A method for orthopedic surgery navigation registration based on UWB marker points

Abstract

A kind of orthopedic surgery navigation registration method based on UWB marker point, UWB module is used as calibration point around the patient's affected part to calibrate, DICOM data is obtained by scanning medical imaging equipment, convolutional neural network is combined with integrated learning algorithm Xgboost to pre-process medical imaging data, finally, three-dimensional reconstruction is carried out by surface rendering Marching Cubes method, the spatial coordinate system of calibration point is obtained by UWB positioning algorithm, according to the number of registration points in actual spatial coordinate system and virtual spatial coordinate system, the optimal transformation matrix is found, so that the transformed coordinates are closest to virtual coordinate system coordinates.Translation transformation matrix and rotation transformation matrix are solved by SVD matrix decomposition method, scale transformation parameter is solved by quaternion algorithm combined with homogeneous coordinate theory, when marker point moves, repeat the steps to complete real-time registration and registration based on UWB marker point.Avoid the movement of intraoperative non-active factors and need to recalibrate.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of surgical navigation positioning, in particular to a bone surgery navigation registration method based on UWB marker points. BACKGROUND

[0002] The spatial registration technology is the core of the positioning technology of the surgical navigation system, and its purpose is to obtain the spatial conversion relationship between the reference entity part of the patient's skeleton and its three-dimensional virtual model, and to provide a prerequisite guarantee for the accurate and efficient positioning performance of the orthopedic surgery navigation system. At present, the spatial registration method applied in the clinic is to complete point cloud registration by using corresponding reference points on the skeleton entity and its three-dimensional virtual model, which can be divided into the following two types:

[0003] The paired marker registration method based on medical markers. When this method is adopted, medical markers are usually implanted into the patient's skeleton to make these markers clearly displayed in the image space and collected in the patient space, so as to ensure high point cloud sample collection accuracy.

[0004] The non-paired marker registration method based on anatomical feature structure characteristics. Point cloud data capable of describing the anatomical structure of the patient's skeleton is respectively acquired in the patient space and the image space, and then the two sets of point cloud data are registered through a spatial registration algorithm.

[0005] Although the paired marker registration method based on medical markers has high accuracy, the implantation of markers increases the pain and psychological burden of the patient, and in the operation, the markers may be slightly moved, which requires re-registration after moving, increases the difficulty of the operation, and even in some cases, the markers cannot be implanted smoothly.

[0006] The non-paired marker registration method based on anatomical feature structure characteristics needs to find two sets of point cloud data in space and then register them through an algorithm. This method is theoretically strong, but has a large error in implementation. The ICP algorithm is the most commonly used algorithm to solve the problem of three-dimensional point cloud registration at present. Its main idea is to find the best affine transformation to minimize the average distance between the two nearest corresponding points. The current ICP algorithm has many defects, such as a large difference between the initial positions and attitudes of the two sets of point clouds, and no good initial transformation value is provided. The number of iteration operations will increase, the registration result is easy to fall into a local optimal solution, and registration failure will occur. In addition, with the increase of the number of point cloud samples, the calculation amount in the process of searching for the nearest point will increase, and then the registration efficiency and real-time performance will be poor. SUMMARY

[0007] The purpose of the present application is to provide a bone surgery navigation registration method based on UWB marker points to solve the problems in the background art.

[0008] To achieve the above object, the application provides the following technical scheme: a bone surgery navigation registration method based on UWB marker points, comprising S1, three-dimensional reconstruction, and S2, marker point registration.

[0009] The three-dimensional reconstruction is to scan the marked affected part by using a medical image scanning device, and then to perform image segmentation on the medical image data obtained after scanning, wherein the convolutional neural network is used as a feature trainer for image segmentation, the convolutional neural network architecture comprises a convolutional layer, a pooling layer, a fully connected layer and an output layer, the last layer of the trained convolutional neural network model, i.e., the output layer, is replaced by a feature vector input of an integrated learning xgboost classifier, an image probability map is output by the xgboost classifier, automatic threshold segmentation and isolated region removal are performed, and finally a segmented image is output.

