Search

CN-122023250-A - Scoliosis screening method, device, equipment and medium based on pure vision

CN122023250ACN 122023250 ACN122023250 ACN 122023250ACN-122023250-A

Abstract

The invention provides a pure vision-based scoliosis screening method, device, equipment and medium, which comprise the steps of collecting back images of a human body under two set shooting postures, extracting back key points by adopting a HRNet algorithm, calculating the inclination angles of back 4 groups of symmetrical points, obtaining symmetrical point inclination angle scores after normalization, extracting back left and right boundary curves by utilizing the back key points, calculating curvature difference standardization scores, smoothness standardization scores and DTW distance standardization scores, extracting back contours by adopting a YOLOV-Seg algorithm, extracting contour top curves, fitting approximate straight lines, calculating the inclination angles of the approximate straight lines, carrying out standardization according to a first priori threshold, converting the standardized back images into Adam forward-bending posture scores, combining analysis results, calculating to obtain scoliosis scores, and realizing accurate quantitative assessment of the back postures and the top curves of the human body without depending on special hardware such as infrared sensors, depth cameras, medical equipment and the like.

Inventors

  • GONG YIFEI
  • ZHANG DENGPAN
  • ZHANG ZHUMING

Assignees

  • 恒鸿达(福建)体育科技有限公司

Dates

Publication Date
20260512
Application Date
20251225

Claims (10)

