Human posture-based video measurement method and device, and storage medium

Abstract

The application relates to the field of video measurement, and provides a video measurement method and device based on human body posture and a storage medium, the method comprising the following steps: acquiring a normal posture image of a measurement object, labeling original vertebra coordinates of spine reference points of the measurement object in the normal posture, and performing normalization processing to obtain relative position coordinates of the spine reference points; acquiring a real-time activity video of the measurement object, and extracting two-dimensional image coordinates of shoulder key points and hip key points from the real-time activity video in real time; constructing a quadratic Bezier curve based on the relative position coordinates of the spine reference points and the two-dimensional image coordinates of the shoulder key points and the hip key points; obtaining a spine coordinate sequence through equidistance resampling of the quadratic Bezier curve; and generating a visual video measurement report according to the spine coordinate sequence, and pushing the report to the measurement object and related users. The application adopts video measurement to reduce measurement errors caused by static images and experience judgment, and improves measurement accuracy and measurement efficiency.

Description

Technical Field

[0001] This application relates to the field of video measurement, and more specifically to a video measurement method, apparatus, and storage medium based on human posture. Background Technology

[0002] In existing technologies, human skeletal posture measurement is often achieved through static images. Taking spinal posture measurement as an example, existing technologies typically use fluoroscopic static images (such as MRI, CT, and X-rays) to obtain pixel information of the spinal bony structures, and then manually calculate the relevant angles, positions, and other geometric physical quantities of the spine. This method can only extract the physical quantities of the spine at a single moment and cannot measure the internal activity of the spine in real time when the patient is undergoing rehabilitation training.

[0003] Currently, rehabilitation physicians can only subjectively compare static fluoroscopic images with visual observations at the rehabilitation site to obtain the skeletal activity status. However, this method of activity assessment has large measurement errors and low efficiency. Furthermore, since visual observation can only see the external outline of the object being measured, this method cannot obtain the true activity status of skeletal structures such as the spine, leading to a mismatch between the measurement results and the actual required internal spinal motion parameters, greatly reducing the effectiveness and accuracy of the measurement results.

[0004] In summary, there is an urgent need to propose a dynamic measurement scheme to address the technical problems in existing technologies, such as large measurement errors, low measurement efficiency, and insufficient measurement accuracy in human skeletal posture measurement due to reliance on static medical images and subjective assessment. Summary of the Invention

[0005] This application provides a video measurement method, device, and storage medium based on human posture. By using video measurement to achieve dynamic measurement of spinal motion state, it effectively avoids measurement errors caused by static images and experience judgment during spinal motion state assessment, improves measurement accuracy and efficiency, and ensures measurement precision and real-time measurement results.

[0006] In a first aspect, embodiments of this application provide a video measurement method based on human posture, the method comprising: Acquire the orthogonal posture image of the object being measured; Based on the positive posture image, the original vertebral coordinates of the spinal reference point of the measured object in the positive posture are labeled, and the relative position coordinates of the spinal reference point are obtained by normalization. Acquire real-time activity video of the measurement object, and extract two-dimensional image coordinates of shoulder key points and hip key points from the real-time activity video in real time using a pose estimation model; Based on the relative position coordinates of the spinal reference point, the two-dimensional image coordinates of the shoulder key point and the hip key point, a quadratic Bézier curve is constructed to simulate the natural curvature path of the spinal centerline of the measured object in multiple consecutive video frames. By resampling the quadratic Bézier curve at equal intervals and mapping the physical coordinates, the spine coordinate sequence corresponding to the spine reference point in multiple consecutive video frames is obtained. The spinal change trend of the measured object, the segmental angle time series matrix used to quantify the degree of local spinal deformation, and the activity symmetry parameters are calculated based on the spinal coordinate sequence. The spinal change trend of the measured object and the segment angle time series matrix are converted into a visualized video measurement report and pushed to the measured object and / or relevant users.

[0007] Secondly, embodiments of this application provide a video measurement device based on human posture, which has functions corresponding to the video measurement method based on human posture provided in the first aspect above. These functions can be implemented in hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above functions, and these modules can be software and / or hardware.

[0008] In one embodiment, the human posture-based video measurement device includes: The input / output module is configured to acquire an orthogonal orientation image of the measured object; The processing module is configured to: label the original vertebral coordinates of the spinal reference points of the measured object in the orthostatic posture image, and normalize them to obtain the relative position coordinates of the spinal reference points; acquire real-time activity videos of the measured object, and extract the two-dimensional image coordinates of shoulder key points and hip key points from the real-time activity videos in real time through a posture estimation model; construct quadratic Bézier curves based on the relative position coordinates of the spinal reference points and the two-dimensional image coordinates of the shoulder and hip key points to simulate the natural curvature path of the spinal centerline of the measured object in multiple consecutive video frames; obtain the spinal coordinate sequence corresponding to the spinal reference points in multiple consecutive video frames by equidistant resampling of the quadratic Bézier curves and physical coordinate mapping; and calculate the spinal change trend of the measured object, the segmental angle time series matrix used to quantify the degree of local spinal deformation, and the activity symmetry parameters based on the spinal coordinate sequence. The input / output module is also configured to convert the spinal change trend of the measured object and the segment angle time series matrix into a visualized video measurement report and push it to the measured object and / or relevant users.

[0009] Thirdly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the video measurement method based on human posture as described in the first aspect.

[0010] Fourthly, embodiments of this application provide a computing device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the video measurement method based on human posture described in the first aspect.

[0011] Fifthly, embodiments of this application provide a video measurement system based on human posture. The video measurement system based on human posture includes a processor for implementing the functions involved in the first aspect above, such as generating or processing the information involved in the video measurement method based on human posture provided in the first aspect above.

[0012] In one possible design, the aforementioned human posture-based video measurement system further includes a memory connected to the processor via a circuit structure. This memory stores necessary program instructions and data for the terminal. The human posture-based video measurement system can be constructed from a chip or may include the chip and other discrete components. Further optionally, the chip includes a communication interface to which the processor is connected. The communication interface receives data and / or information that needs to be processed. The processor obtains the data and / or information from the communication interface, processes the data and / or information, and outputs the processing results through the communication interface. This communication interface can be an input / output interface.

[0013] In a sixth aspect, embodiments of this application provide a computer program product containing instructions that, when run on a computer, cause the computer to execute the video measurement method based on human posture provided in the first aspect.

[0014] Compared to existing technologies, in this embodiment, the relative position coordinates of the spinal reference points are obtained by acquiring the orthogonal posture image of the measurement object and normalizing it. The coordinates of key points of the shoulder and hip extracted from the real-time activity video are combined to construct a quadratic Bézier curve to simulate the spinal centerline. Then, a continuous spinal coordinate sequence is obtained by equidistant resampling and mapping with physical coordinates. Subsequently, quantitative parameters such as the trend of spinal changes are automatically calculated and a visual video measurement report is generated.

[0015] Compared to existing technologies that rely on static perspective images for manual calculations and subjective assessments based on visual observation of the external contour, this application's embodiments utilize video measurement and spinal structure reconstruction. This allows for continuous, temporal dynamic tracking of the internal spinal activity during real-time movement of the measured object, achieving dynamic quantification and automated measurement of the spinal internal activity, thus improving measurement efficiency and accuracy. Furthermore, automated calculation of spinal-related parameters further avoids subjective errors caused by manual assessment. In addition, this application uses the relative coordinates of spinal reference points as the absolute metrological benchmark. Through Bézier curve modeling, the reconstructed spinal centerline maintains metrological consistency with the actual bony structure. All measurement results are quantified temporal data based on video frames, possessing traceability and reproducibility. The visual report intuitively presents the dynamic changes of the spine, preserving the true characteristics of the spinal bony structure and achieving real-time capture of the internal spinal activity. It also solves the problems of existing technologies that only acquire external contour information and where measurements do not match actual requirements. Therefore, the quantitative parameters such as the spinal change trend and segment angle time series matrix obtained by the embodiments of this application, as well as the visualized video measurement report, can meet the actual needs of spinal rehabilitation training for dynamic measurement of the internal activity state of the spine and measurement accuracy, and solve the technical problems of large measurement error, low efficiency and insufficient accuracy in the prior art. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the video measurement method based on human posture according to an embodiment of this application; Figure 2 This is a schematic diagram of the spine centerline of a video measurement method based on human posture according to an embodiment of this application; Figure 3 This is another schematic diagram of the spine centerline for the video measurement method based on human posture according to an embodiment of this application; Figure 4 This is a schematic diagram of the spinal curvature direction of a video measurement method based on human posture according to an embodiment of this application. Figure 5 This is a comparison image of the maximum lateral bending posture in the video measurement method based on human posture according to an embodiment of this application; Figure 6 This application provides a thermal image of the local curvature angle of a spinal segment using a video measurement method based on human posture. Figure 7 This is a schematic diagram of the structure of a video measurement device based on human posture according to an embodiment of this application. Detailed Implementation

[0017] This application also provides a video measurement method, apparatus, and storage medium based on human posture. This can be applied to video measurement systems based on human posture in scenarios such as spinal rehabilitation training monitoring, clinical assessment of spinal posture, and remote spinal rehabilitation consultation, which can achieve dynamic quantitative measurement of human spinal posture. The human posture-based video measurement system may include an image processing device and a measurement device, which can be integrated or deployed separately. The image processing device is at least used to perform specialized image processing operations on the acquired orthostatic posture image and the real-time activity video of the measured object to obtain various processing result data related to video measurement. The measurement device is used to identify and detect the input orthostatic posture image and real-time activity video frames to obtain image recognition and detection results related to video measurement. The image processing device can be an application program that implements image processing logic such as normalization of spinal reference point coordinates, construction of quadratic Bézier curves, calculation of spinal coordinate sequences, calculation of spinal quantification parameters, and generation of visualization reports, or a server or terminal device with the application program installed. The measuring device can be an image recognition program that implements image recognition and detection logic such as spinal reference point annotation and extraction of key points of the human shoulder and hip. The image recognition program is, for example, a pose estimation model or a key point detection model of the spinal vertebrae. The measuring device can also be a terminal device, server, or other computing device that has deployed such an image recognition model.