[0010] The marker point registration is to obtain the actual space coordinates of the marker points by UWB positioning, in the conversion of the actual space coordinates of the marker points to virtual space coordinates, according to the number of registration points in the actual space coordinate system and the virtual space coordinate system, an optimal transformation matrix is found to make the transformed coordinates closest to the coordinates in the virtual coordinate system, then the marker point transformation matrix is decomposed into a translation transformation matrix, a rotation transformation matrix and a scale transformation parameter which are independent of each other, finally the translation transformation matrix and the rotation transformation matrix are solved by using the quaternion algorithm combined with the homogeneous coordinate theory, the translation matrix can be represented by a translation vector, and the rotation matrix can be represented by Euler angles and quaternions, when the marker point moves, the steps are repeated to complete real-time space registration and registration based on the UWB marker points.

[0011] Preferably, the specific steps of the step S1, three-dimensional reconstruction, are as follows:

[0012] S11, a feature extractor is trained by using a convolutional neural network, the convolutional neural network comprises an input layer, a hidden layer and an output layer;

[0013] The hidden layer comprises two pairs of feature mapping layers and convolutional sampling layers;

[0014] The first feature mapping layer comprises six surfaces, and the down-sampling rate is 3;

[0015] The second feature mapping layer comprises 12 surfaces, and the down-sampling rate is 3;

[0016] The convolutional filter kernel size of the feature layer is 5*5;

[0017] To balance time and speed, the training model in this paper is trained for 100 rounds with a window size of 61. Based on this convolutional neural network framework, the features of the last layer of the trained convolutional neural network model are used as the input of the xgboost classifier, that is, the output layer of the last layer of the original network structure is replaced with the input of the feature vector of the xgboost classifier.

[0018] S12, using XGBoost as the pattern recognition algorithm. In the previous step, the feature extractor based on the convolutional neural network has been trained. Then, the test set is input into the feature extractor to learn the feature set. The last output layer of the entire framework is replaced with XGBoost. The learned 108-dimensional feature vector is used to train the entire XGBoost classifier. After the random forest classifier is trained, the test set can be segmented to obtain the probability distribution of the image.

[0019] S13, Image post-processing;

[0020] S14 employs a surface rendering algorithm. First, it determines the volume elements containing isosurfaces. Then, it uses linear interpolation to find the intersection points between the isosurfaces and the volume elements. Finally, it calculates the normal vector of the isosurfaces to complete the 3D reconstruction of the medical image.

[0021] Preferably, step S13, the specific operation of image post-processing, is as follows:

[0022] S131, Automatic thresholding was performed using the automatic thresholding method provided by Fiji based on the cell membrane probability map returned by the xgboost algorithm;

[0023] S132, iteratively remove irrelevant isolated regions, where the region removal criterion is a series of threshold operations based on region attributes;

[0024] The region properties used include region area, Euler's totient function, eccentricity of an ellipse with the same standard second-order central moment as the region, and pixel ratio in both the region and its smallest convex polygon.

[0025] Preferably, the specific steps of step S2, marker registration, are as follows:

[0026] S21, Obtaining the coordinates of the marker point:

[0027] TDOA ranging technology obtains the time difference of arrival of signals from two different base stations to the positioning tag through a UWB module, and multiplies it by the electromagnetic wave propagation speed c to obtain the distance difference.

[0028] r i，j =r i -r j =c(t) i -tj

[0029] wherein r i，j is the distance difference between the marker point and the ith base station and the marker point and the jth base station, r i is the ranging value of the marker point to the ith base station, i≠j; i, j = 1, 2, …, n; n is the total number of base stations;

[0030] t is the propagation time of the signal between the base station and the tag;

[0031] Taking the first base station as the reference base station, the TDOA hyperbolic model can be expressed as:

[0032]

[0033] wherein r i，1 is the distance difference between the marker point and the ith base station and the positioning tag and the reference base station, (x, y) represents the real position coordinates of the marker point, and (x i , y i ) represents the plane coordinates of the ith base station;

[0034] S22, coordinate conversion:

