A device and method for detecting torticollis in children

Abstract

The present invention provides an imaging and detection device and method for torticollis in children. The image processing of the three cameras can obtain 3D data of the entire head, or at least the coordinates (x, y, z) of the marker points and the head contour. Taking the middle position as the reference, the rotation angle and the tilt angle of the head can be calculated from the marker points. The marker points can be seen by at least two cameras, so the spatial position (x, y, z) of the marker points can be measured by the binocular camera. Taking the middle position as the reference, the rotation angle and the tilt angle of the head can be calculated from the marker points. By identifying the baseline of the shoulders, the angle of the head relative to its shoulders can be inferred. The present invention is based on multi-camera image acquisition, pure image processing, and three-dimensional measurement technology to realize the assessment of torticollis in children, quickly and easily realize digital detection and multi-dimensional assessment of torticollis, meet clinical requirements, and realize quantitative assessment of the following clinical symptoms.

Description

Technical Field

[0001] The present invention relates to detection of torticollis in children, and in particular to an imaging detection device and method for torticollis in children. Background Art

[0002] Torticollis refers to a disease in which the neck is tilted due to contracture of the sternocleidomastoid muscle on one side, with the head tilting toward the affected side and the face turning toward the healthy side.

[0003] The main clinical manifestations of torticollis are as follows:

[0004] Torticollis: After birth, mothers may notice that their infant's head tilts toward the affected side, the face rotates toward the unaffected side, and the lower jaw points toward the unaffected shoulder. Torticollis becomes more pronounced two to three weeks later. Turning the head toward the unaffected side is significantly limited, and even mild symptoms require careful observation to detect. These symptoms worsen as the child grows.

[0005] Neck mass: A neck mass can usually be felt after birth or within 2 weeks. It is located in the middle and lower part of the sternocleidomastoid muscle and is more common on the right side. This mass is spindle-shaped and non-tender. It usually reaches its maximum size after 1 to 2 months and then gradually shrinks until it disappears completely. In some children, the mass persists and develops muscle fibrosis and contracture, leading to torticollis.

[0006] Facial deformities: If congenital muscular torticollis is not effectively treated early, facial deformities may appear after the age of two. The main manifestation is facial asymmetry, with the distance from the outer corner of the eye to the corner of the mouth being asymmetrical on both sides, shortened on the affected side and longer on the healthy side. The eye position on the affected side is lowered, and because the eyes are not on the same horizontal line, visual fatigue and decreased vision may occur. The healthy side of the face is round and full, while the affected side is narrow and flat. Compensatory scoliosis may occur in the cervical spine. In addition, the entire face, including the nose and ears, may also show asymmetry.

[0007] Current treatment evaluation criteria: (1) Cured: The child's head and neck can rotate freely to both sides with normal range of motion, and can remain in a neutral position for a long time, and the deformity disappears. (2) Improved: The child's head and neck can rotate freely to both sides with normal range of motion, and can remain in a neutral position, but is accustomed to a mild torticollis position (<10°), or the torticollis has been significantly improved compared to before treatment (≥15°). (3) Uncured: The child's head and neck still cannot remain in a neutral position, or the short-term treatment effect is acceptable, but the torticollis deformity recurs in the long term.

[0008] Current clinical examination and treatment evaluation methods lack traceable and quantifiable means. Clinically, rulers, calipers, and protractors are still used to measure the angles of movement of children's heads and necks, and facial symmetry is assessed through clinical observation. This lack of effective instrumentation and quantifiable evaluation indicators has led to a lag in pediatric physical examinations, clinical examinations, and the evaluation of rehabilitation treatment outcomes. Summary of the Invention

[0009] 1. Technical problems to be solved:

[0010] How to quickly and easily realize digital detection and multi-dimensional evaluation of torticollis to meet clinical requirements.