[0018] The solutions provided in this application can be applied to scenarios such as dynamic quantitative measurement of human spinal posture and monitoring of spinal rehabilitation training. The following explains some concepts involved in the embodiments of this application: Video measurement refers to an engineering surveying discipline that uses ordinary video equipment (including smartphone cameras) to achieve high-precision measurement of physical quantities such as length, angle, and deformation through image processing, camera calibration, and geometric inverse calculation. This application applies video measurement to the dynamic three-dimensional measurement of the human spine.

[0019] Fluorescent imaging refers to medical images with penetrating power, including but not limited to standing full-spine X-rays, MRI, and CT. In this application, it specifically refers to standing anteroposterior or lateral X-rays used as initial measurement benchmarks.

[0020] The dynamic Cobb angle is the instantaneous scoliosis angle calculated in each frame of the rehabilitation training video using the classic Cobb method (the angle between the vertical lines of the upper edge of the upper vertebra and the lower edge of the lower vertebra) based on the real-time reconstructed spinal centerline.

[0021] The internal movement of the spine refers to the real-time position, rotation, and angular changes of the vertebrae in the sagittal, coronal, and horizontal planes, rather than just the surface outline of the trunk.

[0022] In existing technologies, human skeletal posture measurement generally relies on static images. Taking spinal posture measurement as an example, the industry typically uses fluoroscopic static images to acquire pixel information of the spinal bony structures, and then manually calculates the relevant angles, positions, and other geometric physical quantities of the spine. On the one hand, due to limitations in the usage scenarios and operating procedures of medical imaging equipment, fluoroscopic image acquisition cannot be performed in real-time during the subject's rehabilitation training, daily activities, or other dynamic processes. On the other hand, existing technologies lack dynamic measurement techniques that can capture the internal activity state of the spine in real time, making it difficult to achieve continuous quantification and extraction of spinal bony structure motion parameters. Therefore, it is only possible to rely on static images to complete the measurement of spinal parameters at a single moment, combined with human subjective judgment to complete the spinal posture assessment in dynamic scenarios. This assessment method is greatly affected by human subjective factors, resulting in large measurement errors and low measurement efficiency. Furthermore, manual observation can only capture the external contour of the torso and cannot obtain the true activity state of the spinal bony structures, leading to a mismatch between the measurement results and the actual required internal spinal motion parameters, significantly reducing the effectiveness and accuracy of the measurement results. In summary, there is an urgent need to propose a dynamic measurement scheme to address the technical problems in existing technologies, such as large measurement errors, low measurement efficiency, and insufficient measurement accuracy caused by reliance on static medical images and subjective human assessment.

[0023] Compared to existing technologies, this application proposes a video measurement scheme based on human posture by integrating video metrology with artificial intelligence and computer vision technologies. The scheme involves acquiring an orthogonal posture image of the object being measured, normalizing the coordinates of spinal reference points, extracting the coordinates of key human points from real-time activity video, constructing a quadratic Bézier curve to simulate the spinal centerline, resampling and mapping to physical coordinates to obtain a spinal coordinate sequence, calculating spinal quantification parameters, and generating and pushing a visual measurement report. This comprehensive process addresses various problems of existing technologies: First, by relying on camera equipment (such as mobile phones, tablets, or other terminal devices equipped with image acquisition modules) to acquire real-time activity video, and combining it with a spinal measurement benchmark constructed from perspective images, continuous temporal measurement of the internal activity state of the spine in dynamic scenes is achieved, no longer limited by the spatiotemporal constraints of static images. Second, by automatically completing the entire process of key point extraction, coordinate calculation, and parameter solving, measurement errors caused by manual calculation and subjective comparison are avoided, improving measurement accuracy and efficiency. Third, a measurement benchmark is constructed based on the spinal bony structure information of fluoroscopic images. Combined with Bézier curve modeling, the accurate reconstruction of the spinal centerline is achieved, which can capture the real activity state inside the spine and make the measurement results highly matched with the actual spinal internal motion parameters, further improving the effectiveness and accuracy of the measurement results.

[0024] In some embodiments, the image processing device and the measuring device in this application can be deployed separately. The video measurement method provided in this application embodiment can be implemented based on a video measurement system composed of a server and a terminal device, which specifically includes a server 01 and a terminal device 02.

[0025] The server 01 serves as an image processing device, internally deploying image processing programs such as image preprocessing, normalization of spinal reference point coordinates, construction of quadratic Bézier curves, calculation of spinal coordinate sequences, calculation of spinal quantification parameters, and generation of visualization reports. The terminal device 02 serves as a measurement device, internally deploying image recognition models trained based on machine learning and computer vision technologies, such as pose estimation models for extracting human key points and spinal vertebral key point annotation models.

[0026] The terminal device 02 can forward the acquired orthostatic posture image and real-time activity video of the measured object to the server 01. Based on the received image and video, the server 01 sequentially completes the normalization of the spine reference point coordinates, the construction of Bézier curves, the calculation of the spine coordinate sequence, the calculation of the spine change trend and related parameters, and generates a visualized video measurement report, which is then sent to the terminal device 02. The terminal device 02 can push the received video measurement report to the measured object and / or relevant users, or it can display and store the report locally.

[0027] It should be noted that the computing devices involved in the embodiments of this application can be servers and / or terminal devices. The servers involved in the embodiments of this application can be independent physical servers, server clusters or distributed systems composed of multiple physical servers, or cloud servers providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The terminal devices involved in the embodiments of this application can be devices that provide voice and / or data connectivity to users, handheld devices with wireless connectivity, or other processing devices connected to a wireless modem. Examples include mobile phones and computers with mobile terminals; for instance, these can be portable, pocket-sized, handheld, computer-embedded, or vehicle-mounted mobile devices that exchange voice and / or data with a wireless access network.

[0028] Reference Figure 1 , Figure 1 This is a flowchart illustrating a video measurement method based on human posture, provided in an embodiment of this application. The method can be executed by a video measurement device based on human posture and can be applied to scenarios such as spinal rehabilitation training monitoring, clinical assessment of spinal posture, and remote spinal rehabilitation consultation. The method includes steps 101-107: Step 101: Obtain the orthogonal posture image of the object being measured.

[0029] The anteroposterior posture image can be the fluoroscopic image described in this application. For example, it can be a medical image with penetrating power. Specifically, a standing full-spine anteroposterior X-ray film taken under standard projection conditions with the measurement object in a natural standing position is also the highest metrological source used to establish an absolute metrological benchmark in dynamic video measurement of the spine. This image is captured by professional medical imaging equipment and can clearly present the bony structural pixel information of 10 key vertebrae from C4 to S1 of the measurement object. The bony features are clear and unobstructed, and can be directly used for accurate annotation of spinal reference points. At the same time, the anteroposterior posture image does not contain redundant privacy information. After subsequent processing, it can generate lightweight bony measurement information (such as structured data in a preset format), which is the core foundation for subsequent spinal coordinate normalization and dynamic reconstruction of the spinal centerline in video frames, providing a traceable spinal bony structural benchmark reference for the entire video measurement process.

[0030] Step 102: Based on the positive posture image, label the original vertebral coordinates of the spinal reference point of the measurement object in the positive posture, and normalize them to obtain the relative position coordinates of the spinal reference point.

[0031] Spinal reference points are feature points selected from anteroposterior images (standing full-spine anteroposterior X-rays) that characterize the core locations of the spinal bony structures. For example, spinal reference points can be set as the center locations of 10 key vertebrae: C4 (4th cervical vertebra), C6 (6th cervical vertebra), T1 (1st thoracic vertebra), T4 (4th thoracic vertebra), T7 (7th thoracic vertebra), T10 (10th thoracic vertebra), L1 (1st lumbar vertebra), L3 (3rd lumbar vertebra), L5 (5th lumbar vertebra), and S1 (1st sacral vertebra). These reference points cover the core functional segments of the spine from the cervical to the sacral vertebrae. They are the core anchor points for constructing the spinal centerline and achieving quantitative measurement of spinal motion. Their locations are directly related to the actual distribution of the spinal bony structures, providing accurate bony benchmarks for subsequent dynamic reconstruction.

[0032] The original vertebral coordinates of the spinal reference points refer to the two-dimensional pixel coordinates of the above 10 vertebral centers in the anteroposterior posture image, expressed as (x...). i y i The form of ) is (i=1, 2, ..., 10), where x i The x-coordinate is the pixel value, y iThis represents the pixel value of the ordinate. This coordinate is based on the global pixel coordinate system of the orthogonal posture image and is obtained through automatic identification and annotation or semi-automatic annotation review by physicians. It locates the pixel position of each vertebral body center on the X-ray film and serves as the raw data basis for subsequent coordinate normalization processing. The original vertebral body coordinates contain pixel units, and their scale is inconsistent across different scenarios due to factors such as imaging equipment parameters, shooting distance, and the height and body type of the measured subject. Therefore, they cannot be directly used for dynamic video measurements across different scenarios. Thus, this step requires targeted processing of the original vertebral body coordinates, as described below.

