the touch display structure 41. The touch display structure 41 can be rotated and adjusted by the control unit 40 according to requirements, so that the image capturing unit 20 is The preferred image capture angle and the temperature sensing unit 30 has a preferred body temperature sensing angle.

About image heart rate signal acquisition: The purpose of this algorithm is to effectively extract the user's heart rate signal during exercise. A major advantage of this algorithm over prior art is the reliable extraction of heart rhythms from complex motion signals in a non-contact manner. During the movement of the motion unit 10, the image capturing unit 20 captures (ie, the image data 81 shown in FIG. 4) the facial image 91 of the face 90 of the user. Temperature sensing unit 30 temperature signal 90B, and then according to the flow chart shown in Figure 4, the calculation, divided into seven main steps, explained below one by one.

[1] Face Area Detection and Tracking 82. At the initial frame where the video starts, the face detector 911 is used to detect the face area 911 of the user, and the feature point 912 is extracted in the face area 911 (as shown in FIGS. 3A and 3B). In the next face image 91, only a plurality of feature points 912 are extracted, and then a coordinate transformation matrix is estimated based on the coordinates of the feature points 912 of two consecutive frames, and the matrix is applied to the face region 911 of the previous frame. The coordinates can be used to get the coordinates of the face area of the current frame.

For example, when the resolution is 800x600, the camera will capture 30x4=120 frames of image data in 0-4 seconds, assuming that the top, bottom, left and right edges of the "face area 911" are The uppermost edge of the eyes, the upper edge of the lips, and the end of the eyes of the eyes are bounded, and the clear area and coordinate data of the "face area 911" can be determined. Of course, it can also be done by other similar conventional image processing techniques.

Further, regarding the "face area detection algorithm", it is a conventional technique. For example, most cameras give the position of a face (usually represented by a rectangular box) when taking a picture. The algorithm used in the present invention is a recently proposed algorithm (eg, reference: Bertinetto L, Valmadre J, Golodetz S, et al) .Staple: Complementary learners for real-time tracking, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2016: 1401-1409.).

Each of the foregoing feature points 912 may be a Speeded Up Robust Feature (SURF), using a determinant value of a Hessian matrix for feature point detection, and accelerating the operation by using an integral graph. It is mainly used for robust image recognition and description algorithms (known techniques published in 2006 are not mentioned).

[2] RGB three-channel color values take an average of 83. Extracting pixel values of three channels of RGB (ie, the red pixel signal 91A, the green pixel signal 91B, and the blue pixel signal 91C) in the face area 911 of each frame, and averaging all pixel values of each channel , get a three-dimensional vector.

[3] Color space projection 84. Mapping a three-dimensional vector of each frame to a point in RGB space will have several consecutive points within a time window. Principal components analysis (PCA) was used.

For example, if there are 120 points of R, G, and B in 0-4 seconds, three curves can be formed. In fact, the time difference can be estimated from points a and b in Figure 5. = (1.5-0.75) = 0.75 seconds, so its heart rate is about 1.33 Hz (that is, the heart rate is 80 / min).

The aforementioned principal component analysis is a technique for analyzing and simplifying data sets. Principal component analysis is often used to reduce the dimensionality of a data set while maintaining the features that contribute the most to the variance in the data set.

Principal component analysis is the easiest way to analyze multivariate statistical distributions by feature quantities. This kind of operation can be seen as a better way to expose the internal structure of the data and thus explain the variables of the data. If a multivariate data set can be visualized in a high dimensional data space coordinate system, principal component analysis can provide a lower dimensional image, which is the original object at the point where the message is most A "projection". This allows you to reduce the dimensions of your data with a small amount of principal components.

Principal component analysis calculates the first principal component of these points in space, projecting the coordinates of each point to this first principal component, thus forming a waveform that changes over time.

[4] Bandpass filtering 85. This waveform is bandpass filtered to remove noise from noise that is not related to heart rhythm.

For example, since the normal heart rate of a normal person is about 70/min to 90/min, noises higher than 180/min and below 35/min can be filtered first.

[5] Fourier transform 86. Perform a Fourier transform on the bandpass filtered signal.

[6] Extract the peak frequency and the second harmonic frequency 87. According to research, the heart rhythm signal contains two high-frequency signals, one is the main frequency corresponding to the heart rhythm (ie, the peak frequency is 92A), and the other is twice the main frequency, called the second harmonic (ie, the second time). Harmonic frequency 92B). We determine the frequency with the largest energy as the main frequency of the heart rhythm according to the signal after the Fourier transform, and the frequency of the second harmonic is the second harmonic frequency.

