System and method for detecting chest compression depth, rate and posture based on binocular vision

Abstract

The application discloses a chest compression depth and frequency and posture detection system and method based on binocular vision, which comprises an identification system and a detection method, wherein the identification system comprises a binocular camera, a checkerboard wristband and a computer, and the detection method comprises a compression depth and frequency detection method and a compression posture detection method. The compression depth and frequency detection method is based on target detection and color region segmentation technology, extracts the center point of the checkerboard mark of the checkerboard wristband worn on the wrist of a rescuer as a binocular matching point, and calculates the coordinates of the center point in a camera coordinate system, so that the real-time calculation of the compression depth and frequency is realized; the compression posture detection method evaluates the compliance of the chest compression posture through human key point detection technology. The system provides a non-contact real-time evaluation method for the quality of chest compression of cardiopulmonary resuscitation, does not need to install any tool between the hands of rescuers and the chest of a patient, does not affect the compression quality, and can be used for cardiopulmonary resuscitation training and clinical assistance.

Description

Technical Field

[0001] This invention relates to the interdisciplinary fields of smart healthcare, artificial intelligence, instrument science, and computer science, specifically to a system and method for detecting chest compression depth, frequency, and posture based on binocular vision. Background Technology

[0002] In today's fast-paced society, people face numerous pressures from work and life, leading to frequent cases of sudden cardiac arrest due to physical or psychological issues. In such situations, timely cardiopulmonary resuscitation (CPR) is crucial. CPR is a key survival technique used in emergency care to stimulate breathing and maintain blood flow to the heart. To achieve high-quality CPR, medical personnel need regular training in basic life support. However, due to the complexity of CPR, medical workers may make mistakes during the procedure, which can worsen the patient's condition or even lead to death. Therefore, accurate monitoring of CPR procedures, especially chest compression depth (CCD) and chest compression rate (CCR), is essential for providing effective first aid. To ensure the effectiveness of chest compressions, on the one hand, according to the requirements of the European Committee for Cardiopulmonary Resuscitation (ECPR), the CCD should be maintained at 50mm-60mm, and the CCR should reach 100-120 compressions / minute. This requires the cardiopulmonary resuscitation (CPR) behavior monitoring system to have good real-time performance; otherwise, it may introduce large errors due to missing peak and trough detection. On the other hand, compliance with the Chest Compression Posture (CCP) also affects the effectiveness of chest compressions. During chest compressions, the operator's elbows should be straight and the compression direction should be perpendicular to the patient's chest cavity. Based on the above analysis, a real-time CPR behavior monitoring device with good real-time performance and measurement accuracy, and capable of evaluating CCP, is extremely important. Chinese patent CN104486989A proposes a device to assist rescuers in performing CPR. It monitors CPR quality parameters by capturing images through electronic medical devices and cameras, and provides auditory, visual, or tactile cues. It mainly determines whether to change rescuers by monitoring compression depth and frequency, but it lacks monitoring of compression posture. Chinese patent CN115586834A proposes an intelligent cardiopulmonary resuscitation (CPR) training system. It assesses compression movements using visually acquired skeletal points and combines this with a multi-sensor mannequin to detect hand touch, touch location, artificial respiration, compression frequency, and compression depth to ensure compliance. However, the mannequin's sophisticated design limits its widespread application in public health education and training for professional medical personnel. Chinese patent CN117115255A proposes a 3D human key point detection method based on an RGB-D camera. This method accurately identifies and locates 3D human key point coordinates, such as those of the head, arms, and legs, through 2D human pose estimation and depth images from the camera. This helps rescuers and rescue robots quickly assess the injured person's condition and improves rescue efficiency. However, calculating 3D human key points using depth images consumes significant computational resources, resulting in limited accuracy in compression depth detection.Chinese patent CN202110713123.X proposes a machine vision-based cardiopulmonary resuscitation (CPR) compression depth feedback system. This system overcomes the inconvenience of connecting cables by using independently designed feedback pads and a main unit, enabling CCD measurement and feedback. However, it lacks the ability to measure compression parameters such as CCR and CCP. This patent, based on binocular vision technology, provides a non-contact measurement solution that can detect CCD, CCR, and CCP without requiring any tools between the rescuer's hand and the patient's chest. It is convenient to use and has high application value in medical human-computer interaction. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention proposes a system and method for detecting chest compression depth, frequency, and posture based on binocular vision. This system provides a non-contact method for real-time assessment of the quality of chest compressions during cardiopulmonary resuscitation (CPR). It does not require any tools to be installed between the rescuer's hand and the patient's chest, does not affect the compression quality, and can be used for CPR training and clinical assistance.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] A binocular vision-based chest compression depth, frequency, and posture detection system includes a binocular camera, a checkerboard wristband, and a computer. The binocular camera is capable of binocular visual localization and object recognition. The checkerboard wristband includes a rectangular elastic band and a checkerboard pattern affixed to its center surface. The checkerboard pattern is called a MARK and is worn on the rescuer's wrist or other location related to compression depth. The computer is connected to the binocular camera, receives data from the camera, and calculates compression depth, frequency, and posture in real time based on the detection methods described above. The detection methods include compression depth and frequency detection methods and compression posture detection methods.