[0033] The relative position coordinates of the spinal reference points are dimensionless coordinates obtained by normalizing the original vertebral coordinates, expressed as (r ix r i The coordinates are represented in the form of (i=1, 2, ..., 10), and are standardized coordinates that can be reused across scenarios and have a unified measurement standard. In the process of obtaining the relative position coordinates of the spinal reference points, the reference shoulder midpoint (the midpoint of the pixel coordinates of the bony key points of the left and right shoulders) and the reference hip midpoint (the midpoint of the pixel coordinates of the bony key points of the left and right hips) are first extracted from the anteroposterior pose image. The Euclidean distance between the reference shoulder midpoint and the reference hip midpoint is calculated as the length of the human body principal axis. Then, a normalization operation is performed on the original vertebral coordinates of each spinal reference point. For example, the difference is obtained by subtracting the pixel coordinates of the reference hip midpoint from the original vertebral coordinates, and then dividing this difference by the length of the human body principal axis to finally obtain the relative position coordinates. The numerical range of these coordinates is strictly controlled within [-1, 1], where r ix ∈[-0.5, +0.5], representing the left and right offset of the vertebral body relative to the midline of the spine (0 is the midline position), r i ∈[0.00, +1.00], representing the longitudinal position of the vertebral body relative to the midpoint of the hip and the midpoint of the shoulder (0 is at the level of the midpoint of the hip, 1 is at the level of the midpoint of the shoulder). The relative position coordinates eliminate the differences between the shooting and individual scales, retaining only the relative positional relationship between the 10 vertebral bodies, and finally forming a lightweight bony measurement template with a volume of less than 2KB. It contains no privacy information and can be directly stored in mobile devices, providing a unified absolute measurement benchmark for the dynamic reconstruction of the spinal centerline in subsequent video frames, facilitating real-time initialization of video measurements.

[0034] As an optional embodiment, in step 102, the normalization process to obtain the relative position coordinates of the spinal reference point includes: extracting the trunk physiological feature points of the measurement object in the orthopedic posture image; the trunk physiological feature points include the reference shoulder midpoint and the reference hip midpoint; and normalizing the original vertebral coordinates based on the original pixel coordinates of the trunk physiological feature points to obtain the relative position coordinates of the spinal reference point.

[0035] Specifically, in step 102, feature points of the spinal bony structures in the anteroposterior posture image are first annotated. Ten key vertebral centers, from C4 to S1, are selected as spinal reference points. The original vertebral pixel coordinates of each spinal reference point in the anteroposterior posture image are obtained through AI automatic annotation or semi-automatic annotation by a physician. The original vertebral coordinates are then normalized to eliminate scale differences caused by the shooting equipment, shooting distance, and the height and body shape of the measured object, forming relative position coordinates that can be reused across scenes. Further, optionally, the normalization process involves: firstly, extracting the trunk physiological feature points of the measured object from the anteroposterior posture image, including the reference shoulder midpoint and the reference hip midpoint. The reference shoulder midpoint is the midpoint of the pixel coordinates of the bony key points of the left and right shoulders, and the reference hip midpoint is the midpoint of the pixel coordinates of the bony key points of the left and right hips. Then, based on the original pixel coordinates of the trunk physiological feature points, the original vertebral coordinates of each spinal reference point are normalized. Specifically, the original pixel coordinates of the reference hip midpoint are subtracted from the original vertebral coordinates of the spinal reference point to obtain the difference. Next, the Euclidean distance between the original pixel coordinates of the reference shoulder midpoint and the reference hip midpoint is calculated as the length of the human body's principal axis. Finally, the difference is divided by the length of the human body's principal axis to obtain the relative position coordinates of the spinal reference point. The normalized relative position coordinates are dimensionless values ​​with an overall value range strictly controlled within [-1, 1]. The horizontal coordinate range is [-0.5, +0.5], representing the left-right offset of the vertebral body relative to the center of the spine, with 0 representing the center position. The vertical coordinate range is [0.00, +1.00], representing the longitudinal position of the vertebral body relative to the hip midpoint and shoulder midpoint, with 0 representing the level of the hip midpoint and 1 representing the level of the shoulder midpoint. This ultimately forms coordinate information containing only the relative position of the vertebral body, facilitating real-time initialization for subsequent video measurements.

[0036] Further optionally, in the above steps, the original vertebral coordinates are normalized based on the original pixel coordinates of the trunk physiological feature points to obtain the relative position coordinates of the spinal reference points, including: for each spinal reference point, subtracting the original pixel coordinates of the reference hip midpoint from the original vertebral coordinates of the spinal reference point to obtain the difference; calculating the Euclidean distance between the original pixel coordinates of the reference shoulder midpoint and the original pixel coordinates of the reference hip midpoint to obtain the length of the human body principal axis; dividing the difference by the length of the human body principal axis to obtain the normalized relative position coordinates of the spinal reference point.

[0037] For example, suppose the original pixel coordinates of 10 spinal reference points (corresponding to the centers of the C4 to S1 vertebral bodies) in the X-ray are: .in, The x-coordinate pixel value. The vertical coordinate is the pixel value. Based on this assumption, firstly, the core reference parameters required for normalization are defined, including: First, the midpoint of the shoulder, S, is obtained by averaging the pixel coordinates of the left shoulder bony key point (LEFT_SHOULDER) and the right shoulder bony key point (RIGHT_SHOULDER) in the X-ray, expressed as follows: S = (LEFT_SHOULDER + RIGHT_SHOULDER) / 2. Second, the midpoint of the hip, H, is obtained by averaging the pixel coordinates of the left hip bony key point (LEFT_HIP) and the right hip bony key point (RIGHT_HIP) in the X-ray, expressed as follows: H=(LEFT_HIP + RIGHT_HIP) / 2. Third, the principal axis vector V of the human body is obtained by the coordinate difference between the midpoint of the shoulder, S, and the midpoint of the hip, H, i.e., V=S. H. Fourth, the length of the human body's principal axis, Dsh, is the Euclidean distance (L2 norm) of the human body's principal axis vector V, i.e., Dsh = ||V||², in pixels. Fifth, the template origin O is set to the midpoint of the hip, H, i.e., O = H. Sixth, the positive direction of the template's Y-axis is the normalized unit vector of the human body's principal axis vector V, i.e., V / Dsh, pointing from the hip to the shoulder.

[0038] After defining the above parameters, for each spinal reference point i (i.e., the center of the vertebral body), perform normalization operations according to the following steps. First, subtract the pixel coordinates of the hip midpoint H from the original pixel coordinates Pixray of the spinal reference point to obtain the coordinate difference (vector form). Then, divide the above coordinate difference by the length of the human body principal axis Dsh (i.e., ||S||). H‖2), to obtain the relative position coordinates ri of the spinal reference point, the calculation formula is: .in, , which represents the left and right offset of the vertebral body relative to the center of the spine, where a value of 0 corresponds to the center of the spine, a positive value indicates offset to the right, and a negative value indicates offset to the left. , which represents the longitudinal position of the vertebral body relative to the midpoint of the hip and the midpoint of the shoulder, where a value of 0 corresponds to the level of the midpoint of the hip and a value of 1 corresponds to the level of the midpoint of the shoulder.

[0039] Step 103: Obtain real-time activity video of the measurement object, and extract the two-dimensional image coordinates of the shoulder key points and hip key points from the real-time activity video in real time using a pose estimation model.

[0040] Step 103 is used to acquire dynamic activity images of the measured object and extract the coordinates of key control points. The real-time activity video is acquired through ordinary camera equipment (such as the monocular camera of a smartphone or tablet), which can capture continuous video images of the measured object in dynamic scenes such as rehabilitation training and daily activities, ensuring the continuity and clarity of video frames. Then, through the pose estimation model based on artificial intelligence and computer vision technology deployed in the measurement device, continuous and real-time image recognition and analysis are performed on each frame of the real-time activity video to extract the two-dimensional image pixel coordinates of four key human body points: left shoulder, right shoulder, left hip, and right hip in each frame. The extraction process achieves pixel-level accuracy and avoids errors caused by manual annotation. The extracted shoulder and hip key point coordinates will be used as the basic control point data for subsequent construction of the simulated curve of the spinal centerline.

[0041] Pose estimation models can be human keypoint detection models trained using artificial intelligence and computer vision technologies (such as the MediaPipe model). Specifically, pose estimation models can analyze input video frames in real time, automatically identifying and locating the two-dimensional image coordinates of key human joints. These models require no human intervention, capture joint position changes during dynamic human activity, and are characterized by strong real-time performance, high positioning accuracy (pixel-level), and compatibility with images captured by ordinary camera equipment. They are then used to extract human surface feature points (shoulder and hip keypoints) related to spinal posture from dynamic videos, providing reliable control point data support for subsequent spinal centerline modeling.

[0042] For example, a pre-trained MediaPipe model can be deployed in a measurement device (terminal device or server). Upon startup, the model automatically loads the core human detection algorithm and pre-trained parameters, completing initialization to enable real-time processing. The model parses the real-time video stream frame by frame, processing each frame sequentially to ensure no pose information is missed, guaranteeing the continuity of dynamic capture. Using its built-in human detection algorithm, the model identifies the human contour in a single frame, detecting the positional information of key joints (such as the head, shoulders, elbows, hips, and knees), and outputs indices and probability arrays containing all joints. The indices distinguish different joints (e.g., a specific index corresponds to the left shoulder, right shoulder, left hip, and right hip), while the probability array filters joints with the required confidence level (excluding low-confidence false detections).

[0043] Furthermore, based on the preset target keypoint index (corresponding to four key points related to spinal posture: left shoulder, right shoulder, left hip, and right hip), these four key points are selected from the joint point data output by the model, and their corresponding two-dimensional image pixel coordinates (i.e., left shoulder coordinates, right shoulder coordinates, left hip coordinates, and right hip coordinates) are extracted. The model extraction process achieves pixel-level positioning accuracy by default, and the positioning error caused by occlusion and posture changes is reduced through algorithm optimization. At the same time, low-confidence coordinate data is automatically filtered out (if the confidence of a keypoint in a certain frame is lower than the threshold, the coordinates of adjacent frames are interpolated to supplement it), ensuring that the output shoulder and hip keypoint coordinates are continuous and reliable, avoiding subjective errors caused by manual annotation.