[0011] 2. Technical solution:

[0012] In order to solve the above problems, the present invention provides a device for detecting torticollis in children. Figures 1 to 3 As shown, it can be installed beside the test bed where the child is lying flat. Figure 1 As shown in the figure, the surface color of the detection bed is obviously different from the color of the child's head and shoulders, which is convenient for image identification. The detection head of the device has three cameras, such as Figure 2 As shown, the orthogonal camera is located in the middle of the detection head and is used to collect the orthogonal image of the head. The left eye camera is located on the left side of the orthogonal camera and is used to collect the left image. The right eye camera is located on the right side of the orthogonal camera and is used to collect the right image.

[0013] The imaging device for detecting torticollis in children comprises an imaging probe head and a probe bracket. The probe head is equipped with three cameras: an orthostatic camera, a left-eye camera, and a right-eye camera. The camera bracket is mounted on the probe head, and the camera wiring passes through the camera bracket and the probe bracket to connect to an external computer. The camera assembly also includes rear and front covers for the orthostatic camera, the left-eye camera, and the right-eye camera.

[0014] When in use, the child's head is located in the lower center of the orthogonal camera, facing upward. In the image of the orthogonal camera, the head is in the orthogonal position, and the left and right cameras on both sides can see the head image; the camera is connected to a host as follows Figure 3 As shown, the host collects images and obtains torticollis analysis data after analysis and processing.

[0015] The detection method of the present invention comprises the following image acquisition steps: step S1: attaching two image-identifiable markers to the center of the baby's forehead and chin, wherein the markers can be distinguished from the child's skin color and the background color; step S2: acquiring at least one frontal image through the frontal camera; acquiring at least one side image through the left and right cameras, such as Figure 5As shown; Step S3: restore the child's head to the right position, then tilt the child's head to the left to its natural limit position, confirm that the head is within the field of view of the three cameras, and collect at least one image respectively, as shown Figure 6 As shown; Step S4: Then tilt the child's head to the right to its natural extreme position, confirm that the head is within the field of view of the three cameras, and collect at least one image respectively; Step S5: Rotate the child's head to the left to its natural extreme position, confirm that the head is within the field of view of the three cameras, and collect at least one image respectively, as shown Figure 7 As shown; Step S6: Rotate the child's head to the right to its natural extreme position, confirm that the head is within the field of view of the three cameras, and capture at least one image for each camera; at this point, image acquisition is completed.

[0016] The method of the present invention involves performing image analysis on a collected image group, and the analysis method includes contour recognition: using an edge extraction image processing method to identify the boundaries of the head and shoulders, as shown in Figure 9, thereby establishing the basic coordinates of the child's head and shoulders; face recognition: using face recognition technology to obtain the positional characteristics of facial organs, as shown in Figure 9; landmark point recognition: identifying the forehead and chin landmarks, as shown in Figure 9, to establish the directional position of the face.

[0017] Based on the identification of the boundaries between the head and shoulders, basic coordinate compensation is performed on each image. Taking the first positive image as a reference, the x, y position and rotation of each image are adjusted to make the shoulder baseline of each image consistent. The adjusted images are then processed for subsequent processing, such as Figure 14 shown.

[0018] The present invention processes the frontal image and identifies the facial image, such as Figure 10 As shown, the following facial evaluation parameters are calculated from the processed face images: left-right symmetry of the eyes, difference in horizontal eye position, symmetry of the left and right cheeks (measured by area difference), and difference in jaw shape. The following facial symmetry indices are analyzed and calculated from the processed images:

[0019] index unit illustrate Difference in upper and lower eye position mm C Difference in left and right eye position mm Abs(ab) Difference in facial area mm2 (Q1+Q2)-(Q3+Q4) Difference in upper and lower facial area mm2 (Q1+Q4)-(Q3+Q4) Relative Index Eye left and right position difference index ％ Abs(ab) / (a+b)*100 Facial left-right difference index ％ [(Q1+Q2)-(Q3+Q4)] / (Q1+Q2+Q3+Q4)*100 Difference in upper and lower facial area ％ [(Q1+Q4)-(Q3+Q4)] / (Q1+Q2+Q3+Q4)*100