[0006] This invention provides a method for detecting chest compression depth, frequency, and posture based on binocular vision, comprising the following steps:

[0007] Step 1: Image acquisition, distortion removal, and stereo correction; MARK center point coordinate calculation;

[0008] Step 2: Detect the depth, frequency, and posture of chest compressions based on the corrected images;

[0009] Step 3: Chest compression posture detection and classification;

[0010] During chest compressions, a human key point detection algorithm locates key points on the shoulder, elbow, and wrist to assess whether the elbow is straight and whether the compression direction is perpendicular to the patient's chest cavity. These parameters are calculated by taking the angle α between the line connecting the shoulder and elbow and the line connecting the elbow and wrist, and the angle β between the line connecting the shoulder and wrist and the vertical direction. When calculating the compression posture based on the acquired human key points, if ∠α is in the range of [165°, 180°], it is considered that the elbow is straight during compression. If ∠β is in the range of [0°, 10°], the compression direction is identified as vertical. Based on the set angle range, four different compression postures are classified: Elbow (elbow not straight, ∠α does not meet the requirement); Right (compression is correct, ∠α and ∠β both meet the requirement); Vertical (compression is not vertical, ∠β does not meet the requirement); and Wrong (elbow not straight, compression is not vertical, ∠α and ∠β both do not meet the requirement).

[0011] Furthermore, the specific steps of step 1 are as follows;

[0012] Step 1-1, Camera Calibration: Before CCD measurement, calibrate the stereo cameras to obtain the intrinsic parameters K and distortion parameter distCoeff of both cameras, as well as the extrinsic parameters R and T of the right camera relative to the left camera.

[0013]

[0014] distCoeff=[k1 k2 p1 p2 k3] (2)

[0015]

[0016] T = [t1 t2 t3] (4)

[0017] Among them, c x and c y The coordinates of the principal point are in pixels; if we assume f is the focal length of the camera, then f x =α·f, f y =β·f, where α and β are the width and height of the two-dimensional image, i.e., the size of the pixels;

[0018] Steps 1-2, Distortion Removal and Stereo Correction: The radial and tangential distortions of the camera are as follows:

[0019]

[0020]

[0021] Where k1, k2, and k3 represent the coefficients of radial distortion, r 2 =x 2+y2, p1, p2 represent the coefficients of tangential distortion, x corrected and y corrected Given the positions of x and y in the distorted image, and combining radial and tangential distortion, we only consider the first three terms of the radial distortion coefficient and the two terms of the tangential distortion coefficient to obtain a comprehensive distortion correction formula.

[0022]

[0023] The correct position of a point in the pixel plane can be determined using five distortion coefficients k1, k2, k3, p1, and p2.