[0044] Through the above process, we finally obtain the two-dimensional image pixel coordinates of four key points in each video frame: left shoulder, right shoulder, left hip, and right hip. These coordinates will be directly used for the calculation of the midpoint of the shoulder and the midpoint of the hip in subsequent steps, becoming the core basic control point data for constructing the quadratic Bézier curve and simulating the center line of the spine.

[0045] Step 104: Based on the relative position coordinates of the spinal reference point, the two-dimensional image coordinates of the shoulder key point and the hip key point, construct a quadratic Bézier curve to simulate the natural curvature path of the spinal centerline of the measured object in multiple consecutive video frames.

[0046] This step is used to achieve dynamic simulation of the spinal centerline. Specifically, based on the extracted two-dimensional coordinates of the shoulder and hip key points in each frame, the midpoint of the shoulder and the midpoint of the hip in the current video frame are calculated frame by frame. The midpoint of the shoulder is the midpoint of the coordinates of the left and right shoulder key points, and the midpoint of the hip is the midpoint of the coordinates of the left and right hip key points. Then, combined with the relative position coordinates of the spinal reference points, the trunk vector and trunk vector magnitude of the current frame are calculated. The trunk vector is the coordinate difference between the midpoint of the shoulder and the midpoint of the hip, and its magnitude is the Euclidean distance between the two. Subsequently, combined with the trunk tilt angle of the measured object, the lateral offset is calculated. The product of the hip midpoint coordinates, the preset coefficient, and the trunk vector, plus the lateral offset, is added to obtain the dynamic control point of the lumbar region in the current frame. This control point is used to simulate the physiological curvature of the spine in the sagittal and coronal planes, adapting to the nonlinear deformation characteristics of the spine. Finally, using the midpoint of the shoulder in the current video frame as the starting point of the curve, the midpoint of the hip as the ending point of the curve, and the dynamic control point of the waist as the control point of the curve, a quadratic Bézier curve is constructed frame by frame. This curve can accurately fit the physiological structural characteristics of the human spine, ensuring the smoothness and continuity of the curve, thereby achieving accurate simulation of the natural curvature path of the spine centerline in multiple consecutive video frames, and the curve maintains measurement consistency with the bony structure of the spine.

[0047] As an optional embodiment, in step 104, a quadratic Bézier curve is constructed using the midpoint of the shoulder as the starting point and the midpoint of the hip as the ending point, combined with the waist control points of the measured object. This includes: calculating the midpoint of the shoulder in the current video frame based on the two-dimensional image coordinates of the shoulder key points; the shoulder key points include the left and right shoulders; calculating the midpoint of the hip in the current video frame based on the two-dimensional image coordinates of the hip key points; the shoulder key points include the left and right hip joints; calculating the trunk vector and trunk vector magnitude based on the two-dimensional image coordinates of the midpoint of the shoulder and the midpoint of the hip in the current video frame; calculating the lateral offset by combining the trunk tilt angle, trunk vector, and trunk vector magnitude of the measured object; adding the product of the midpoint of the hip, a preset coefficient, and the trunk vector, and the lateral offset, to obtain the waist control point in the current video frame; constructing a quadratic Bézier curve using the midpoint of the shoulder in the current video frame as the starting point, the midpoint of the hip in the current video frame as the ending point, and the waist control point in the current video frame as the control point, wherein the control point is used to control the curvature of the curve.

[0048] For example, suppose the midpoint of the shoulder and mid-hip The formulas are as follows: , .in, and These are the pixel coordinates of the key points on the left and right shoulders (i.e., the two-dimensional image coordinates of the shoulder key points). and These are the pixel coordinates of the key points on the left and right hips (i.e., the two-dimensional image coordinates of the key points on the hips).

[0049] Furthermore, based on the shoulder midpoint in the current video frame and mid-hip Calculate the torso vector from the two-dimensional image coordinates. and trunk vector magnitude .

[0050] Specifically, the torso vector From the midpoint of the shoulder Midpoint of the hip The coordinate difference is obtained and used to characterize the longitudinal direction of the torso. The calculation formula is: Among them, the torso vector magnitude is The Euclidean distance (L2 norm) of the torso vector Vt, used to characterize the longitudinal length of the torso, is calculated as follows: Lumbar control points are used to simulate the physiological curvature of the spine in the sagittal plane (flexion and extension) and the coronal plane (lateral curvature), ensuring that the curves can capture the nonlinear deformation of the spine and improve the adaptability of dynamic measurements. For example, lumbar control points... From the midpoint of the hip Preset coefficients and torso vector The product of, lateral offset The sum of the three is given by the following formula: The coefficient 0.25 is a preset value used to constrain the control points within a reasonable range in the longitudinal direction of the trunk, ensuring that they match the physiological curvature of the spine.

[0051] Further, optionally, lateral offset The torso tilt angle of the object is measured based on the current frame. The calculation is used to adapt to the scoliosis or tilt of the spine. An example calculation formula is as follows: The value of 0.1 is a preset scaling factor that can be adjusted according to the actual scenario.

[0052] In video metrology, furthermore, quadratic Bézier curves can be used to parameterize the path of the spinal centerline on the image plane. For example, Figure 2 The spinal centerline shown can be obtained by fitting a quadratic Bézier curve. The quadratic Bézier curve can be extracted and displayed as... Figure 3 The curve shown has points on the left that represent the waist control points in the current video frame.

[0053] Specifically, the quadratic Bézier curve is located at the midpoint of the shoulder. (starting point ), waist control points (control points) ) and mid-hip (end Based on this, the curve is made smooth and continuous to simulate the natural curvature of the spine. Control points control the degree of curvature of the curve; for example, control points are pre-annotated by doctors using video samples and used for model training. Control points can also be pre-set as fixed parameters. In this embodiment, based on the foregoing example, the parametric equation of the quadratic Bézier curve satisfies the following curve parametric equation: Where u∈[0,1], The coordinates of the shoulder midpoint in the current video frame. These are the coordinates of the waist dynamic control point in the current video frame. The coordinates of the midpoint of the hip in the current video frame.

[0054] Step 105: By resampling the quadratic Bézier curve at equal intervals and mapping the physical coordinates, the spine coordinate sequence corresponding to the spine reference point in multiple consecutive video frames is obtained.

[0055] This step converts the simulated spinal centerline curve into physical coordinates that match the spinal reference point.

[0056] Specifically, firstly, the total arc length of each of the quadratic Bézier curves constructed frame by frame in step 104 is calculated using numerical integration. Then, based on the number of spinal reference points (10), the total arc length of the curves is divided into equal segments. The parameter position corresponding to each segment on the Bézier curve is determined through iterative solving, resulting in equidistantly distributed spinal sampling points. These sampling points correspond one-to-one with the 10 spinal reference points from C4 to S1, avoiding measurement deviations caused by parameter nonlinearity and achieving sub-pixel-level vertebral localization. Subsequently, physical coordinate mapping is performed on the spinal sampling points. First, the Euclidean distance between the shoulder midpoint and hip midpoint in the current video frame is calculated as a scale factor. The relative position coordinates of the spinal reference points are multiplied by the scale factor and then added to the hip midpoint coordinates in the current frame to reconstruct the template physical coordinates of the spinal reference points in the current frame. The corresponding rotation matrix and translation vector are then obtained using the least-squares rigid registration algorithm. These matrices and vectors are used to spatially align the spinal sampling points with the physical coordinates of the template. Coordinate transformations are then performed on each spinal sampling point sequentially to obtain the final physical coordinates of each spinal reference point in the current video frame. Finally, the final physical coordinates of all spinal reference points in each frame of multiple consecutive video frames are arranged sequentially according to the playback time of the video frames, forming a sequence of spinal coordinates for the spinal reference points across multiple consecutive video frames. This provides continuous and accurate geometric coordinate data for subsequent calculations of spinal motion parameters.

[0057] As an optional embodiment, in step 105, the spine coordinate sequence corresponding to the spine reference point in multiple consecutive video frames is obtained by equidistant resampling of the quadratic Bézier curve and physical coordinate mapping, including: calculating the total arc length of the quadratic Bézier curve by numerical integration; performing equidistant resampling of the quadratic Bézier curve based on the total arc length to obtain spine sampling points that correspond one-to-one with the spine reference point; and obtaining and converting the spine sampling points from the real-time active video to obtain the spine coordinate sequence corresponding to the spine reference point in multiple consecutive video frames.

[0058] Specifically, in step 105, the total arc length of the quadratic Bézier curve is divided into equal segments based on the number of spinal reference points, and the arc length segment value corresponding to each spinal sampling point is determined. Through iteration, the parameter position corresponding to each arc length segment value on the quadratic Bézier curve is determined sequentially, and the corresponding sampling point is determined on the quadratic Bézier curve based on this parameter position. This process of determining sampling points corresponding to all arc length segment values ​​is continued until spinal sampling points are obtained that perfectly match the number of spinal reference points and correspond one-to-one in the order of the spinal physiological structure.

[0059] In this embodiment, the process of equidistant resampling and vertebral body center mapping involves obtaining 10 sampling points that match the bony measurement template to simulate the vertebral body center positions from C4 to S1. During the calculation, the total arc length of the quadratic Bézier curve is first obtained through numerical integration. Then follow the formula Determine the arc length value corresponding to each sampling point, and then iteratively solve for the parameter u, so that the arc length from the starting point of the curve to... The arc length is equal to This allows us to obtain the sampling points of the current video frame. ,These This refers to the two-dimensional image coordinates of the 10 estimated vertebral body center points in the current frame. This operation uses the curve arc length as the measurement basis, ensuring the uniform distribution of sampling points along the spine centerline, effectively avoiding measurement deviations caused by the nonlinear distribution of parameter u, achieving sub-pixel-level vertebral body localization, and improving the accuracy and consistency of spine reconstruction in dynamic video.