[0020] The method of the present invention processes the left oblique image group and the right oblique image group, and calculates the left tilt degree α1 and the right tilt degree α2 of the head by using the multi-camera space reconstruction positioning method, thereby obtaining the symmetry index of the child's head tilt activity, as shown in the following table

[0021] Tilt motion symmetry index table

[0022] Tilt indicator unit illustrate Left tilt degrees Spend α1 Right tilt degrees Spend α2 Left and right tilt degree difference Spend Abs(α1-α2) Left-right tilt difference index ％ Abs(α1-α2) / (α1+α2)*100

[0023] The method of the present invention processes the left-oblique image group and the right-oblique image group, and uses a multi-camera spatial reconstruction positioning method to calculate the left rotation degree β1 and the right rotation degree β2 of the left-rotation image group and the right rotation degree β1; thereby providing a symmetry index of the child's head rotation. The table is as follows:

[0024] Rotational symmetry index table

[0025] Rotating indicator unit illustrate Degrees of left rotation Spend β1 Right rotation degrees Spend β2 Left and right rotation degree difference Spend Abs(β1-β2) Left-right rotation difference index ％ Abs(β1-β2) / (β1+β2)*100

[0026] The method of the present invention includes calibrating the camera device to obtain the internal parameters of each camera and the external parameters of the camera relative to the detection bed coordinate system, thereby establishing the relationship between the image coordinates of the camera and the actual external world coordinates (such as Figure 4 ), from which the distance from the child's head to the camera, the symmetry of its facial features, the angle of head rotation, and the tilt can be detected from the image processing of the three cameras.

[0027] The specific steps of the specific calibration method of the present invention are as follows: Step C1: Print a checkerboard and place it on the bed at the position where the child's head is placed as a calibration object; Step C2: Take a series of photos of the calibration object by adjusting the position and orientation of the calibration object; Step C3: Extract the checkerboard corner points from the photos; Step C4: Estimate the five intrinsic parameters and six extrinsic parameters of each camera under an ideal distortion-free condition, thereby establishing the coordinate relationship of each camera in the present device.

[0028] The coordinate relationship is: dx and dy are the physical size of a pixel, f is the focal length, γ is the distortion factor of the image physical coordinates, μ and υ are the vertical and horizontal offsets of the image origin relative to the optical center imaging point, R is the rotation matrix for converting the world coordinate system to the camera coordinate system, T is the translation matrix for converting the world coordinate system to the camera coordinate system, and Xw, Yw, and Zw are a three-dimensional coordinate system of the camera position.

[0029] In addition, the optical system of the camera often has distortion, which is usually calibrated using the distortion coefficients [k1, k2, k3, p1, p2], that is, the radial distortion coefficients k1, k2, k3 of the camera, and the tangential distortion coefficients p1, p2 of the camera.

[0030]

[0031] Among them, r is the distance from the image pixel to the image center, that is, r 2 =x 2 +y 2 These parameters can also be calculated from the calibration image of the checkerboard.

[0032] The three cameras can respectively form three pairs of binocular cameras, namely the left camera and the right camera, the middle camera and the left camera, and the middle camera and the right camera; based on the collected checkerboard image group, the binocular camera parameter matrix of each pair of cameras can be obtained by the binocular camera calibration method, and the three-dimensional coordinates (Xw, Yw, Zw) can be calculated through the image processing of the binocular camera.

[0033] In the present invention, in each set of captured images, the marker points are visible to at least two cameras. Using the binocular camera principle, the three-dimensional coordinates (Xai, Yai, Zai) and (Xbi, Ybi, Zbi) of marker points A and B can be calculated, where i represents the i-th image set. From the index values ​​of these marker points, α1, α2, β1, and β2 can be calculated.