For example, as shown in Fig. 6, its main frequency (i.e., peak frequency 92A) is 1.38 Hz, that is, heart rate = 82.8 / min. Of course, it can be verified by the calculation of the second harmonic frequency 92B: 2.7/2 (times) = 1.35, which is very close to 1.38.

[7] Heart Rhythm Signal Reconstruction 88. According to the main frequency and the second harmonic frequency, the heart rate signal 90A can be obtained by inverse Fourier transform.

That is, the heart rate = 82.8 / min.

Similarly, in the case of 0-4 seconds, there are 120 frames, and the heart rate can be estimated to be 82.8/min. Of course, the time can be lengthened or shortened depending on the situation.

Regarding the body temperature detection, the temperature sensing unit 30 performs temperature sensing on the face area 911, and the temperature signal 90B is obtained by the control unit 40. The formula for calculating the body surface temperature is as follows: T = T _s + T _c ; where: T _{s is} the temperature measured by the temperature sensing unit 30; T _c is the environmental compensation temperature, and the calculation formula is as follows: T _c = α × D ^-1 + β ; wherein: α and β are constants associated with the temperature sensing unit 30, and D is the distance of the target from the lens.

The calculation formula for the energy consumption value 90C is as follows (the following parameters can be adjusted according to actual conditions): Male: (1+sign( T - T _μ )× )(-55.0969+0.6309× f ( HR ^' < HR _μ )+0.1988× W +0.2017× A )/4.184×60(1)

Female: (1+sign( T - T _μ )× )(-20.4022+0.4472× f ( HR ^' < HR _μ )-0.1263× W +0.074× A )/4184×60(2)

Where W = weight (kg), A = age (years).

and ; among them: Is the rate of change of heart rhythm with time; T ₀ is the sampling period; HR _μ is the upper limit of heart rate change rate; HR _m = HR (( k -1) T ) is the heart rate value of the previous moment; sign( ) is the switching function; T _μ The body temperature constant for the preset movement.

In equations (1) and (2), the original heart rhythm is replaced by a function that uses the current heart rhythm when the rate of heart rate change is less than a certain threshold (valve) value; when the rate of heart rate change is greater than this threshold (valve) value When using the heart rate of the previous moment. Since the rate of change of the human heart rhythm is very slow under normal conditions, when the heart rate changes drastically, it is likely to be caused by external stimuli and does not represent an increase in the amount of exercise. In order to avoid the error of the exercise energy consumption calculation caused by this situation, we use the first few heart rate averages to calculate. In addition, the study found that for every 1 degree Celsius temperature increase, the energy consumption will increase by 10%, so we design (1+sign( T - T _μ )× This item takes into account the body surface temperature. The above two formulas are instantaneous sports energy consumption, and the total energy consumption of the exercise period can be obtained by integrating the above formula with time.

Further, referring to FIG. 7, the closed loop control architecture of the present invention includes a motion machine parameter control unit 71 (provided in the control unit 40), a motion machine 72 (eg, the motion unit 10), An exerciser 73 (such as a user of the present invention) and a physiological signal sensor 74 (for example, the image capturing unit 20 and the temperature sensing unit 30), wherein the exercise machine 72 and the athlete 73 are controlled objects . The physiological signal sensor 74 is a photographic lens in a touch-sensitive display screen, including a normal camera lens and an infrared sensing lens. The core of this closed loop control system is the motion parameter control unit 71. The at least one integrated circuit hardware structure is combined with the touchable display screen to form a separate device, or embedded in the motion In the calculation unit of machine 72. The physiological signal sensor 74 extracts the body energy consumption information of the heart rate, body temperature, and the like of the athlete 73, and compares it with the selected motion mode reference input as an input of the exercise machine parameter control unit 71. A control variable for controlling the voltage, current, and the like of the exercise machine 72 is generated. Thus, the exercise machine 72 is manipulated to generate ideal slope, belt speed, running time and other parameters to form a target exercise training mode. In the exercise training mode, the athlete 73 generates a physiological signal change, and the signals are transmitted back to the control unit 40 by the physiological signal sensor 74, and the control unit 40 adjusts according to the reference input, and then generates A corresponding output is used to control the exercise machine 72. A closed loop control architecture is formed to actively control the motion machine 72 to produce a specific training effect.