[0024] Steps 1-3, MARK center point coordinate calculation: Using the center point of the MARK marker as the matching point, the binocular matching point detection based on object detection and color region segmentation is used to obtain a matching point that allows the binocular camera to obtain an accurate matching point. Combining the triangulation model and the camera's geometric modeling model, the coordinates of the center point of the MARK marker in the camera coordinate system are calculated. After obtaining the position of the marker in the μOv pixel coordinate system based on object detection, this method uses color detection to identify the color region of the MARK and finds the center point of the MARK based on the distribution of different color regions.

[0025] Furthermore, the specific steps of step 2 are as follows;

[0026] Step 2-1, Chest Compression Depth and Frequency Detection: Based on target detection and color region segmentation, the center point of the MARK marker is extracted and used as the binocular matching point to calculate the compression depth and frequency. Assuming the imaging plane coordinate system is xO'y, the pixel coordinate system is μOν, and the camera coordinate system is OXYZ, the camera and the human body are placed horizontally on the ground. Using the camera geometry model and triangulation principles, the Y-axis coordinate change of the MARK marker center point in OXYZ is determined, and the CCD and CCR are obtained. The steps are as follows:

[0027] Step 2-1-1, Coordinates of the MARK Center in the Imaging Plane Coordinate System: Based on the camera's geometric model and the principle of triangulation, the two-dimensional pixel coordinates are restored to three-dimensional space through the new camera intrinsic and extrinsic parameters after distortion correction and stereo correction. The physical coordinates of the MARK center in the camera coordinate system OXYZ are calculated. Then, based on the μ coordinates of the MARK center in the left and right camera coordinate systems... L and μ R The coordinates x of the MARK center in the xO′y coordinate system of the imaging plane are calculated. The specific calculation method is Equation (8).

[0028]

[0029] Where f is the camera's true focal length, in millimeters. xand f y Focal length is measured in pixels, c x and c y The principal point coordinates are in pixels, f x f y c x and c y Both are derived from intrinsic parameters obtained from camera calibration, where α and β are the width and height of the two-dimensional image, i.e., the pixel size;

[0030] Step 2-1-2, Y-axis coordinate of the MARK center in the camera coordinate system: The x-axis coordinate of the MARK center in the xO′y coordinate system of the left and right camera imaging planes is obtained. L and x R After determining the coordinates, the disparity dis = x is calculated. L -x R Then, the Z-axis coordinate of the MARK center in the camera coordinate system OXYZ is calculated as Equation (9), where B is the baseline of the binocular camera;

[0031]

[0032] The relationship between the camera coordinate system OXYZ and the pixel coordinate system μOv is as follows:

[0033]

[0034] Based on the Z-axis coordinates in OXYZ, obtain the Y-axis coordinates of the MARK center;

[0035] Step 2-1-3, Calculate the maximum chest compression depth based on the baseline value: After obtaining the x-coordinate of the center point of the MARK marker in the left and right camera imaging plane coordinate systems, obtain the parallax dis = x. L -x R Then, based on the triangulation model, the Y-axis coordinate of the MARK marker in the camera coordinate system is obtained through the Z-axis coordinate of the object in the camera coordinate system. This Y-axis coordinate is denoted as Y(t) with time t as the independent variable. After accurately calculating the position of the center point of the MARK marker, the initial position of the MARK marker is obtained as the reference value B0 at the initial moment of compression. Then, during chest compressions, the CCD changes over time, specifically CCD(t) and MCCD, based on the difference in position between the center of the MARK marker and the reference value in the compression direction. The calculation method is as follows:

[0036]

[0037] Where Y max The maximum value of Y(t) is obtained by directly measuring the difference between the palm's initial position and the maximum pressing depth as the CCD.