[0034] 3.Beneficial effects:

[0035] The present invention proposes a new method and device for detecting torticollis. This method is based on multi-camera image acquisition, uses pure image processing, and three-dimensional measurement technology to realize the assessment of torticollis in children. It can quickly and easily realize digital detection and multi-dimensional assessment of torticollis. It is non-contact, non-destructive, non-invasive, meets clinical requirements, and realizes quantitative assessment of the following clinical symptoms. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 Schematic diagram of the detection device structure.

[0037] Figure 2 Schematic diagram of the detection head structure.

[0038] Figure 3 Schematic diagram of the detection head composition.

[0039] Figure 4 Schematic diagram of camera coordinate transformation.

[0040] Figure 5 Schematic diagram of the head in the frontal position (left, center, and right camera images).

[0041] Figure 6 Schematic diagram of the head tilted to the left (left, center, and right camera images).

[0042] Figure 7 Schematic diagram of turning the head left (left, center, and right camera images).

[0043] Figure 8 Schematic diagram of head turning left + tilting left (left, center, and right camera images).

[0044] Figure 9a a Schematic diagram of the head frontal camera; b Boundary and landmark point recognition of the head frontal image.

[0045] Figure 10Diagram of facial symmetry index

[0046] Figure 11 Diagram of left tilt activity

[0047] Figure 12 Rotation activity diagram

[0048] Figure 13a Rotate top view (no tilt) b Rotate top view (with tilt)

[0049] Figure 14 Illustration of systematic position and orientation changes DETAILED DESCRIPTION

[0050] The present invention will be described in detail below with reference to the accompanying drawings.

[0051] like Figure 1 As shown, a device for detecting torticollis in children is installed at the side of a detection bed. The detection head of the device includes three cameras with fixed positions and directions. The orthogonal camera is located directly above the detection bed and is used to collect orthogonal images. The left eye camera is located on the left side of the orthogonal camera and is used to collect left-turned images and left-oblique side images. The right eye camera is located on the right side of the orthogonal camera and is used to collect right-turned images and right-oblique right-side images. The child's head is located in the center of the first camera, facing directly upward. In the image of the orthogonal camera, the head is in the orthogonal position, and the left eye camera and the right eye camera on both sides can see the head image. The camera is connected to an external host, and the host collects the image and sends it to the processor, which obtains the torticollis data after analysis and processing.

[0052] The imaging device for detecting torticollis in children comprises an imaging probe 1 and a probe support 2. The probe is equipped with three cameras: an orthographic camera 3, a left-eye camera 4, and a right-eye camera 5. A camera support 6 is provided, and camera wiring 7 passes through the camera support 6 and the probe support 2 to connect to an external computer. The camera assembly also includes a rear cover 8 and a front cover 9 for the orthographic camera, a rear cover 10 and a front cover 11 for the left-eye camera, and a rear cover 12 and a front cover 13 for the right-eye camera.

[0053] System calibration:

[0054] After the device is installed, the cameras are calibrated to obtain the intrinsic parameters of each camera, the extrinsic parameters of the camera in the spatial coordinate system, and the parameters measured by the binocular camera. The specific steps are as follows: Step C1: Print a checkerboard and place it on the bed at the position where the child's head should be placed as a calibration object; Step C2: Adjust the position and orientation of the calibration object. The position adjustment refers to the conventional camera calibration method. Each adjustment is made, and the three cameras simultaneously capture images, thus collecting 15 or more sets of images; Step C3: Extract the checkerboard corner points from the photos; Step C4: Calculate the intrinsic and extrinsic parameters of each camera, thereby establishing the coordinate relationship of each camera in the device.

[0055] This checkerboard calibration method is known as the Zhang Zhengyou calibration method in vision technology. It is a single-plane checkerboard camera calibration method proposed by Professor Zhang Zhengyou in 1998. Because it can be completed using only a printed checkerboard, the Zhang calibration method is widely used in computer vision.