The present invention replaces the heart rate curve with an energy consumption curve, so that a training mode based on energy consumption control can be designed. Referring to Fig. 8A, the initial curve L1 of the training represents the target heart rhythm. At the beginning of the exercise, the heart rhythm is intermittently increased, and the exercise intensity is gradually increased. In the later stage of exercise (such as the training curve L2 shown in Figure 8B), the heart rate gradually decreases and gradually returns to a stable state. By measuring the actual heart rhythm of the user, the closed loop control architecture controls the exercise intensity of the exercise machine, so that the user's heart rate reaches the goal and achieves the training purpose. Through the feedback of the heart rhythm, the exercise machine knows when to enter the next stage. Compared with the preset program, this method can control the time of each stage more flexibly. In the case of coaching, the training time at each stage is usually determined by the coach's experience, or through feedback from the coach and the user. The proposed method is a direct feedback that does not require the coach to communicate with the user. It is also possible to design the coach's experience in the rhythm curve so that the training effect under the guidance of the coach can be achieved. In addition, the proposed method is more flexible and more suitable for the special situation of the individual user. Because each human body can be different, the rate of rise and recovery of heart rhythm is not the same, and a fixed time is not suitable for each user. For example, some people have entered the next stage without fully recovering, which will weaken the training effect; some people have fully recovered within the preset time, which will reduce the training efficiency. The proposed algorithm completely overcomes these shortcomings and achieves the goal of tailoring for each person.

The security protection mechanism of the present invention is as follows: the facial image 91 of the image capturing unit 20 is continuous shooting of the image, and the number of frames may be 30 frames per second or 60 frames, and the processing of the face recognition is performed. In the process, if the image capturing unit 20 captures a face target (ie, the face image 91) at the nth frame, and no face is captured after the n+1th frame, the control unit 40 operates. After processing, when the presence of the face target (ie, the face image 91) cannot be detected in the n+1th frame, it indicates that the user may fall or leave the motion unit 10, and then disappears in the observation screen. The control unit 40 will stop the operation of the motion unit 10 in a very short time, preventing the user from being injured twice.

In addition, the correlation between the body temperature and the heart rhythm is established by the acquisition of the body temperature signal and the heart rate signal to determine whether there is abnormal heart rhythm change:

[a] For example, when the exercise unit 10 is increasing the intensity, the heart rate signal 90A is declining, and the heart is swaying, and the temperature signal 90B is judged to be an abnormal heart rhythm drifting phenomenon.

[b] Further, for example, when the exercise unit 10 reduces the exercise intensity, the heart rate signal 90A is increased upward or palpitations.

[c] For example, when the temperature signal 90B of the user is sensed to exceed 40 degrees (the body center temperature does not exceed 40 degrees under normal exercise, otherwise there is a possibility of heat exhaustion).

When the above various conditions [a], [b], and [c] occur, the control unit 40 is used to send an alert command (which may be a combination of text, a pattern, and any alert that can be reached), and the touch display structure 41 is presented. And simultaneously sending a control command to lower the rotational speed of the motion unit 10 or adjust related settings to alleviate the physiological condition of the user.

When the exercise unit 10 is a treadmill, the training mode can be determined after the treadmill is started (the treadmill first records the physiological data of the user under normal physical fitness), and then the image capturing unit 20 and the temperature sense are obtained. The measuring unit 30 is adjusted to detect the position of the face 90. The touch display structure 41 can present various physiological messages, treadmill information and function menus.

Then, the control unit 40 performs tracking of the plurality of feature points 912 in the face area 911, and obtains three different pixel signals of red, green, and blue for image analysis, and smoothes the obtained three kinds of pixel signals, and then applies the tape. The pass filter removes the noise, and the remaining signal is subjected to Fast Fourier Transform to extract the largest energy spectrum and its harmonics in the signal, which is the detected heart rate signal. On the body temperature detection, the temperature sensing unit 30 converts the color corresponding to the infrared light of the specific area of the face (ie, the face area 911), and then estimates the temperature to which the color is close, which is the detected temperature. Signal 90B. The heart rate, body temperature and exercise time are combined to analyze the corresponding energy consumption value 90C through the relevant physiological data.

In the past treadmills, after the user selected the training mode before the exercise, the machine sequentially controlled the motor speed and the slope according to the training mode, and its operation was completely independent of the user's own immediate physiological state.

Different from the conventional treadmill, the special feature of the present invention is that after the user selects the training mode, the control unit 40 (through the closed loop control mechanism) controls the training algorithm according to the heart rhythm, and performs cyclic training on the physiological conditions of different users. First, set the upper and lower limits of the heart rhythm in motion. For example, the upper limit of the rhythm is set 180 times per minute, and the lower limit of the heart rhythm is set 90 times per minute. When the treadmill is just starting, the heart rhythm of the user is still less than the lower limit. Then, the control unit 40 sends a control command to the treadmill, increases the treadmill rotation speed or the gradient to increase the exercise intensity, raises the user's heart rhythm to the upper limit, and when the detected heart rate has reached the upper limit, the control unit 40 controls the treadmill. Start to reduce the speed or the slope of the runway, and use the high and low intensity cycle changes to achieve effective training.