[0060] Further optionally, in an optional embodiment of step 105, obtaining and converting the spinal coordinate sequence corresponding to the spinal reference point in multiple consecutive video frames from the real-time activity video using spinal sampling points includes: calculating the Euclidean distance between the midpoint of the shoulder and the midpoint of the hip in the current video frame, and using the Euclidean distance as the scale factor in the current video frame; combining the scale factor in the current video frame with the relative position coordinates of the spinal reference point to restore the physical coordinates of the spinal reference point in the current video frame; solving for the corresponding rotation matrix and translation vector using the least squares rigid registration algorithm, and using the rotation matrix and translation vector to spatially align the coordinates of the spinal sampling point with the physical coordinates of the spinal reference point; performing coordinate transformation on each spinal sampling point sequentially according to the solved rotation matrix and translation vector to obtain the final physical coordinates corresponding to each spinal reference point in the current video frame; and arranging the final physical coordinates of all spinal reference points corresponding to each frame in multiple consecutive video frames in the order of video frame playback time to form the spinal coordinate sequence corresponding to the spinal reference point in multiple consecutive video frames.

[0061] The above process involves a rigid mapping between video reconstruction points and X-ray skeletal templates to achieve scale restoration and spatial alignment. First, the midpoint of the shoulder in the current video frame is calculated. Midpoint of the hip The Euclidean distance is used as a scale factor. The calculation formula is: Then, based on this scaling factor... Combined with the normalized coordinates of the spinal reference point According to the formula The physical coordinates of the spine reference point in the current frame are then reconstructed using a template. Next, the rotation matrix is ​​solved using a least-squares rigid registration algorithm (e.g., Umeyama's algorithm). Translation vector This makes the spinal sampling points physical coordinates of the template Achieve spatial alignment. Finally, follow the formula. After completing the coordinate transformation, the final physical coordinates of the center point of each vertebra in the current frame are obtained.

[0062] The rigid mapping process described above conveys absolute metrological significance to video measurements. By minimizing the registration mapping error through least squares optimization, it ensures that the spinal coordinates obtained from video measurements are consistent with the patient's actual bony structure. This is suitable for the accurate calculation of core clinical parameters such as the Cobb angle, greatly improving the reliability and clinical usability of dynamic spinal reconstruction. At the same time, it achieves the effective transfer of absolute scale, allowing the final spinal coordinate sequence to truly reflect the dynamic positional changes of the spinal bony structure.

[0063] Step 106: Calculate the spinal change trend of the measured object, the segmental angle time series matrix used to quantify the degree of local spinal deformation, and the activity symmetry parameter based on the spinal coordinate sequence.

[0064] Step 106 converts the spinal coordinate sequence into quantifiable and interpretable spinal motion parameters.

[0065] Specifically, in step 106 above, the spinal change trend is first calculated. Based on the spinal coordinate sequence, the spinal reference points corresponding to the upper and lower vertebrae are selected frame by frame, and their tilt vectors are calculated respectively. The angle between the two tilt vectors is substituted into the vector angle calculation function as the original Cobb angle. Its absolute value is taken as the Cobb angle of the current frame. Then, the direction of spinal curvature is determined by the cross product sign. The Cobb angles and corresponding curvature directions of multiple consecutive video frames are arranged in chronological order to form the spinal change trend of the measured object. Second, a segment angle time series matrix is ​​constructed. The coordinates of all spinal reference points are extracted from the spinal coordinate sequence frame by frame. According to the physiological structure of the spine, three adjacent spinal reference points are divided into a segment group. The absolute value of the vector angle between the first two reference points and the last two reference points in each group is calculated as the segment angle of that segment. All segment angles of each frame are arranged into a segment angle array according to the segment order. Then, the segment angle arrays of multiple consecutive video frames are arranged in frame order. A segment angle time series matrix is ​​constructed with spinal segments as rows and video frames as columns to quantify the degree of deformation of each local part of the spine. Finally, the activity symmetry parameter is calculated. The spinal segments are divided into left-side and right-side segment groups according to their physiological structure. The absolute value of the difference between the sum of the segment angles of the two groups is calculated frame by frame. The mean of the absolute values ​​of this difference for all video frames is calculated. The average value is the activity symmetry parameter. The value of this parameter is negatively correlated with the degree of symmetry of spinal activity, thus realizing the quantitative assessment of spinal activity symmetry.

[0066] As an optional embodiment, step 106 involves calculating the spinal change trend of the measured object, the segmental angle time series matrix for quantifying the degree of local spinal deformation, and the activity symmetry parameter based on the spinal coordinate sequence. This includes: for multiple consecutive video frames, calculating the spinal change trend of the measured object using the vector angle calculation function of the Cobb angle and the vector cross product sign, combined with the spinal coordinate sequence. The spinal trend includes the angle between the tilt vectors of the upper and lower vertebrae and the direction of spinal curvature of the measured object; calculating the segmental angle between adjacent segments in each video frame based on the spinal coordinate sequence, and forming a segmental angle time series matrix corresponding to multiple consecutive video frames; and statistically analyzing the left-right distribution of the segmental angle time series matrix to obtain the symmetry value as the activity symmetry parameter.

[0067] First, in the above embodiments, for multiple consecutive video frames, the spinal change trend of the measurement object is calculated by combining the vector angle calculation function of the Cobb angle and the vector cross product sign with the spinal coordinate sequence. Further optionally, for each video frame, spinal reference points corresponding to the upper and lower vertebrae are selected from the spinal coordinate sequence, and the first tilt vector of the upper vertebra and the second tilt vector of the lower vertebra are calculated respectively. Specifically, the upper and lower vertebrae are key bony landmarks for defining the core segments of scoliosis and calculating the Cobb angle (a core indicator for assessing the degree of scoliosis). They are the core reference vertebrae for quantitative measurement of scoliosis, and together they determine the upper and lower boundaries of the scoliosis arc. The upper vertebra refers to the uppermost vertebra in the scoliosis arc with the largest tilt angle towards the convex side of the curve. It is the landmark vertebra of the upper boundary of the scoliosis arc, and the tilt state of its vertebral endplate can intuitively reflect the degree of deformity in the upper segment of the scoliosis. The lower vertebra refers to the vertebra below which the angle of inclination towards the convex side of the scoliosis arc is greatest. It is the lower boundary landmark of the scoliosis arc and, together with the upper vertebra, defines the main affected segments of the scoliosis. In traditional clinical Cobb angle measurement, the angle between the vertical lines drawn from the superior endplate of the upper vertebra and the inferior endplate of the lower vertebra is used as the scoliosis angle. Accurate identification of the upper and lower vertebrae is a prerequisite for ensuring the accuracy of Cobb angle measurement.

[0068] Substituting the first and second tilt vectors into the Cobb angle calculation function, the angle between the first and second tilt vectors is obtained; the angle is the original Cobb angle in the current video frame, and the absolute value of the original Cobb angle is used as the Cobb angle of the current video frame; the sign of the cross product result of the upper and lower vertebral tilt vectors is determined by the sign of the cross product result, and the direction of spinal curvature of the measured object in the current video frame is determined according to the sign of the cross product result; the Cobb angles and corresponding spinal curvature directions of multiple consecutive video frames are arranged sequentially according to the time order of the video frames to form the trend of spinal change of the measured object.

[0069] For example, before the calculation, time-series data is read from the stored coordinate text file, specifically the frame sequence. Time series and point sequence Each of them An array consisting of the final physical coordinates of 10 spine reference points corresponding to a single frame. This method generates a dynamic temporal sequence of the spinal centerline, ensuring the continuity of measurements based on video frames, effectively avoiding data loss or noise interference, and providing a reliable temporal data foundation for subsequent calculations of various spinal parameters.

[0070] Based on this, the trend of spinal changes is calculated. Further, optionally, for each video frame, the corresponding spinal reference points of the upper and lower vertebrae are selected from the point sequence of the spinal coordinate sequence, and the first tilt vector of the upper vertebra and the second tilt vector of the lower vertebra are calculated respectively.

[0071] Then, the two tilt vectors are substituted into the vector angle calculation function of the Cobb angle to obtain the angle between them as the original Cobb angle of the current frame. The absolute value of the original Cobb angle is the Cobb angle of the current video frame. At the same time, the sign of the cross product result of the tilt vectors of the upper and lower vertebrae is determined by the sign of the cross product, and the direction of spinal curvature in the current frame is determined based on the sign of the result. Finally, the Cobb angles and corresponding spinal curvature directions of multiple consecutive video frames are arranged sequentially according to the time order of the video frames to form the trend of spinal changes of the measured object. In the video measurement scheme, the clinical standard Cobb method is used to measure the degree of scoliosis for Cobb angle calculation and left-right convexity direction determination. C4 to C6 are selected as the upper vertebrae, which is the core functional segment of the upper spine and the key monitoring area for dynamic deformation of the upper scoliosis; L5 to S1 are selected as the lower vertebrae, which is the connection between the lower spine and the pelvis and the key monitoring area for dynamic deformation of the lower scoliosis. The tilt vectors in the above embodiment are the tilt vectors of the upper vertebrae. ,in, Coordinates of the C4 vertebral body center The coordinates of the C6 vertebral body center are given, with the vector direction pointing from C4 to C6, representing the tilt direction of the upper segment of the spine (superior vertebral region). The tilt vector of the lower vertebrae... ,in, Coordinates of the center of L5 vertebral body Let L5 be the coordinates of the center of the S1 vertebral body, and let L5 be the vector direction pointing from S1, representing the tilt direction of the lower segment of the spine (lower vertebral region). Substitute vector angle calculation function ,in (correspond )and (correspond Let be the two vectors whose included angle is to be calculated. Then, the original Cobb angle is obtained by solving. Ultimately, Cobb's Corner and based on The positive and negative signs determine the direction of bending. ,Right now This process enables the traceability and measurement of the Cobb angle under dynamic video, and the determination of the curvature direction conforms to the clinical back observation standards. It supports continuous temporal tracking of the scoliosis direction, making the quantitative results of spinal change trends more in line with clinical assessment needs.