[0056] In this example, we use the OpenCV software library to calculate calibration parameters. OpenCV (OpenSource Computer Vision Library) is an open-source computer vision library developed by Intel. It implements many common algorithms for image processing and computer vision. Originally consisting of a series of C functions and a small number of C++ classes, it now supports most software platforms (such as Python and Matlab). Specifically, for camera calibration, OpenCV implements algorithms such as discovery and tracking calibration modes, calibration, fundamental matrix estimation, and homogeneous matrix estimation. By calling these interface functions, users can easily implement the required functions without performing complex mathematical calculations.

[0057] The calibration method of the orthographic camera is specifically as follows: using the chessboard image set collected in steps S01 to S04: using the orthographic camera image data set, the orthographic camera calibration is performed, specifically using the findChessboardCorners function call in the OpenCV open source algorithm to find the position of the chessboard intersection in each image, and the sub-pixel processing function cornerSubPix can be used to further process the obtained intersection position to improve the accuracy; then, the chessboard intersection position identified in the image and the actual chessboard size information are used as input, and the calibrateCamera function is called to obtain the orthographic camera calibration parameters K0 (intrinsic reference matrix), D0 (distortion matrix), R0 (rotation matrix), and T0 (translation matrix).

[0058] The calibration method for the left and right cameras is the same as the calibration method for the above-mentioned orthogonal camera, except that the image datasets of the corresponding cameras are used separately. The same calculation steps and algorithms as above can be used to calculate the calibration parameters K1 (intrinsic parameter matrix), D1 (distortion matrix), R1 (rotation matrix), and T1 (translation matrix) of the left camera; and the calibration parameters K2 (intrinsic parameter matrix), D2 (distortion matrix), R2 (rotation matrix), and T2 (translation matrix) of the right camera.

[0059] The measuring head of the device of the present invention is equipped with three cameras with fixed positions and directions. The three cameras respectively form the relationship of three pairs of binocular cameras, namely the left eye camera and the right eye camera, the middle camera and the left eye camera, and the middle camera and the right eye camera. Based on the principle of binocular cameras, after calculating the calibration parameters of each camera, we can use the calculation method of the binocular camera to obtain the coordinate relationship between the two cameras. Here, the algorithm program of OpenCV is also used to implement the calibration of the binocular camera. The specific embodiment takes the calibration relationship between the left eye camera and the right eye camera as an example. After obtaining the left eye internal parameter matrix K1 and the left eye distortion coefficient vector D1; the right eye internal parameter matrix K2 and the right eye distortion coefficient vector D2 are obtained. The parameters K1, K2, D1, and D2 obtained by measuring the left and right eyes are used as input, and then the chessboard picture with one-to-one correspondence between the left and right eyes is used at the same time. The OpenCV open source algorithm stereoCalibrate function is called to calculate and output the rotation matrix R that determines the position relationship between the left and right cameras. 12 , translation vector T 12 ; With the calibration parameters K1, K2, D1, D2, R of the two cameras of the present invention 12 、T 12 After that, binocular vision image calculation can be performed, and the 3D coordinates (X, Y, Z) of a feature point in space can be calculated from the image coordinates (x1, y1) and (x2, y2) of the two cameras.

[0060] Measurement image acquisition

[0061] After calibration is completed, the child's head and neck symmetry can be measured. The specific measurement steps include:

[0062] Step S1: Place two markers on the center of the baby's forehead and chin, where the markers can be distinguished from the child's skin color and background color; Figure 5 As shown, the marker can be a yellow, red or blue dot that can be distinguished from the child's skin color and the background color. You can also add markers on the tip of the pen or other characteristic parts, such as the corner of the eye, to assist in image recognition and processing. Place the baby on the bed with the head in the middle of the lower right position. Figure 1 As shown, the color of the bed surface is white (or black), which makes it easy to image-recognize the outline of the head and shoulders.