  1. 1. A scoliosis screening method based on pure vision is characterized by comprising the following steps: step 1, acquiring back images of a human body under two set shooting postures; extracting back key points by adopting HRNet algorithm, calculating the inclination angles of 4 groups of symmetrical points of the back, and obtaining symmetrical point inclination angle scores after normalizing according to a second priori threshold value; step 3, extracting a back contour by adopting YOLOV-Seg algorithm, extracting a contour top curve, fitting an approximate straight line, calculating the inclination angle of the approximate straight line, and converting the inclination angle into an Adam anteversion gesture score after standardization according to a first priori threshold; And step 4, combining the analysis results of the step 2 and the step 3, and calculating to obtain a scoliosis score.
  2. 2. The method for screening scoliosis based on pure vision according to claim 1, wherein the step1 specifically comprises the steps of collecting back images of a human body in two set shooting postures, wherein the set shooting postures and the corresponding camera positions are as follows: the back standing posture is that the feet of the shot object are gathered together, the shoulders are relaxed, the two arms naturally drop or the two hands are crossly placed in front of the stride, and the camera is aligned to the midpoint of the connecting line of the lower corner of the scapula; Adam forward-bending gesture, namely, the knee of a shooting object is straightened, the arms are drooped, the back is parallel to the ground, a camera is opposite to the highest bulge of the back, the resolution of the image acquired by the camera is more than or equal to 1920 multiplied by 1080 pixels, the frame rate is more than or equal to 30fps, and the lens distortion rate is less than or equal to 1%.
  3. 3. The method for screening scoliosis based on pure vision according to claim 1, wherein the step 2 is characterized in that key points of the back are extracted by adopting HRNet algorithm, the inclination angles of 4 groups of symmetrical points of the back are calculated, the inclination angle scores of the symmetrical points are obtained after the calibration is carried out according to a second priori threshold, and the 4 groups of symmetrical points of the back comprise left and right shoulder peak points, left and right scapula lower corner points, left and right thoracic spine protrusion points, left and right posterior upper iliac spine points; Extracting left and right boundary curves of the back by using key points of the back, and calculating curvature difference standardization score, smoothness standardization score and DTW distance standardization score, wherein: And (3) calculating the maximum curvature, namely fitting the discrete key points into a continuous curve through cubic spline interpolation, calculating the curvature of each point on the curve based on a curvature formula, taking the maximum value as a quantization index of the bending degree, and carrying out moving average filtering denoising treatment on the discrete points of the discrete boundary curve before calculating the maximum curvature, wherein the size of a sliding window is 5 pixels, and the curvature formula is as follows: ; The curve is characterized by comprising a curve, a parameter variable, x, a y-axis, curv max , a maximum curvature of the back boundary curve, i.e. a curvature maximum value of all points in the range of 0,1, wherein the curve is used for quantifying the curvature of a certain point on the curve, and the larger value is used for representing the curve to be bent more obviously at the point; the curvature difference is calculated based on the maximum curvature of the left and right back curves by the formula: ; Wherein curv diff is the curvature difference of the left and right back boundary curves, curv max,left is the maximum curvature of the left back boundary curve, and curv max,right is the maximum curvature of the right back boundary curve; Calculating a curvature difference and normalizing according to a third prior threshold value to obtain a curvature difference normalization score; Smoothness calculation by the formula: ; The curve degree-of-curvature analysis method comprises the steps of delta curv k , namely, calculating the curvature change rate between a kth sampling point and a kth+1th sampling point, wherein the curvature change rate is used for quantifying the mutation condition of the bending degree of two adjacent points of a curve, wherein k is the sequence number of the sampling point, the value range is 1-99, curvatures (t k+1 ) is the curvature of the kth+1th sampling point, curvatures (t k ) is the curvature of the kth sampling point, smoothness is the smoothness of a back boundary curve, is used for quantifying the continuity degree of the curve to calculate the average value of the curvature change rate, and is standardized according to a fourth priori threshold value to obtain a smoothness standardization score; And calculating the DTW distance, namely calculating the shape distance of the left back curve and the right back curve through a dynamic time warping algorithm, normalizing according to a fifth priori threshold value to obtain a DTW distance normalization score, and setting Sakoe-Chiba constraint windows with the width of 10% of the sequence length by adopting Euclidean distance as a measure mode of curvature difference of sampling points during the calculation of the DTW distance.
  4. 4. The method for screening scoliosis based on pure vision according to claim 1, wherein the step 3 is characterized in that a YOLOV-Seg algorithm is adopted to extract a back contour, a contour top curve is extracted and an approximate straight line is fitted, the inclination angle of the approximate straight line is calculated, the contour top curve is standardized according to a first priori threshold and then is converted into an Adam forward-flexion gesture score, when the contour top curve is fitted and the approximate straight line is fitted, a least square method is adopted, the regional weight of the middle point of the curve is 1.0, and the regional weight of the two ends is 0.5; The step 4 specifically comprises the steps of carrying out weight distribution on the symmetrical point inclination angle score, the curvature difference standardization score, the smoothness standardization score, the DTW distance standardization score and the Adam anteflexion posture score under the back standing posture, then carrying out weighted summation, obtaining the range of [0,1], and carrying out scoliosis screening according to the preset interval of the obtained scores.