[0035] Let the coordinates of the marker point in the real-time surgery space obtained by the TDOA ranging technology be {P n}, the coordinates of the marker point in the computer space obtained after the image three-dimensional visualization be {I n}, n = 1, 2, …, N, N is the total number of marker points, W is the translation transformation matrix, V is the rotation transformation matrix, and s is the scaling factor. When the transformation matrix S of {P n} and {I n} satisfies the condition , the current transformation matrix S P-1 is the optimal transformation matrix of {P n} and {I n}. The unit quaternion and homogeneous coordinate theory can be used to obtain S P-1 . The specific algorithm process is as follows:

[0036] The center coordinates of the marker point group in the real-time space and the computer space are obtained, wherein P0 is the center coordinates of the marker point group in the real-time surgery space, and I0 is the center coordinates of the marker point group in the computer space;

[0037]

[0038] Then, the coordinates of the marker point group in the two spaces are converted into relative coordinate values about the center coordinates, wherein P ′ n is the relative coordinate value of the nth marker point in the real-time surgery space, and I​′ n the relative coordinate value of the nth marker point in the computer space;

[0039]

[0040] According to matrix theory and quaternion theory, the optimal rotation transformation matrix V can be expressed as q (q0, q1, q2, q3) ;

[0041]

[0042] Solve the scaling factor s, so that the optimal transformation matrix From the above formula, we have Since the optimal rotation transformation matrix V is a rigid body transformation, we have ‖V (P ′ n )‖ = ‖I ′ n ‖, which is brought into the optimal transformation matrix S, so that the scaling factor s that satisfies the condition can be solved:

[0043]

[0044] Solve the optimal translation transformation matrix:

[0045] W = P n -sV (I n )

[0046] There is the following formula to solve the optimal transformation matrix:

[0047]

[0048] Using the optimal transformation matrix, the spatial mapping relationship from the real-time surgery space to the computer space in the surgical navigation system can be solved, and the registration process in the surgery process can be completed.

[0049] Preferably, in step S21, the original TDOA value is substituted into the Chan algorithm in the marker point coordinate acquisition, the preliminary positioning coordinates of the UWB positioning tag are calculated, and then the residual sum of squares is calculated, and the calculation formula is:

[0050] Then the error is removed, and the role of the residual is to measure the closeness between a group of TDOA values and the corresponding positioning results. In the case that the main source of error in the UWB positioning system is NLOS error, the greater the influence of NLOS error on a group of measured values, the greater the residual value calculated;

[0051] Therefore, the threshold value is set as R ws ≤ Δ, and the residual value is compared, and the TDOA measured value containing large NLOS error is removed;

[0052] When the residual value is greater than the threshold value, it indicates that the distance difference of the estimated coordinates calculated by the Chan algorithm to the base station has a large deviation from the TDOA measurement value between the tag and the base station measured by the sensor, and when the NLOS error is the main error of the positioning system, it indicates that the TDOA measurement value contains a large NLOS error, and a corresponding group of TDOA measurement values is removed

[0053] When the residual value is less than or equal to the threshold value, it is considered that the error of the measurement value is small.

[0054] Finally, the position coordinate estimation of the positioning tag is screened out, and the TDOA measurement value with a smaller error is brought into the standard EKF filtering algorithm, so that the real and accurate position coordinates of the marker point of the surgical instrument can be obtained.

[0055] Compared with the prior art, the beneficial effects of the present application are that the present application proposes a new marker point registration method, uses a UWB model as a marker, and because of the chip-level wireless positioning technology of UWB, when the UWB module is used as a marker, slight movement of the intraoperative marker or movement of the patient can be corrected in coordinates through the real-time positioning technology of UWB, and re-calibration and registration are not needed, and the current UWB positioning algorithm is mature, has high accuracy, and has strong visualization means. The spatial coordinate system of the marker can be obtained through the positioning algorithm that has been set, and finally the virtual coordinate system of the marker point is obtained through coordinate conversion. While the accuracy is ensured, re-registration is also eliminated. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 The present application is a whole flow chart. DETAILED DESCRIPTION