[0038] Step 2-1-4, error compensation to obtain accurate maximum chest compression depth:

[0039] Preliminary experiments were conducted, and analysis of chest compression data revealed that when calculating the CCD based on visual measurements of the MARK mark center, errors caused by hand muscle deformation and slight arm tilt during force application necessitate error compensation. This error is defined as deviation b, and an error compensation method is proposed. This method measures the deviation b for each individual by pressing on a hard tabletop before chest compressions, and then uses b for error compensation during chest compressions. The error compensation method is as follows:

[0040]

[0041] Step 2-2, calculate the chest compression rate:

[0042] The CCR(t) is calculated based on the time difference ΔT between adjacent MCCDs, in seconds. The formula is shown in equation (13), where the time between two adjacent MCCDs is set as t. i-1 and t i Then ΔT = t i -t i-1 ;

[0043]

[0044] Beneficial effects:

[0045] This invention proposes a real-time detection system for chest compression depth, frequency, and posture based on binocular vision. (1) A binocular matching point extraction method combining target detection and color region segmentation is proposed. This method achieves fast and effective target ranging by detecting the motion trajectory of the MARK marker on the rescuer's forearm. (2) Color region segmentation avoids edge detection, has low resolution requirements, and is not affected by bounding box offset and object rotation. In addition, color region segmentation only requires color space conversion, which has sufficiently high operating efficiency. (3) We also propose a CCD detection error compensation method caused by the rescuer's hand deformation and arm tilt, which can improve the detection accuracy of compression depth. (4) Our evaluation method provides a non-contact measurement solution that does not require any tools to be installed between the rescuer's hand and the patient's chest. It is convenient and quick, and has high application value in medical human-computer interaction. Attached Figure Description

[0046] Figure 1 The diagram shows the structure of the chest compression recognition system and detection method based on binocular vision provided by this invention.

[0047] Figure 2A flowchart of CCD, CCR, and CCP detection methods based on a binocular vision system;

[0048] Figure 3 This refers to the geometric model of the camera and its reference coordinate system.

[0049] Figure 4 A schematic diagram of the checkerboard mark and its wristband;

[0050] Figure 5 A schematic diagram of the center of the MARK logo is extracted to perform color region segmentation based on the HSV color space;

[0051] Figure 6 This is a schematic diagram illustrating the calculation of the Y-axis coordinate of the center point of the MARK marker in the camera coordinate system based on stereo matching.

[0052] Figure 7 This diagram illustrates the error caused by the relative movement between the MARK mark on the wrist and the pressed position during pressure measurement.

[0053] Figure 8 The results of the pressure depth measurement before error compensation;

[0054] Figure 9 The measurement results of the pressing depth after error compensation;

[0055] Figure 10 Scatter plot of the 95% confidence interval for the absolute error of chest compression rate;

[0056] Figure 11 This diagram illustrates the parameters that need to be calculated for quantitative assessment of CCP based on the human keypoint detection algorithm.

[0057] Figure 12 The results of human keypoint recognition based on HRNet;

[0058] Figure 13 The results are based on HRNet for pressure posture classification. Detailed Implementation

[0059] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0060] This invention provides a real-time detection system for chest compression depth, frequency, and posture based on binocular vision. The structural diagram of the binocular vision-based chest compression recognition system and detection method is shown below. Figure 1 As shown, the flowchart of the CCD, CCR, and CCP detection methods based on a binocular vision system is as follows: Figure 2 As shown; the camera's geometric model and its reference coordinate system are as follows. Figure 3The diagram shows an identification system and detection methods. The identification system includes a binocular camera, a checkerboard wristband, and a computer. The binocular camera is capable of binocular visual positioning and object recognition. The checkerboard wristband consists of a rectangular elastic band and a checkerboard pattern affixed to its center. It is worn on the rescuer's wrist or other location related to compression depth. The computer connects to the binocular camera, receives data from the camera, and uses the detection methods to calculate compression depth, frequency, and posture in real time. The detection methods include compression depth and frequency detection methods and compression posture detection methods. The checkerboard pattern is also known as a marker.

[0061] Example:

[0062] For step 2-1, refer to Figure 4 The following steps are taken to identify the checkerboard-patterned MARK markers and extract their center points:

[0063] (1) MARK identification: The YOLOv5 algorithm was chosen to detect MARKs. We trained the yolov5s.pt pre-trained model on our own MARK dataset. Then, the trained model was quantized by fp16 (float point 16) and accelerated using the NVIDIA TensorRT high-performance deep learning inference framework. Multi-threaded programming technology was used to optimize the processing speed of the software.