  5. 5. The scoliosis screening device based on pure vision is characterized by comprising the following components: the image acquisition module is used for acquiring back images of a human body under two set shooting postures; The first characteristic analysis module is used for extracting back key points by adopting HRNet algorithm, calculating the inclination angles of 4 groups of symmetrical points of the back, and obtaining the inclination angle scores of the symmetrical points after normalization; The second feature analysis module is used for extracting a back contour by adopting YOLOV-Seg algorithm, extracting a contour top curve, fitting an approximate straight line, calculating the inclination angle of the approximate straight line, and converting the inclination angle into an Adam forward-bending gesture score after standardization according to a first priori threshold; And the spine judgment module is used for calculating and obtaining a scoliosis score by combining analysis results of the first characteristic analysis module and the second characteristic analysis module.
  6. 6. The scoliosis screening device based on pure vision according to claim 5, wherein the image acquisition module is specifically configured to acquire back images of a human body in two set shooting postures, and the set shooting postures and the corresponding camera positions are as follows: the back standing posture is that the feet of the shot object are gathered together, the shoulders are relaxed, the two arms naturally drop or the two hands are crossly placed in front of the stride, and the camera is aligned to the midpoint of the connecting line of the lower corner of the scapula; Adam forward-bending gesture, namely, the knee of a shooting object is straightened, the arms are drooped, the back is parallel to the ground, a camera is opposite to the highest bulge of the back, the resolution of the image acquired by the camera is more than or equal to 1920 multiplied by 1080 pixels, the frame rate is more than or equal to 30fps, and the lens distortion rate is less than or equal to 1%.
  7. 7. The scoliosis screening device based on pure vision according to claim 5, wherein the first feature analysis module is specifically configured to extract back key points by adopting HRNet algorithm, calculate inclination angles of back 4 groups of symmetrical points, and normalize according to a second prior threshold to obtain symmetrical point inclination angle scores, wherein the back 4 groups of symmetrical points comprise left and right shoulder peak points, left and right scapula lower corner points, left and right thoracic spine spinous points, left and right posterior iliac upper spine points; Extracting left and right boundary curves of the back by using key points of the back, and calculating curvature difference standardization score, smoothness standardization score and DTW distance standardization score, wherein: And (3) calculating the maximum curvature, namely fitting the discrete key points into a continuous curve through cubic spline interpolation, calculating the curvature of each point on the curve based on a curvature formula, taking the maximum value as a quantization index of the bending degree, and carrying out moving average filtering denoising treatment on the discrete points of the discrete boundary curve before calculating the maximum curvature, wherein the size of a sliding window is 5 pixels, and the curvature formula is as follows: ; The curve is characterized by comprising a curve, a parameter variable, x, a y-axis, curv max , a maximum curvature of the back boundary curve, i.e. a curvature maximum value of all points in the range of 0,1, wherein the curve is used for quantifying the curvature of a certain point on the curve, and the larger value is used for representing the curve to be bent more obviously at the point; the curvature difference is calculated based on the maximum curvature of the left and right back curves by the formula: ; Wherein curv diff is the curvature difference of the left and right back boundary curves, curv max,left is the maximum curvature of the left back boundary curve, and curv max,right is the maximum curvature of the right back boundary curve; Calculating a curvature difference and normalizing according to a third prior threshold value to obtain a curvature difference normalization score; Smoothness calculation by the formula: ; Wherein, delta curv k is the curvature change rate between the kth sampling point and the (k+1) th sampling point and is used for quantifying the abrupt change condition of the bending degree of two adjacent points of the curve, k is the sampling point serial number, the value range is 1 to 99, curvatures (t k+1 ) is the curvature of the (k+1) th sampling point, curvatures (t k ) is the curvature of the kth sampling point, smoothness is the smoothness of the back boundary curve and is used for quantifying the continuity degree of the curve Calculating an average value of the curvature change rate, and normalizing according to a fourth priori threshold value to obtain a smoothness normalization score; And calculating the DTW distance, namely calculating the shape distance of the left back curve and the right back curve through a dynamic time warping algorithm, normalizing according to a fifth priori threshold value to obtain a DTW distance normalization score, and setting Sakoe-Chiba constraint windows with the width of 10% of the sequence length by adopting Euclidean distance as a measure mode of curvature difference of sampling points during the calculation of the DTW distance.
  8. 8. The scoliosis screening device based on pure vision according to claim 5, wherein the second feature analysis module specifically comprises the steps of extracting a back profile by adopting YOLOV-Seg algorithm, extracting a profile top curve and fitting an approximate straight line, calculating an inclination angle of the approximate straight line, normalizing according to a first priori threshold, converting into an Adam forward-bending gesture score, and adopting a least square method when the profile top curve and fitting the approximate straight line, wherein the regional weight of the midpoint of the profile is 1.0, and the regional weight of the two ends is 0.5; The spine judgment module specifically comprises the calculation modes of a scoliosis score, namely, a symmetrical point inclination angle score, a curvature difference standardization score, a smoothness standardization score, a DTW distance standardization score and an Adam anteflexion posture score under a back standing posture are subjected to weight distribution, then weighted summation is carried out, the score range is 0,1, and scoliosis screening is carried out according to a preset score range.
  9. 9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any one of claims 1 to 4 when the program is executed by the processor.
  10. 10. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the method according to any one of claims 1 to 4.