[0063] Step S2: Collect the orthotopic image, such as Figure 5 As shown, keep the child's head in the middle of the frontal position, facing upward. In the image of the frontal camera, the head is in the frontal position, and the cameras on both sides can see the head image; if necessary, the position of the child's head can be adjusted to obtain a normal image effect; then collect the frontal group of images, at least one for each camera.

[0064] Step S3: Acquire left oblique image, such as Figure 6 As shown, keep the child's body still, use toys or hands to assist, then tilt the child's head to the left to its natural extreme position, confirm that the head is within the field of view of the three cameras, and collect images from the three cameras, at least one from each camera. The collected multiple images form a left oblique group image, as shown Figure 3 shown.

[0065] Step S4: Capture a right oblique image. Keep the child's body still, use a toy or hand as an aid, then tilt the child's head to the right to its natural extreme position, confirm that the head is within the field of view of the three cameras, and capture images from the three cameras, at least one from each camera. The multiple images captured constitute a right oblique group image.

[0066] Step S5: Collect left-turn images, such as Figure 7 As shown, keep the child's body still, use toys or hands to help rotate the child's head to the left to its natural extreme position, confirm that the head is within the field of view of the three cameras, and capture at least one image for each camera. The multiple images captured constitute a left-turn group image.

[0067] Step S6: Capture right-turn images, rotate the child's head to the right to its natural extreme position, confirm that the head is within the field of view of the three cameras, and capture images from the three cameras, at least one image from each camera. The captured multiple images constitute a right-turn group image.

[0068] Facial symmetry analysis

[0069] Using orthographic camera images and face recognition technology, we analyze the head outline and facial features. We then use the markers on the forehead and chin to analyze facial symmetry:

[0070] Contour recognition: Uses edge extraction image processing to identify the boundary between the head and shoulders, thereby establishing the basic coordinates of the child's head and shoulders. Landmark recognition: Identifies the forehead and chin landmarks to establish the midline of the face. Face recognition: Uses face recognition technology to obtain the positional characteristics of facial organs. Because the head's color is different from the bed board, conventional image processing (edge ​​extraction) methods can easily identify the boundary between the head and shoulders. This establishes the basic coordinates of the baby's head and shoulders.

[0071] Marker recognition: Place easily recognizable markers (such as a red dot) in the center of the baby's forehead and chin. You can also add markers on the tip of a pen or other parts of the face to facilitate identification. Recognizing the forehead and chin markers establishes the facial midline, as shown in Figure 9. This serves not only as a reference point for facial analysis but also for activity analysis.

[0072] Face recognition: Face recognition technology is very mature. Our images are of good quality. Using conventional face recognition technology, we can obtain the positional features of facial organs. The following parameters can be calculated from the pixels of the image: left-right symmetry of the eyes, the difference in horizontal position of the eyes, the symmetry and area difference of the left and right cheeks, and the difference in the shape of the two jaws.

[0073] The positive group of images is processed by the face recognition algorithm, such as Figure 10 As shown in the figure, the boundary recognition, landmark recognition, feature recognition, and region segmentation of the child's face can be performed. The frontal image is processed to calculate the following facial evaluation parameters: left-right symmetry of the eyes, the difference in horizontal position of the eyes, the symmetry and area difference of the left and right cheeks, and the difference in the shape of the two jaws. Figure 11 As shown; from the processed image, the following symmetry parameters can be obtained, as shown in Table 1,

[0074] Table 1: Facial symmetry index

[0075] index unit illustrate Difference in upper and lower eye position mm c Difference in left and right eye position mm Abs(ab) Difference in facial area mm2 (Q1+Q2)-(Q3+Q4) Difference in upper and lower facial area mm2 (Q1+Q4)-(Q3+Q4) Relative Index Eye left and right position difference index ％ Abs(ab) / (a+b)*100 Facial left-right difference index ％ [(Q1+Q2)-(Q3+Q4)] / (Q1+Q2+Q3+Q4)*100 Difference in upper and lower facial area ％ [(Q1+Q4)-(Q3+Q4)] / (Q1+Q2+Q3+Q4)*100