In the security detection process, if the system cannot detect the user's face during the process of face recognition, it may indicate that the user falls or leaves the treadmill, and the control unit immediately stops the operation of the treadmill to prevent the user from being subjected to the second The secondary injury can also be compared with the comprehensive data of body temperature and heart rhythm to determine whether there is abnormal physiological condition. For example, when the treadmill is increasing the intensity, the heart rate value (signal) is decreased, or when running When the machine reduces the intensity, the heart rate value (signal) increases, and the control unit sends a warning command to prompt the user to ease the exercise intensity on the screen display unit, or by the main The dynamic adjustment mechanism (which can be set by the user to set the active intervention authority of the controller) sends a control command to lower the rotation speed of the treadmill, etc., to alleviate the physiological condition of the user.

The advantages and functions of the present invention are as follows:

[1] can improve the safety of exercise. The invention integrates the heart rhythm and the body surface temperature in the course of exercise, and once the data exceeds the threshold of the training mode, the strength of the treadmill is reduced, and the physiological function of the user is alleviated, so as to avoid sports injury. Therefore, the safety of exercise can be improved.

[2] can improve the exercise effect. In the process of the invention, the heart rhythm and the body surface temperature are integrated in real time, and when the data does not exceed the threshold of the training mode, the exercise intensity generated by the treadmill is increased, and the amount of exercise of the user is automatically increased. Therefore, the exercise effect can be improved.

[3] It has a safety protection mechanism to reduce secondary damage. The image capturing unit of the present invention performs instant continuous shooting on the face, and the number of frames may be 30 frames per second or 60 frames per second, and in the process of face recognition, if the image capturing unit is in the first When the n frame is photographed to the face, and after the n+1th frame, no face is photographed, indicating that the user may fall or leave the motion unit, the control unit stops the motion unit in a very short time. Operate to prevent users from being harmed twice.

The present invention has been described in detail with reference to the preferred embodiments of the present invention, without departing from the spirit and scope of the invention.

Claims (7)

An image-based rhythm-controlled exercise training machine includes: a motion unit for a user to move, the user has a face; an image capture unit is disposed on the motion unit for The face captures a face image; the face image has a red pixel signal, a green pixel signal, and a blue pixel signal; a temperature sensing unit is disposed on the motion unit for facing the face The part performs temperature sensing on the body surface and obtains a temperature signal; a control unit is disposed on the motion unit, and is coupled to the image capturing unit and the temperature sensing unit, wherein the control unit has a built-in heart rate threshold and a temperature threshold; when the red pixel signal, the green pixel signal, and the blue pixel signal are received, the Fourier transform is used to obtain the largest energy spectrum and its harmonics in the signal, which is the user's heart rhythm The heartbeat signal and the temperature signal are used to convert the amount of exercise of the user, and the control unit is configured to control the motion unit to move the user's movement amount with the preset target motion. When the control unit compares and obtains that the heart rhythm signal is greater than the heart rate threshold, and the temperature signal is greater than the temperature threshold, automatically controlling the motion unit to reduce the amount of motion of the user, and reducing the The damage caused by excessive exercise by the user. The image-based rhythm-controlled exercise training machine according to claim 1, wherein the exercise unit is at least one of a treadmill and a fitness equipment. The image-based rhythm-controlled exercise training machine according to claim 1, wherein the image capturing unit is at least one of a photosensitive coupling element and a complementary metal oxide semiconductor. The image-based rhythm-controlled exercise training machine according to claim 1, wherein the temperature sensing unit is an infrared temperature sensing device. The image-based rhythm-controlled exercise training machine of claim 1, wherein the control unit further comprises: a touch display structure for touching and operating the image-based rhythm-controlled exercise training machine, and for displaying The facial image, the heart rhythm signal, the temperature signal, and an energy consumption value. The image-based rhythm-controlled exercise training machine according to claim 5, wherein the touch display structure is used to display a running image of a virtual reality, and the variability of the exercise can be improved. The image-based rhythm-controlled exercise training machine according to claim 5, wherein: the exercise unit increases the exercise intensity of the user, and the heart rate signal is decreased, one of the heartbeats, and the temperature signal is matched When it is determined that the abnormal heart rhythm drifts, the control unit is configured to send a warning command, which is displayed on the touch display structure, and the control unit automatically controls the motion unit to reduce the amount of movement of the user to achieve safety. a protection mechanism; and when the motion unit reduces the exercise intensity of the user, and the heart rate signal is rising or one of the heart, the control unit is configured to send a warning command, which is displayed on the touch display The structure, and the control unit automatically controls the motion unit to reduce the amount of movement of the user and achieve a safety protection mechanism.