Description

Scoliosis screening method, device, equipment and medium based on pure vision Technical Field The invention relates to the technical field of computer vision, in particular to a scoliosis screening method, a device, equipment and a medium based on pure vision. Background The scoliosis detection needs to be compatible with universality of large-scale screening, specificity of accurate diagnosis and use safety, and meanwhile, different scenes such as basic medical treatment, schools and the like are adapted, so that the core requirements of low cost, easiness in operation and objective results are met. The prior art comprises the following steps: 1. Traditional physical examination techniques (visual inspection, anteflexion test + scoliosis instrument); 2. traditional imaging detection techniques (X-ray examination, CT examination); 3. Novel detection technology (common ultrasonic detection, AI vision/back image detection and spine three-dimensional measuring instrument). The method has the defects that 1, the result highly depends on doctor experience or professional medical equipment, 2, the radiation hazard is remarkable, and 3, the technology does not realize the balance of accuracy, safety and universality. Disclosure of Invention The invention aims to solve the technical problem of providing a pure vision-based scoliosis screening method, a device, equipment and a medium, which realize accurate quantitative evaluation of the back gesture and the top curve of a human body, do not need to rely on special hardware such as an infrared sensor, a depth camera and the like, can complete the evaluation only by a common RGB camera, and reduce the hardware cost by more than 90 percent. In a first aspect, the invention provides a pure vision-based scoliosis screening method, comprising the following steps: step 1, acquiring back images of a human body under two set shooting postures; Extracting back key points by adopting HRNet algorithm, calculating the inclination angles of back 4 groups of symmetrical points, and obtaining symmetrical point inclination angle scores after normalization; step 3, extracting a back contour by adopting YOLOV-Seg algorithm, extracting a contour top curve, fitting an approximate straight line, calculating the inclination angle of the approximate straight line, and converting the inclination angle into an Adam anteversion gesture score after standardization according to a first priori threshold; And step 4, combining the analysis results of the step 2 and the step 3, and calculating to obtain a scoliosis score. In a second aspect, the present invention provides a pure vision-based scoliosis screening device, comprising: the image acquisition module is used for acquiring back images of a human body under two set shooting postures; The first characteristic analysis module is used for extracting back key points by adopting HRNet algorithm, calculating the inclination angles of 4 groups of symmetrical points of the back, and obtaining the inclination angle scores of the symmetrical points after normalization; The second feature analysis module is used for extracting a back contour by adopting YOLOV-Seg algorithm, extracting a contour top curve, fitting an approximate straight line, calculating the inclination angle of the approximate straight line, and converting the inclination angle into an Adam forward-bending gesture score after standardization according to a first priori threshold; And the spine judgment module is used for calculating and obtaining a scoliosis score by combining analysis results of the first characteristic analysis module and the second characteristic analysis module. In a third aspect, the invention provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of the first aspect when executing the program. In a fourth aspect, the present invention provides a computer readable storage medium having stored thereon a computer program which when executed by a processor implements the method of the first aspect. The one or more technical schemes provided by the invention have at least the following technical effects or advantages: 1. The device has extremely simple requirements that the detection can be finished by only a single RGB camera, large-scale imaging devices such as CT and X-ray are not needed, auxiliary tools such as additional sensors and marking points are not needed, the device is flexible to deploy (the device can be applied to scenes such as physical examination departments of hospitals, school screening and community health service centers), and the investment threshold of sites and devices is reduced. 2. The invention has simple shooting flow, only needs to collect images of two standardized postures (standing on the back and bending forwards at 90 degrees), does not need naked backs during shooting, has low matching degree requirements on