[0076] Symmetry analysis of neck tilt movement

[0077] By processing the left oblique image group and the right oblique image, the outline of the head and shoulders, as well as the facial markers A and B, are extracted. The three-dimensional coordinates (Xai, Yai, Zai) and (Xbi, Ybi, Zbi) of the markers A and B can be calculated based on the binocular camera principle, where i represents the i-th image set. A straight line passing through two marker space points (such as Figure 11 ), the angle of inclination of the straight line parallel to the bed surface is the angle of inclination of the child's head; the angle α1 of the left inclination diagram is calculated based on this, as Figure 11 As shown, and the angle α2 of right tilt; thereby giving an index of symmetry of the child's head tilt.

[0078] Table 3: Symmetry indicators of head tilt

[0079] Tilt indicator unit illustrate Left tilt degree Spend α1 Right tilt degrees Spend α2 Left and right tilt degree difference Spend Abs(α1-α2) Left-right tilt difference index ％ Abs(α1-α2) / (α1+α2)*100

[0080] Symmetry analysis of neck rotation

[0081] By processing the left-turn image group and the right-turn image group, we can extract the outline of the head and shoulders, as well as the facial landmarks A and B. Based on the binocular camera principle, we can calculate the three-dimensional coordinates (Xai, Yai, Zai) and (Xbi, Ybi, Zbi) of the landmarks A and B, where i represents the i-th image set. In addition, from processing the frontal image set, we can derive the coordinates of the rotation center point C (Xc, Yc, Zc). Assuming that there is no parallel movement or tilt of the head during rotation, the rotation through points A and B is the same (e.g. Figure 12 ), the angle β of the point's rotation can be calculated from the spatial coordinates of a single marker point. Specifically, it is calculated as the angle between the line connecting point A and point C and the plane perpendicular to the bed surface. Applying this calculation to the left-turn image group and the right-turn image group, the degrees of left and right head rotation β1 and β2 can be calculated, thereby providing a symmetry index for the child's head rotation. In actual applications, head rotation may be accompanied by tilt, such as Figure 8 As shown, the rotation angles calculated using the markers A and B are different (as shown in FIG13 ). At this time, we take the average of the two rotation angles βa and βb, which can reflect the rotation angle of the head.

[0082] Table 2: Symmetry index of head rotation

[0083] Rotating indicator unit illustrate Degrees of left rotation Spend β1 Right rotation degrees Spend β2 Left and right rotation degree difference Spend Abs(β1-β2) Left-right rotation difference index ％ Abs(β1-β2) / (β1+β2)*100

[0084] Compensation for systemic changes

[0085] In actual use, even if the operator tries to position the child as standard as possible, the head and shoulders are in the middle of the camera and aligned with the camera coordinates when the child is in the upright position; when tilting or rotating, only rotation or oblique movement is performed. However, in reality, there will be changes in position (translation) and orientation (rotation) (such as Figure 14 In the method of the present invention, we use the boundary extraction method of image processing to obtain the boundary outline of the head and shoulders, and use this to calculate the systematic changes of each set of images during acquisition: translation (x0, y0) and rotation δ. Before performing the above-mentioned face, tilt angle, and rotation angle calculations, the images are first compensated so that the baseline of each set of images is consistent.