[0064] (2) Color Region Segmentation for MARK Center Point Extraction: The trained YOLOv5 model can accurately locate the MARK, and then color segmentation can precisely locate the center of the MARK. This segmentation involves converting the RGB image to the HSV color space and then identifying black regions based on HSV values. Other color region segmentation schemes, such as image segmentation techniques, can also be used, and it is not limited to HSV color conversion methods. Color region segmentation avoids edge detection, has low resolution requirements, and is unaffected by bounding box offsets and object rotation. In addition, color region segmentation only requires color space conversion operations, resulting in sufficiently high running efficiency.

[0065] (3) Test results of target detection and color region segmentation: The effect of color region segmentation is as follows: Figure 5 As shown, the stereo matching results of MARK center points based on object detection combined with color region segmentation are as follows: Figure 6 As shown, after detecting the MARK using object detection, the center of the MARK is detected within the color region. Then, the Y-axis coordinate of the MARK's center point is calculated using stereo matching, and the measurement is 30mm.

[0066] For step 2-1-4, regarding the proposed deviation b, we compared and presented the measured and actual values ​​of the pressure applied. For example... Figure 7 As shown, the negative direction of the y-axis is the same as the pressing direction. By aligning the data in the pressing direction, it can be seen that there is a deviation b at the box when pressing from light to dark. At the same time, the curve trend of the measured value and the true value in the red box is the same, only the change of the measured value is larger, which verifies the rationality of our conjecture.

[0067] Experiments were conducted on 20 people before and after error compensation. First, the pressing depth was measured directly according to formula (11), and the results are as follows: Figure 8 As shown, the average error is 14.3063 mm. Before measurement, each person pressed on a hard table to measure the deviation b caused by hand muscle deformation and arm tilt. This value was then subtracted from the measured MCCD during the actual measurement to obtain the precise pressing depth. Subsequently, we analyzed the experimental data and, according to equation (12),... Figure 10 Subtracting each person's deviation 'b' from their data yields the absolute error curve, as shown below. Figure 9 As shown, the pressing depth results that meet the error requirements were obtained, with an average absolute error of -1.0773 mm.

[0068] For step 2-2, we calculated CCR according to equation (13), and the result is as follows: Figure 10 As shown, a scatter plot of the absolute error values ​​for compress frequency calculation is presented. Based on the distribution of the absolute error of compress frequency within the 95% confidence interval, the arithmetic mean of the absolute error at a 95% confidence level is -0.1117 compressions / minute. Most data points are closely distributed around the mean, indicating that our method has high accuracy in compress frequency measurement, with a 95% confidence interval of [-9.6186, 8.2067] compressions / minute.

[0069] For step 3, to determine whether the elbow joint is straight and whether the arm is perpendicular to the patient's chest in the direction of compression, we designed a CCP assessment method based on 17 human body key points (i.e., nose, left eye, right eye, left ear, right ear, left shoulder, right shoulder, left elbow, right elbow, left wrist, right wrist, left hip, right hip, left knee, right knee, left ankle, and right ankle) from the COCO train 2017 dataset. A scheme containing only six key points—left shoulder, right shoulder, left elbow, right elbow, left wrist, and right wrist—can complete CCP identification. The judgment steps are as follows:

[0070] (1) Calculation of angles α and β: The flexion of the elbow joint is determined by the angle between the elbow and the shoulder joint. This angle is calculated by the angle between the lines connecting the key points of the elbow joint and the key points of the shoulder joint, and the lines connecting the key points of the shoulder joint and the key points of the wrist joint. Figure 11 As shown by ∠α. Whether the compression direction is perpendicular to the patient's chest is determined by the angle between the compression direction and the vertical line of the patient's chest. This angle is calculated by the angle between the line connecting the key points of the shoulder joint and the key points of the wrist joint and the vertical direction, such as... Figure 11 Shown as ∠β.

[0071] Where α is calculated based on triangle ACD, and β is calculated based on triangle ABC. The formulas are as follows:

[0072]

[0073] (2) Selection of judgment criteria: Considering the existence of individual differences, after communicating with professional medical staff, our judgment criteria is to perform key point detection when pressing the deepest point. At this time, when ∠α is in the range of [165°, 180°], it is considered as pressing with the elbow straight. When ∠β is in the range of [0°, 10°], it is considered as pressing in a direction perpendicular to the patient's chest.