[0072] Furthermore, in the above embodiments, the segment angles between adjacent segments in each video frame are calculated based on the spinal coordinate sequence, forming a segment angle time series matrix corresponding to multiple consecutive video frames. Optionally, for each video frame, the final physical coordinates of all spinal reference points in the current video frame are extracted from the spinal coordinate sequence. Three adjacent spinal reference points are divided into a spinal segment group according to the physiological structure of the spine, and segment groups corresponding to all adjacent spinal segments are obtained sequentially. For each spinal segment group, the first vector of the first two spinal reference points and the second vector of the last two spinal reference points are calculated respectively. The angle between the first and second vectors is solved using a vector angle calculation function, and the absolute value of the angle between the first and second vectors is used as the segment angle of the corresponding adjacent spinal segment. After calculating the segment angles of all adjacent spinal segments in the current video frame, all segment angles are arranged into a segment angle array corresponding to the current video frame according to segment order. The segment angle arrays of multiple consecutive video frames are arranged in the time order of the video frames, and a matrix is ​​constructed with segments as rows and video frames as columns to obtain a segment angle time series matrix corresponding to multiple consecutive video frames.

[0073] Local curvature angles were calculated for eight adjacent segments of the spine, including C6 to T1 and T1 to T4. The degree of local deformation of the spine was quantified by the angle difference between adjacent vectors. This was done for a point sequence in each frame. Its segment angle ( The calculation method for ) is as follows: , , The segment angles of all frames are integrated to form an 8-row, N-column timing matrix. This measurement method originates from the X-ray bony measurement template, ensuring the accuracy of segmental angle calculation. At the same time, it realizes the temporal quantification of local deformation of multiple segments of the spine, making the assessment dimensions more comprehensive, supporting multidimensional assessment of the spine under dynamic video, and improving the quantitative accuracy of clinical diagnosis.

[0074] Finally, in the above embodiments, the left-right distribution of the segment angle time series matrix is ​​statistically analyzed to obtain a symmetry value as an activity symmetry parameter. Further optionally, when obtaining the symmetry value as the activity symmetry parameter, adjacent spinal segments in the segment angle time series matrix can be divided into a left segment group and a right segment group according to the physiological structure of the spine. For each video frame, the first summation value of all segment angles in the left segment group and the second summation value of all segment angles in the right segment group are calculated for the current video frame. The absolute value of the difference between the first and second summation values ​​is then calculated. The mean of the absolute values ​​of the corresponding differences for all video frames is calculated, and the obtained average value is used as the activity symmetry parameter of the measured object. The magnitude of the activity symmetry parameter is negatively correlated with the degree of symmetry of the spinal activity of the measured object.

[0075] For example, the symmetry exponent σ is calculated as follows: This method enables the quantitative assessment of spinal mobility symmetry. Through the aforementioned full-process calculations, the system can accurately extract Cobb angle sequences from the real-time activity video of the measured object. Spinal curvature direction sequence Segment angle time series matrix This allows for multi-parameter time-series measurement of spinal motion status, providing a more accurate quantitative data foundation for scenarios such as spinal rehabilitation training monitoring and clinical posture assessment.

[0076] Step 107: Convert the spinal change trend of the measured object and the segment angle time series matrix into a visualized video measurement report and push it to the measured object and / or relevant users.

[0077] In step 107, firstly, based on the trend of spinal changes, a time series plot is drawn. x-axis, Cobb angle plot the dynamic Cobb angle time series curve with the vertical axis as the ordinate. Different colors are used to code and label the direction of spinal curvature. For example, color coding is used to distinguish the direction of spinal curvature, ensuring that clinicians can quickly identify changes in the direction of scoliosis. See [link to documentation]. Figure 4 As shown. Furthermore, the point of maximum lateral bending during the measurement period is marked on the curve. And the corresponding Cobb angle value. Wherein, , The time point corresponding to the maximum lateral bending. This represents the maximum Cobb angle value. This visualization method, generated based on frame-by-frame measurement data, preserves temporal continuity while highlighting key monitoring points, significantly improving the intuitiveness of the report and diagnostic efficiency, allowing physicians to quickly grasp the dynamic changes in the subject's scoliosis.

[0078] Simultaneously, key nodes of dynamic posture are captured through static charts, enabling a direct comparison of the degree of scoliosis. For example, spinal posture data from the starting frame (frame 1) and the frame with the maximum scoliosis (frame m) of a real-time video are extracted. Corresponding spinal centerline curves for both frames are plotted, and tilt markers for the upper and lower vertebrae are overlaid. The Cobb angle values ​​and spinal curvature directions for each frame are labeled, generating a comparison chart of the starting and maximum scoliosis postures. This is done for the posture point sequence of each frame. Drawing from arrive The spinal centerline curve is used to completely recreate the spinal morphology in that frame. Simultaneously, tilted marker lines for the upper and lower vertebrae are overlaid, with red arrows at the top, indicating the direction from... Along the upper vertebral vector Extend. The lower arrow is blue, from... Along the lower vertebral vector Extend in the reverse direction and clearly mark the Cobb angle values ​​of the corresponding frames in the chart. With bending direction (Right convexity, left convexity). This maximum scoliosis posture comparison chart statically freezes key postures in dynamic videos, creating a clear contrast between the initial and most severe states of scoliosis. This provides intuitive visualization for efficacy assessment and rehabilitation plan adjustments, supporting quantitative comparisons of pre- and post-rehabilitation effects. For example, the maximum scoliosis posture comparison chart can... Figure 5 As shown, the initial posture is the spinal centerline curve corresponding to the first frame, and the maximum lateral bending posture is the spinal centerline curve corresponding to the 671st frame.

[0079] Secondly, based on the segment angle time-series matrix, with the video frame sequence as the horizontal axis and the adjacent segments of the spine as the vertical axis, the bending angle intensity of each segment is encoded by color gradient to generate a local bending angle heatmap of the spine segments, which intuitively shows the dynamic bending distribution of each segment during the measurement period.

[0080] by Figure 6 The illustrated thermal image of local curvature angles of the spinal segments is used as an example, with the segment angle time series matrix as an example. transpose matrix Based on this, a graph is constructed where the horizontal axis corresponds to the video frame sequence (time dimension), and the vertical axis corresponds to eight adjacent segments of the spine (labeled as C6-T1, T1-T4, T4-T7, T7-T10, T10-L1, L1-L3, L3-L5, L5-S1). A heatmap color mapping (cmap=hot) is used to encode the bending angle intensity of each segment, with local angles marked by color bars. The numerical range is indicated by the color. Darker colors represent larger curvature angles in that segment, visually highlighting the dynamic curvature distribution of each spinal segment during the measurement period. This helps physicians identify key segments with asymmetrical activity, providing precise data support for local rehabilitation training guidance. Simultaneously, the spinal segment curvature angle heatmap provides a multi-parameter spatiotemporal view, ensuring the comprehensiveness and clinical value of the report.

[0081] Subsequently, all quantitative data on spinal change trends and segmental angle time-series matrices were integrated to generate a standardized metrological document. This document includes an overview of the measurements, detailed frame-by-frame measurement data, and parameters related to measurement uncertainty. The overview of the measurements includes core indicators such as the average Cobb angle, the maximum Cobb angle, and the degree of improvement in spinal curvature. Next, the completed dynamic Cobb angle time-series curves, the comparison diagram of the initial and maximum scoliosis postures, and the heatmap of local curvature angles of spinal segments were integrated with the standardized metrological document to form a complete video measurement report containing visual charts and quantitative data. The report data is traceable, reproducible, and conforms to clinical metrological standards.

[0082] Finally, the complete video measurement report is sent to the terminal devices of the measurement subject and / or relevant users through a preset push channel. In practical applications, the preset push channel may further include, but is not limited to, in-app messages, emails, and web-based report links. Relevant users may include rehabilitation physicians, medical staff, and remote consultation experts, thereby meeting the needs for using spinal measurement results in multiple scenarios.

[0083] In this embodiment, the relative position coordinates of the spinal reference points are obtained by acquiring the orthostatic posture image of the measured object and normalizing it. Then, real-time activity video is acquired and the two-dimensional image coordinates of key points of the shoulder and hip are extracted. Based on the above coordinates, a quadratic Bézier curve is constructed to simulate the spinal centerline. The spinal coordinate sequence is obtained through equidistant resampling and physical coordinate mapping. Then, the spinal change trend, segment angle time series matrix and activity symmetry parameters are calculated. Finally, a visual video measurement report is generated and pushed. This scheme can realize dynamic and continuous video quantitative measurement of the internal activity state of the human spine, obtain traceable and standardized spinal motion quantitative parameters and visual video measurement reports, thereby reducing human subjective error and measurement deviation in the process of spinal posture measurement and rehabilitation assessment, bringing a better experience to users, reducing the cost for medical staff to carry out spinal rehabilitation training monitoring, clinical assessment, remote consultation and other work, and improving the efficiency of spinal posture measurement and rehabilitation assessment.

[0084] The above describes a video measurement method based on human posture in the embodiments of this application. The following describes the video measurement device based on human posture that performs the above-described video measurement method based on human posture.