[0057] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0058] The present application proposes a new marker point registration method, uses a UWB model as a marker, and because of the chip-level wireless positioning technology of UWB, when the UWB module is used as a marker, slight movement of the intraoperative marker or movement of the patient can be corrected in coordinates through the real-time positioning technology of UWB, and re-calibration and registration are not needed, and the current UWB positioning algorithm is mature, has high accuracy, and has strong visualization means. The spatial coordinate system of the marker can be obtained through the positioning algorithm that has been set, and finally the virtual coordinate system of the marker point is obtained through coordinate conversion. While the accuracy is ensured, re-registration is also eliminated.

[0059] The present application mainly includes two aspects, specifically, preoperative three-dimensional reconstruction and marker point registration. The three-dimensional reconstruction is the basis of marker point registration, and accurate three-dimensional reconstruction can better display the virtual coordinate system of the calibration point. The overall framework diagram is shown in the accompanying Figure 1 , specifically:

[0060] 1) Three-dimensional reconstruction

[0061] The medical image scanning device is used to scan the calibrated affected part, and then the medical image data obtained after scanning is subjected to image segmentation. The convolutional neural network is used as a feature trainer for image segmentation, and the convolutional neural network architecture is composed of convolutional layers, pooling layers, fully connected layers and output layers. The last layer of the trained convolutional neural network model, the output layer, is replaced by the feature vector input of the integrated learning xgboost classifier, and the image probability map is output by the xgboost classifier. Automatic threshold segmentation and isolated region removal are performed, and finally the segmented image (i.e. image key target region extraction and partial tissue quantitative representation) is output. The Marching Cubes method is used for three-dimensional reconstruction. The method is described as follows:

[0062] 1. First, a feature extractor is trained using a convolutional neural network. The convolutional neural network of the present application is composed of an input layer, a hidden layer and an output layer. The hidden layer includes two pairs of feature mapping layers and convolutional sampling layers. The first feature mapping layer includes 6 faces, and the down-sampling rate is 3. The second feature mapping layer includes 12 faces, and the down-sampling rate is 3. The convolutional filter kernel size in the feature layer is 5*5. Under the premise of balancing time and rate, the training number of the trained model in this paper is 100 rounds, and the window size is set to 61. On the basis of the convolutional neural network framework, the features of the last layer of the trained convolutional neural network model are used as the input of the xgboost classifier, i.e. the output layer of the last layer of the original network structure is replaced by the input of the feature vector of the xgboost classifier.

[0063] 2. Second is based on ensemble learning xgboost image segmentation, in the existing algorithm, the accuracy of xgboost algorithm is incomparable; for large data volume can maintain efficient performance. And xgboost can effectively estimate data missing, for those without a lot of input variables can be directly processed and keep the accuracy unchanged. Xgboost in the process of classification, for each variable can give an evaluation to determine the importance of the variable. And, xgboost in the process of building can generate an unbiased internal estimate for the generalization error. For those unbalanced data sets, the error can be effectively balanced. And the trained xgboost can be saved, to solve other data preparation. And clustering analysis can be achieved by calculating the similarity of instances, so that the abnormal points can be determined.

[0064] Therefore, the design adopts xgboost as the pattern recognition algorithm, through the above step, the feature extractor based on convolutional neural network has been trained, then the test set is input into the feature extractor for feature set learning, the output layer of the last layer of the whole framework is replaced by xgboost, the 108-dimensional feature vector learned is used to train the whole xgboost classifier, and after the random forest classifier is trained, the test set can be segmented to obtain the probability distribution of the image.

[0065] 3. Finally, image post-processing, the image post-processing of the application still has two steps: one is to use the automatic threshold method provided by Fiji to perform automatic threshold segmentation according to the cell membrane probability map returned by the xgboost algorithm, which can effectively improve the continuity of the cell membrane; two is to iteratively execute the removal of irrelevant isolated regions, wherein the region removal criterion is a series of threshold operations based on region attributes. The region attributes used include region area, Euler function, eccentricity of an ellipse with the same standard second central moment as the region, and pixel ratio in the region and its minimum convex polygon (Solidity).