[0074] (3) Human Keypoint Detection Scheme and Experimental Testing: The pre-trained model selected for the human keypoint detection algorithm is the HRNet-W32 pre-trained model, which is trained on the COCO train2017 dataset containing 57K images and 150K human instances, and evaluated on the val2017 validation set containing 5K images and the test-dev2017 test set containing 20K images, respectively. The recognition results are as follows: Figure 12 As shown. To determine whether the elbow joint was straight and whether the arm was perpendicular to the patient's chest in the direction of compression, we classified CCPs according to Equation (14), and the experimental results were as follows. Figure 13 Experimental results show that when the camera is in a fixed position, the arithmetic mean accuracy of the press posture detection based on the HRNet algorithm reaches 90.806% when the elbow is not straight, 90.845% when the press is correct, 95.752% when the press is not perpendicular, and 91.844% when the press is neither perpendicular nor straight. Overall, the average recognition rate of the press posture in the experiment is 92.32%. The experimental results demonstrate that the method based on human keypoint detection can successfully complete the task of estimating and classifying press postures.

[0075] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any modifications or equivalent changes made based on the technical essence of the present invention shall still fall within the scope of protection claimed by the present invention.

Claims

1. A method for detecting chest compression depth frequency and posture based on binocular vision, characterized in that: Includes the following steps: Step 1: Image acquisition, distortion removal and stereo correction, calculation of the coordinates of the center point of the checkerboard marker MARK; Step 2 is detailed below; Step 2-1, calculating the pressing depth, specifically includes: Step 2-1-1, Coordinates of the MARK Center in the Imaging Plane Coordinate System: Based on the camera's geometric model and the principle of triangulation, the two-dimensional pixel coordinates are restored to three-dimensional space through the new camera intrinsic and extrinsic parameters after distortion correction and stereo correction. The physical coordinates of the MARK center in the camera coordinate system OXYZ are calculated. Then, based on the coordinates (μ) of the MARK center in the left-hand camera coordinate system... L ,v L ) and the coordinates of the MARK center in the right camera coordinate system (μ R ,v R ), calculate the coordinate x of the MARK center in the xO′y coordinate system of the imaging plane, and the specific calculation method is Equation (8); (8) Where f is the camera's true focal length, in millimeters. x and f y Focal length is measured in pixels, c x and c y The principal point coordinates are in pixels, f x f y c x and c y All of these are intrinsic parameters obtained from camera calibration. α and β It refers to the width and height of a two-dimensional image, that is, the size of a pixel; Step 2-1-2, Y-axis coordinate of the MARK center in the camera coordinate system: The x-axis coordinate of the MARK center in the xO′y coordinate system of the left and right camera imaging planes is obtained. L and x R After determining the coordinates, the disparity dis = x is calculated. L – x R Then, the Z-axis coordinate of the MARK center in the camera coordinate system OXYZ is calculated as Equation (9), where B is the baseline of the binocular camera; (9) The relationship between the camera coordinate system OXYZ and the pixel coordinate system μOv is as follows: (10) Based on the Z-axis coordinates in OXYZ, obtain the Y-axis coordinates of the MARK center; Step 2-1-3, Calculate the maximum chest compression depth based on the baseline value: After obtaining the x-coordinate of the center point of the MARK marker in the left and right camera imaging plane coordinate systems, obtain the parallax dis = x L − x R， Then, based on the triangulation model, the Y-axis coordinate of the MARK marker in the camera coordinate system is obtained through the Z-axis coordinate of the object in the camera coordinate system. This Y-axis coordinate is denoted as Y(t) with time t as the independent variable. After accurately calculating the position of the center point of the MARK marker, the initial position of the MARK marker is obtained as the reference value B0 at the initial moment of compression. Then, during chest compressions, the chest compression depth CCD changing with time, CCD(t), and the maximum chest compression depth MCCD are calculated based on the difference between the center of the MARK marker and the reference value in the compression direction. The calculation method is as follows: (11) Where Y max The maximum value of Y(t); Step 2-1-4, error compensation to obtain accurate maximum chest compression depth: The deviation b of each individual was measured by pressing on a hard tabletop before chest compressions were performed, and then b was used for error compensation during chest compressions. The error compensation method was as follows: (12) Step 2-2, calculate the chest compression rate: The CCR (t) is calculated based on the time difference ∆T between adjacent MCCDs, in seconds. i The calculation formula is shown in equation (13), where the time between two adjacent MCCDs is set as t. i-1 and t i Then ∆T = t i – t i-1 ; (13) Step 3: Chest compression posture detection and classification; During chest compressions, a human key point detection algorithm is used to locate key points on the shoulder, elbow, and wrist to assess whether the elbow is straight and whether the direction of chest compressions is perpendicular to the patient's chest cavity. These parameters are calculated by taking the angle α' between the line connecting the shoulder and elbow and the line connecting the elbow and wrist, and the angle β' between the line connecting the shoulder and wrist and the vertical direction. When calculating the compression posture based on the acquired human key points, if ∠α' is within the range of [165°, 180°], The pressing is considered to be done with the elbow extended. When ∠β' is within the range of [0°, 10°], the pressing direction is identified as vertical. Based on the set angle range, four different pressing postures are classified: Elbow, the elbow is not extended and ∠α' does not meet the requirements; Right, the pressing is correct and both ∠α' and ∠β' meet the requirements; Vertical, the pressing is not vertical and ∠β' does not meet the requirements; Wrong, the elbow is not extended and the pressing is not vertical and neither ∠α' nor ∠β' meets the requirements.