[0085] See Figure 7 ,like Figure 7 The diagram shows a structural schematic of a video measurement device based on human posture. The video measurement device based on human posture in this embodiment can achieve the above-described... Figure 1 The steps of the human posture-based video measurement method executed in the corresponding embodiments are described above. The functions of the human posture-based video measurement device can be implemented in hardware or by executing corresponding software within hardware. The hardware or software includes one or more modules corresponding to the above functions, and these modules can be software and / or hardware. The human posture-based video measurement device may include an input / output module 701 and a processing module 702. The functional implementation of the processing module 702 and the input / output module 701 can be found in [reference missing]. Figure 1 The operations performed in the corresponding embodiments will not be described in detail here. For example, the processing module 702 can be used to control the sending, receiving, and acquisition operations of the input / output module 701. The input / output module 701 is configured to acquire the orthogonal posture image of the measured object; The processing module 702 is configured to: annotate the original vertebral coordinates of the spinal reference points of the measured object in the orthogonal posture image, and normalize them to obtain the relative position coordinates of the spinal reference points; acquire real-time activity videos of the measured object, and extract the two-dimensional image coordinates of shoulder key points and hip key points from the real-time activity videos in real time through a posture estimation model; construct a quadratic Bézier curve based on the relative position coordinates of the spinal reference points and the two-dimensional image coordinates of the shoulder and hip key points to simulate the natural curvature path of the spinal centerline of the measured object in multiple consecutive video frames; obtain the spinal coordinate sequence corresponding to the spinal reference points in multiple consecutive video frames by equidistant resampling of the quadratic Bézier curves and physical coordinate mapping; and calculate the spinal change trend of the measured object, the segmental angle time series matrix used to quantify the degree of local spinal deformation, and the activity symmetry parameter based on the spinal coordinate sequence. The input / output module 701 is also configured to convert the spinal change trend of the measured object and the segment angle time series matrix into a visualized video measurement report and push it to the measured object and / or relevant users.

[0086] In some optional embodiments, the processing module 702, which normalizes the relative position coordinates of the spinal reference point, is configured to: extract the trunk physiological feature points of the measurement object in the orthopedic posture from the orthopedic posture image; the trunk physiological feature points include the reference shoulder midpoint and the reference hip midpoint; and normalize the original vertebral coordinates based on the original pixel coordinates of the trunk physiological feature points to obtain the relative position coordinates of the spinal reference point.

[0087] In some optional embodiments, the processing module 702, using the midpoint of the shoulder as the starting point and the midpoint of the hip as the ending point in the current state of the measured object, and combining the waist control points of the measured object to construct a quadratic Bézier curve, is configured as follows: Calculate the midpoint of the shoulder in the current video frame based on the two-dimensional image coordinates of the shoulder key points; the shoulder key points include the left and right shoulders; calculate the midpoint of the hip in the current video frame based on the two-dimensional image coordinates of the hip key points; the shoulder key points include the left and right hip joints; calculate the trunk vector and trunk vector magnitude based on the two-dimensional image coordinates of the midpoint of the shoulder and the midpoint of the hip in the current video frame; calculate the lateral offset by combining the trunk tilt angle, trunk vector, and trunk vector magnitude of the measured object; add the product of the midpoint of the hip, a preset coefficient, and the trunk vector, and the lateral offset to obtain the waist control point in the current video frame; construct a quadratic Bézier curve using the midpoint of the shoulder in the current video frame as the starting point, the midpoint of the hip in the current video frame as the ending point, and the waist control point in the current video frame as the control point, wherein the control point is used to control the curvature of the curve.

[0088] In some optional embodiments, the processing module 702, by equidistant resampling of the quadratic Bézier curve and physical coordinate mapping, obtains the spine coordinate sequence corresponding to the spine reference point in multiple consecutive video frames. This is configured to: calculate the total arc length of the quadratic Bézier curve through numerical integration; perform equidistant resampling of the quadratic Bézier curve based on the total arc length to obtain spine sampling points that correspond one-to-one with the spine reference point; and obtain and convert the spine coordinate sequence corresponding to the spine reference point in multiple consecutive video frames from the real-time video using the spine sampling points.

[0089] In some optional embodiments, the processing module 702, which obtains and transforms the spinal coordinate sequence corresponding to the spinal reference point in multiple consecutive video frames from the real-time active video through spinal sampling points, is configured to: calculate the Euclidean distance between the midpoint of the shoulder and the midpoint of the hip in the current video frame, and use the Euclidean distance as the scale factor in the current video frame; combine the scale factor in the current video frame with the relative position coordinates of the spinal reference point to restore the physical coordinates of the spinal reference point in the current video frame; solve for the corresponding rotation matrix and translation vector using the least squares rigid registration algorithm, and use the rotation matrix and translation vector to spatially align the coordinates of the spinal sampling point with the physical coordinates of the spinal reference point; perform coordinate transformation on each spinal sampling point according to the solved rotation matrix and translation vector to obtain the final physical coordinates corresponding to each spinal reference point in the current video frame; and arrange the final physical coordinates of all spinal reference points corresponding to each frame in multiple consecutive video frames in the order of video frame playback time to form the spinal coordinate sequence corresponding to the spinal reference point in multiple consecutive video frames.

[0090] In some optional embodiments, the processing module 702, which calculates the spinal change trend of the measured object, the segmental angle time series matrix for quantifying the degree of local spinal deformation, and the activity symmetry parameter based on the spinal coordinate sequence, is configured to: calculate the spinal change trend of the measured object for multiple consecutive video frames using the vector angle calculation function of Cobb angle and the vector cross product sign, combined with the spinal coordinate sequence; the spinal trend includes the angle between the tilt vectors of the upper and lower vertebrae and the direction of spinal curvature of the measured object; calculate the segmental angle between adjacent segments in each video frame based on the spinal coordinate sequence, and form a segmental angle time series matrix corresponding to multiple consecutive video frames; and statistically analyze the left-right distribution of the segmental angle time series matrix to obtain the symmetry value as the activity symmetry parameter.

[0091] In some optional embodiments, the processing module 702, for multiple consecutive video frames, calculates the spinal change trend of the measurement object by combining the vector angle calculation function of Cobb angle and the cross product sign with the spinal coordinate sequence. The configuration is as follows: For each video frame, selects the corresponding upper and lower vertebral reference points from the spinal coordinate sequence, and calculates the first tilt vector of the upper vertebra and the second tilt vector of the lower vertebra, respectively; substitutes the first and second tilt vectors into the vector angle calculation function of Cobb angle to obtain the angle between the first and second tilt vectors; the angle is the original Cobb angle in the current video frame, and the absolute value of the original Cobb angle is used as the Cobb angle of the current video frame; the sign of the cross product result of the upper and lower vertebral tilt vectors is determined by the cross product sign, and the spinal curvature direction of the measurement object in the current video frame is determined based on the sign of the cross product result; the Cobb angles and corresponding spinal curvature directions of multiple consecutive video frames are arranged sequentially according to the time order of the video frames to form the spinal change trend of the measurement object.

[0092] In some optional embodiments, the processing module 702, which calculates the segment angles between adjacent segments in each video frame based on the spinal coordinate sequence and forms a segment angle time series matrix corresponding to multiple consecutive video frames, is configured as follows: For each video frame, extract the final physical coordinates of all spinal reference points in the current video frame from the spinal coordinate sequence; divide three adjacent spinal reference points into a spinal segment group according to the order of the spinal physiological structure; sequentially obtain the segment groups corresponding to all adjacent spinal segments; for each spinal segment group, calculate the first vector of the first two spinal reference points and the second vector of the last two spinal reference points; solve the angle between the first and second vectors using a vector angle calculation function; use the absolute value of the angle between the first and second vectors as the segment angle of the corresponding adjacent spinal segment; after completing the segment angle calculation of all adjacent spinal segments in the current video frame, organize all segment angles into a segment angle array corresponding to the current video frame according to the segment order; arrange the segment angle arrays of multiple consecutive video frames according to the time order of the video frames, construct a matrix with segments as rows and video frames as columns, and obtain a segment angle time series matrix corresponding to multiple consecutive video frames.

[0093] In some optional embodiments, the processing module 702, which statistically analyzes the left-right distribution of the segment angle time series matrix to obtain a symmetry value as an activity symmetry parameter, is configured to: divide adjacent spinal segments in the segment angle time series matrix into a left segment group and a right segment group according to the physiological structure of the spine; for each video frame, calculate the first summation value of all segment angles in the left segment group and the second summation value of all segment angles in the right segment group in the current video frame, and solve for the absolute value of the difference between the first summation value and the second summation value; calculate the mean of the absolute values ​​of the corresponding differences for all video frames, and use the obtained average value as the activity symmetry parameter of the measured object; the magnitude of the activity symmetry parameter is negatively correlated with the degree of symmetry of the spinal activity of the measured object.

[0094] In this embodiment, the video measurement device based on human posture uses video measurement to achieve dynamic measurement of spinal motion state, effectively avoiding measurement errors caused by static images and experience judgment during spinal motion state assessment, improving measurement accuracy and efficiency, and ensuring measurement precision and real-time measurement results.

[0095] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0096] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, apparatuses, or modules, and may be electrical, mechanical, or other forms.

[0097] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0098] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0099] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0100] The computer program product includes one or more computer instructions. When the computer program is loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state disk (SSD)).

[0101] The technical solutions provided in the embodiments of this application have been described in detail above. Specific examples have been used in the embodiments of this application to illustrate the principles and implementation methods of the embodiments of this application. The description of the above embodiments is only for the purpose of helping to understand the methods and core ideas of the embodiments of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the embodiments of this application. Therefore, the content of this specification should not be construed as a limitation on the embodiments of this application.