[0066] 4. Three-dimensional reconstruction, using surface rendering algorithm, first determine the volume element containing the isosurface, use linear interpolation method to find the intersection point of the isosurface and the volume element, and finally calculate the normal vector of the isosurface to complete the three-dimensional reconstruction of the medical image.

[0067] 2) Calibration point registration

[0068] The actual spatial coordinates of the marker points are obtained by UWB positioning (TDOA), and in the conversion of the actual spatial coordinates of the marker points to virtual spatial coordinates, according to the number of registration points in the actual spatial coordinate system and the virtual spatial coordinate system, an optimal transformation matrix is found to make the transformed coordinates closest to the coordinates in the virtual coordinate system. Then the marker point transformation matrix is decomposed into a translation transformation matrix, a rotation transformation matrix and a scale transformation parameter which are independent of each other. Finally, the translation transformation matrix and the rotation transformation matrix are solved by the quaternion algorithm combined with the homogeneous coordinate theory. For the translation matrix, a translation vector can be used to represent it, and for the rotation matrix, Euler angles and quaternions can be used to represent it. When the marker points move, the steps can be repeated to complete the real-time spatial registration and registration based on the UWB marker points. The specific implementation is as follows:

[0069] 1. Marker point coordinate acquisition:

[0070] The TDOA ranging technology is to obtain the time difference of arrival of signals at two different base stations to the positioning tag, and multiply the electromagnetic wave propagation speed c to obtain the distance difference:

[0071] r i，j = r i - r j = c (t i - t j )

[0072] In the formula, r i，j is the distance difference between the marker point to the i-th base station and the marker point to the j-th base station, r i is the ranging value of the marker point to the i-th base station, i≠j; i, j = 1, 2,..., n; n is the total number of base stations. t is the signal propagation time between the base station and the tag. Take the first base station as the reference base station, and the TDOA hyperbolic model can be expressed as:

[0073]

[0074] In the formula, r i，1 is the distance difference between the marker point to the i-th base station and the positioning tag to the reference base station, (x, y) represents the real position coordinates of the marker point, and (x i , y i ) represents the plane coordinates of the i-th base station.

[0075] First, the original TDOA value is substituted into the Chan algorithm to calculate the preliminary positioning coordinates of the UWB positioning tag, and then the residual sum of squares is calculated, and the calculation formula is:

[0076]

[0077] Then the error elimination is carried out, and the role of the residual error is to measure the closeness between a group of TDOA values and the corresponding positioning results. In the case that the main source of error in the UWB positioning system is NLOS error, the greater the influence of NLOS error on a group of measurement values, the greater the residual value calculated. Therefore, the threshold value R ws ≤ Δ is set, and the TDOA measurement value containing large NLOS error is eliminated by comparing the residual value. When the residual value is greater than the threshold value, it indicates that there is a large deviation between the distance difference of the estimated coordinates calculated by the Chan algorithm to the base station and the TDOA measurement value between the tag and the base station obtained by the sensor measurement. In the case that NLOS error is the main error of the positioning system, it indicates that the TDOA measurement value contains large NLOS error, and the corresponding group of TDOA measurement values is eliminated. If the residual value is less than or equal to the threshold value, it is considered that the error of the measurement value is small.

[0078] Finally, the position coordinate estimation of the positioning tag is carried out, and the TDOA measurement value with small error screened out is brought into the standard EKF filtering algorithm again, so that the final position coordinate of the surgical instrument can be obtained.

[0079] 2. Coordinate conversion:

[0080] Let the coordinates of the marker points obtained by the TDOA ranging technology in the real-time surgery space be {P n}, the coordinates of the marker points obtained by the image three-dimensional visualization in the computer space be {I n}, n = 1, 2,..., N, N is the total number of marker points, W is the translation transformation matrix, V is the rotation transformation matrix, and s is the scaling factor. When the transformation matrix S n of {P n} and {I n} satisfies the condition , the current transformation matrix S P-1 is the optimal transformation matrix of {P n} and {I n}, and S P-1 can be obtained by using unit quaternion and homogeneous coordinate theory. The specific algorithm process is as follows:

[0081] The center coordinates of the marker point group in the real-time space and the computer space are obtained, wherein P0 is the center coordinates of the marker point group in the real-time surgery space, and I0 is the center coordinates of the marker point group in the computer space.