2. The method for detecting chest compression depth frequency and posture based on binocular vision according to claim 1, characterized in that: The specific steps of step 1 are as follows; Step 1-1, Camera Calibration: Before CCD measurement, calibrate the stereo cameras to obtain the intrinsic parameters K and distortion parameter distCoeff of the two cameras, as well as the external rotation parameter R and translation parameter T of the right camera relative to the left camera. (1) (2) (3) (4) Among them, c x and c y The coordinates of the principal point are in pixels; assuming f is the focal length of the camera, then f x = α · f,f y =β · f, where α and β are the width and height of the corresponding two-dimensional image pixel size; Steps 1-2, Distortion Removal and Stereo Correction: The radial and tangential distortions of the camera are as follows: (5) (6) Where k1, k2, and k3 represent the coefficients of radial distortion, r² = x² + y², and p1 and p2 represent the coefficients of tangential distortion, x corrected and y corrected Given the corrected positions of x and y in the distorted image, and combining radial and tangential distortion, considering only the first three terms of the radial distortion coefficient and the two terms of the tangential distortion coefficient, a comprehensive distortion correction formula is obtained. (7) The correct position of a point in the pixel plane can be determined using five distortion coefficients k1, k2, k3, p1, and p2. Steps 1-3, MARK center point coordinate calculation: Using the center point of the MARK marker as the matching point, the binocular matching point detection based on object detection and color region segmentation is used to obtain a matching point that allows the binocular camera to obtain an accurate matching point. Combining the triangulation model and the camera's geometric modeling model, the coordinates of the center point of the MARK marker in the camera coordinate system are calculated. After obtaining the position of the marker in the μOv pixel coordinate system based on object detection, this method uses color detection to identify the color region of the MARK and finds the center point of the MARK based on the distribution of different color regions.

3. The system for detecting chest compression depth, frequency, and posture based on binocular vision according to claim 1, characterized in that: The system includes a binocular camera, a checkerboard wristband, and a computer. The binocular camera is capable of binocular visual positioning and object recognition. The checkerboard wristband includes a rectangular elastic band and a checkerboard pattern attached to its center surface. The checkerboard pattern is called a MARK and is worn on the rescuer's wrist or other location related to compression depth. The computer is connected to the binocular camera, receives data from the camera, and calculates compression depth, frequency, and posture in real time based on the aforementioned detection method.

Images