Claims

1. A video measurement method based on human posture, characterized in that, The method includes: Acquire the orthogonal posture image of the object being measured; Based on the positive posture image, the original vertebral coordinates of the spinal reference point of the measured object in the positive posture are labeled, and the relative position coordinates of the spinal reference point are obtained by normalization. Acquire real-time activity video of the measurement object, and extract two-dimensional image coordinates of shoulder key points and hip key points from the real-time activity video in real time using a pose estimation model; Based on the relative position coordinates of the spinal reference point, the two-dimensional image coordinates of the shoulder key point and the hip key point, a quadratic Bézier curve is constructed to simulate the natural curvature path of the spinal centerline of the measured object in multiple consecutive video frames. By resampling the quadratic Bézier curve at equal intervals and mapping the physical coordinates, the spine coordinate sequence corresponding to the spine reference point in multiple consecutive video frames is obtained. The spinal change trend of the measured object, the segmental angle temporal matrix used to quantify the degree of local spinal deformation, and the activity symmetry parameter are calculated based on the spinal coordinate sequence. This includes: for multiple consecutive video frames, calculating the spinal change trend of the measured object using the Cobb angle vector angle calculation function and the vector cross product sign, combined with the spinal coordinate sequence. The spinal trend includes the angle between the tilt vectors of the upper and lower vertebrae and the direction of spinal curvature of the measured object; calculating the segmental angle between adjacent segments in each video frame based on the spinal coordinate sequence, forming a segmental angle temporal matrix corresponding to multiple consecutive video frames; and statistically analyzing the left-right distribution of the segmental angle temporal matrix to obtain a symmetry value as an activity symmetry parameter. The spinal change trend of the measured object and the segment angle time series matrix are converted into a visualized video measurement report and pushed to the measured object and / or relevant users; The step of calculating the spinal change trend of the measurement object for multiple consecutive video frames, using the Cobb angle vector angle calculation function and the vector cross product sign, combined with the spinal coordinate sequence, includes: for each video frame, selecting the corresponding upper and lower vertebral reference points from the spinal coordinate sequence, and calculating the first tilt vector of the upper vertebra and the second tilt vector of the lower vertebra respectively; substituting the first and second tilt vectors into the Cobb angle vector angle calculation function to obtain the angle between the first and second tilt vectors; the angle is the original Cobb angle in the current video frame, and the absolute value of the original Cobb angle is used as the Cobb angle of the current video frame; determining the sign of the cross product result of the upper and lower vertebral tilt vectors using the vector cross product sign, and determining the spinal curvature direction of the measurement object in the current video frame based on the sign of the cross product result; and arranging the Cobb angles and corresponding spinal curvature directions of multiple consecutive video frames in chronological order to form the spinal change trend of the measurement object.

2. The video measurement method based on human posture according to claim 1, characterized in that, The normalization process yields the relative position coordinates of the spinal reference point, including: Extract trunk physiological feature points of the measured object in the orthopedic posture from the orthopedic posture image; the trunk physiological feature points include the reference mid-shoulder point and the reference mid-hip point; Based on the original pixel coordinates of the trunk physiological feature points, the original vertebral coordinates are normalized to obtain the relative position coordinates of the spinal reference points.

3. The video measurement method based on human posture according to claim 1, characterized in that, The construction of a quadratic Bézier curve based on the relative position coordinates of the spinal reference point, the two-dimensional image coordinates of the shoulder key points and the hip key points includes: Calculate the midpoint of the shoulder in the current video frame based on the two-dimensional image coordinates of the shoulder key points; the shoulder key points include the left and right shoulders; Calculate the midpoint of the hip in the current video frame based on the two-dimensional image coordinates of the hip key points; shoulder key points include the left hip joint and the right hip joint; Based on the two-dimensional image coordinates of the midpoint of the shoulder and midpoint of the hip in the current video frame, calculate the trunk vector and the trunk vector magnitude. The lateral offset is calculated by combining the torso tilt angle, torso vector, and torso vector magnitude of the measured object. The waist control point in the current video frame is obtained by adding the product of the hip midpoint, the preset coefficient and the torso vector and the lateral offset. A quadratic Bézier curve is constructed using the midpoint of the shoulder in the current video frame as the starting point, the midpoint of the hip in the current video frame as the ending point, and the control point of the waist in the current video frame as the control point. The control points are used to control the curvature of the curve.

4. The video measurement method based on human posture according to claim 1, characterized in that, The process of obtaining the spine coordinate sequence corresponding to the spine reference point in multiple consecutive video frames by equidistant resampling of the quadratic Bézier curve and physical coordinate mapping includes: The total arc length of a quadratic Bézier curve is calculated using numerical integration. Based on the total arc length of the curve, the quadratic Bézier curve is resampled at equal intervals to obtain spinal sampling points that correspond one-to-one with the spinal reference point. The spinal coordinate sequence corresponding to the spinal reference point in multiple consecutive video frames is obtained and converted from the real-time activity video by sampling the spinal points.

5. The video measurement method based on human posture according to claim 4, characterized in that, The step of obtaining and converting the spinal coordinate sequence corresponding to the spinal reference point in multiple consecutive video frames from the real-time active video through spinal sampling points includes: Calculate the Euclidean distance between the midpoint of the shoulder and the midpoint of the hip in the current video frame, and use the Euclidean distance as the scale factor in the current video frame; By combining the scale factor in the current video frame with the relative position coordinates of the spine reference point, the physical coordinates of the spine reference point in the current video frame can be reconstructed. The corresponding rotation matrix and translation vector are obtained by solving the least squares rigid registration algorithm. The rotation matrix and translation vector are then used to spatially align the coordinates of the spinal sampling points with the physical coordinates of the spinal reference points. Based on the obtained rotation matrix and translation vector, coordinate transformation is performed on each spine sampling point in sequence to obtain the final physical coordinates of each spine reference point in the current video frame. The final physical coordinates of all spinal reference points corresponding to each frame in multiple consecutive video frames are arranged sequentially according to the playback time of the video frames, forming a sequence of spinal coordinates corresponding to the spinal reference points in multiple consecutive video frames.

6. The video measurement method based on human posture according to claim 1, characterized in that, The step of calculating the segment angles between adjacent segments in each video frame based on the spinal coordinate sequence, and forming a segment angle temporal matrix corresponding to multiple consecutive video frames, includes: For each video frame, the final physical coordinates of all spinal reference points in the current video frame are extracted from the spinal coordinate sequence. The three adjacent spinal reference points are divided into a spinal segment group according to the order of the physiological structure of the spine, and the segment groups corresponding to all adjacent spinal segments are obtained in turn. For each spinal segment group, calculate the first vector of the first two spinal reference points and the second vector of the last two spinal reference points. Solve the angle between the first and second vectors using the vector angle calculation function. Use the absolute value of the angle between the first and second vectors as the segment angle of the corresponding adjacent spinal segments. After calculating the segment angles of all adjacent segments of the spine in the current video frame, organize all segment angles into an array of segment angles corresponding to the current video frame in segment order; Arrange the segment angle arrays of multiple consecutive video frames in the time order of the video frames, construct a matrix with segments as rows and video frames as columns, and obtain the segment angle time sequence matrix corresponding to multiple consecutive video frames.

7. A video measurement device based on human posture, characterized in that, The device includes: The input / output module is configured to acquire an orthogonal orientation image of the measured object; The processing module is configured to: label the original vertebral coordinates of the spinal reference points of the measured object in the orthostatic posture image, and normalize them to obtain the relative position coordinates of the spinal reference points; acquire real-time activity videos of the measured object, and extract the two-dimensional image coordinates of shoulder key points and hip key points from the real-time activity videos in real time through a posture estimation model; construct quadratic Bézier curves based on the relative position coordinates of the spinal reference points and the two-dimensional image coordinates of the shoulder and hip key points to simulate the natural curvature path of the spinal centerline of the measured object in multiple consecutive video frames; obtain the spinal coordinate sequence corresponding to the spinal reference points in multiple consecutive video frames by equidistant resampling of the quadratic Bézier curves and physical coordinate mapping; and calculate the spinal change trend of the measured object, the segmental angle time series matrix used to quantify the degree of local spinal deformation, and the activity symmetry parameters based on the spinal coordinate sequence. The processing module, when calculating the spinal change trend of the measured object, the segmental angle time series matrix used to quantify the degree of local spinal deformation, and the activity symmetry parameter based on the spinal coordinate sequence, is configured to calculate the spinal change trend of the measured object for multiple consecutive video frames using the vector angle calculation function of Cobb angle and the vector cross product sign, combined with the spinal coordinate sequence. The spinal trend includes the angle between the tilt vectors between the upper and lower vertebrae and the direction of spinal curvature of the measured object; calculate the segmental angle between adjacent segments in each video frame based on the spinal coordinate sequence, and form a segmental angle time series matrix corresponding to multiple consecutive video frames; and statistically analyze the left-right distribution of the segmental angle time series matrix to obtain the symmetry value as the activity symmetry parameter. The processing module, when calculating the spinal change trend of the measured object using the Cobb angle vector angle calculation function and the vector cross product sign in combination with the spinal coordinate sequence for multiple consecutive video frames, is configured to: select the corresponding upper and lower vertebral reference points from the spinal coordinate sequence for each video frame; calculate the first tilt vector of the upper vertebra and the second tilt vector of the lower vertebra respectively; substitute the first and second tilt vectors into the Cobb angle vector angle calculation function to obtain the angle between the first and second tilt vectors; the angle is the original Cobb angle in the current video frame, and the absolute value of the original Cobb angle is used as the Cobb angle of the current video frame; determine the sign of the cross product result of the upper and lower vertebral tilt vectors by the vector cross product sign, and determine the spinal curvature direction of the measured object in the current video frame based on the sign of the cross product result; and arrange the Cobb angles and corresponding spinal curvature directions of multiple consecutive video frames in the time sequence of the video frames to form the spinal change trend of the measured object. The input / output module is also configured to convert the spinal change trend of the measured object and the segment angle time series matrix into a visualized video measurement report and push it to the measured object and / or relevant users.

8. A computer-readable storage medium, characterized in that, Includes instructions that, when run on a computer, cause the computer to perform the video measurement method based on human posture as described in any one of claims 1-6.

Images