Claims

1. A pediatric torticollis imaging device, mounted beside a bed, for synchronously capturing images of a child's head while the child lies flat on the bed, characterized by: The detection head of the device is provided with three cameras, the orthogonal camera is located in the middle of the detection head for collecting orthogonal images, the left-eye camera is located on the left side of the orthogonal camera for collecting left-side images, and the right-eye camera is located on the right side of the orthogonal camera for collecting right-side images, including the following image collection steps: step S1: sticking two marking points on the center of the baby's forehead and chin; step S2: collecting at least one orthogonal image through the orthogonal camera; collecting at least one side image through the left and right cameras respectively; step S3: rotating the child's head to its natural extreme position to the left, confirming that the head is within the field of view of the three cameras, and collecting at least one image respectively; step S4: rotating the child's head to its natural extreme position to the right, confirming that the head is within the field of view of the three cameras, and collecting at least one image respectively; step S5: restoring the child's head to the orthogonal position, then tilting the child's head to the left to its natural extreme position, confirming that the head is within the field of view of the three cameras, and collecting at least one image respectively; step S6: then tilting the child's head to the right to its natural extreme position, confirming that the head is within the field of view of the three cameras, and collecting at least one image respectively.

2. The detection method based on the imaging detection device for children's torticollis according to claim 1, characterized in that: The image acquired by the image acquisition includes the following image processing steps: using an edge extraction image processing method to identify the boundaries of the head and shoulders, thereby establishing the basic coordinates of the child's head and shoulders; using face recognition technology to obtain the positions of facial organs; using image marker point recognition to obtain the image coordinates of the forehead and chin marker points, establish the midline of the face, and calculate the three-dimensional coordinates of the marker points using a binocular camera imaging method.

3. The detection method based on the imaging detection device for children's torticollis according to claim 2, characterized in that: Identify the boundaries of the head and shoulders, perform basic coordinate compensation on each image, and use the first orthophoto as a reference to adjust the x, y position and direction of each image so that the shoulder baseline of each image is consistent. The adjusted images are then used for subsequent processing.

4. A detection method based on the imaging detection device for torticollis in children according to claim 2 or 3, characterized in that: The facial features and marker point coordinates obtained after image processing are used to analyze the indicators of children's facial symmetry, tilt and rotation symmetry: the facial symmetry evaluation parameters are obtained by processing the frontal image: the left-right symmetry of the eyes, the difference in horizontal position of the eyes, the symmetry of the left and right cheeks, and the difference in the shape of the two jaws; by processing the left-oblique image group and the right-oblique image group, the degree of left tilt and right tilt of the head are calculated using the multi-camera space reconstruction and positioning method, thereby giving the symmetry index of the child's head tilt activity; by processing the left-rotation image group and the right-rotation image group, the degree of left rotation and right rotation of the head are calculated using the multi-camera space reconstruction and positioning method, thereby giving the symmetry index of the child's head rotation activity.

5. The detection method based on the imaging detection device for children's torticollis according to claim 4, characterized in that: The method for spatial reconstruction and positioning using multiple cameras is described. After the cameras are installed, they are calibrated to obtain the intrinsic parameters of each camera and the extrinsic parameters of the inter-camera coordinate system. The specific steps are as follows: Step C1: Print a chessboard and place it on the bed at the position where the child's head is placed as a calibration object; Step C2: Take a series of photos of the calibration object by adjusting the position and orientation of the calibration object; Step C3: Extract the chessboard corner points from the photos; Step C4: Use Zhang Zhengyou's camera calibration method to calculate the intrinsic and extrinsic parameters of each camera, thereby establishing the spatial coordinate relationship of each camera in the device.

6. A detection method based on the imaging detection device for torticollis in children according to claim 2 or 3, characterized in that: The three cameras form three pairs of binocular cameras, namely the left camera and the right camera, the middle camera and the left camera, and the middle camera and the right camera. As long as the feature points of the object to be measured can be identified in the two cameras, the spatial three-dimensional coordinates of the point can be calculated through binocular vision image processing.

7. The detection method based on the imaging detection device for children's torticollis according to claim 5, characterized in that: The head and shoulders of the child being tested are always within the field of view of the three cameras. The marking points can be distinguished from the child's skin color and background color. The color of the bed surface is white or black, which makes it easy to identify the outline of the head and shoulders in the image.

Images