[0082]

[0083] Then the coordinates of the marker point group in the two spaces are converted into relative coordinate values about the center coordinates, wherein P′ n is the relative coordinate value of the n th marker point in the real-time surgery space, and I′ nThe relative coordinate value of the nth marker point in the computer space.

[0084]

[0085] According to the matrix theory and the quaternion theory, the optimal rotation transformation matrix V can be expressed as q (q0, q1, q2, q3).

[0086]

[0087] Solve the scaling factor s, so that the optimal transformation matrix V From the above formula, we have Since the optimal rotation transformation matrix V is a rigid body transformation, we have ||V(P n )|| = ||I n ||, which is brought into the optimal transformation matrix S, so that the scaling factor s that satisfies the condition can be solved:

[0088]

[0089] Solve the optimal translation transformation matrix:

[0090] W = P n -sV(I n )

[0091] There is the following formula to solve the optimal transformation matrix:

[0092]

[0093] Using the optimal transformation matrix, the spatial mapping relationship from the real-time surgery space to the computer space in the surgical navigation system can be solved, and the registration process in the surgery process can be completed.

[0094] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method for registration of orthopedic surgery navigation based on UWB marker points, characterized in that, Includes S1, 3D reconstruction and S2, marker registration; 3D reconstruction utilizes medical imaging scanning equipment to scan the calibrated affected area, and then performs image segmentation on the obtained medical image data. Image segmentation uses a convolutional neural network as a feature trainer. The convolutional neural network architecture consists of convolutional layers, pooling layers, fully connected layers, and an output layer. The last layer of the trained convolutional neural network model—the output layer—is replaced with the feature vector input of an ensemble-learned XGBoost classifier. The XGBoost classifier outputs an image probability map, which is used for automatic thresholding and isolated region removal. Finally, the segmented image is output. Marker registration involves obtaining the actual spatial coordinates of the markers via UWB positioning. During the transformation from the actual spatial coordinates to the virtual spatial coordinates, an optimal transformation matrix is ​​found based on the number of registered points in both the actual and virtual coordinate systems. This matrix ensures that the transformed coordinates are closest to the virtual coordinates. The marker transformation matrix is ​​then decomposed into independent translation, rotation, and scaling parameters. Finally, the translation and rotation transformation matrices are solved using a quaternion algorithm combined with homogeneous coordinate theory. The translation matrix can be represented by a translation vector, while the rotation matrix is ​​represented by Euler angles and quaternions. Repeating these steps as the markers move completes real-time spatial registration and UWB-based marker registration.

2. The UWB marker point-based orthopedic surgery navigation registration method of claim 1, wherein: The specific steps of the three-dimensional reconstruction in step S1 are as follows: S11 uses a convolutional neural network to train a feature extractor, which consists of an input layer, hidden layers, and an output layer. The hidden layer includes two pairs of feature mapping layers and a convolutional sampling layer; The first feature mapping layer contains 6 faces with a downsampling rate of 3; The second feature mapping layer contains 12 faces and has a downsampling rate of 3; The size of the convolutional filter kernel in the feature layer is 5*5; To balance time and speed, the training model was trained for 100 rounds with a window size of 61. Based on this convolutional neural network framework, the features of the last layer of the trained convolutional neural network model were used as the input of the xgboost classifier. In other words, the output layer of the last layer of the original network structure was replaced with the input of the feature vector of the xgboost classifier. S12, using XGBoost as the pattern recognition algorithm. In the previous step, the feature extractor based on the convolutional neural network has been trained. Then, the test set is input into the feature extractor to learn the feature set. The last output layer of the entire framework is replaced with XGBoost. The learned 108-dimensional feature vector is used to train the entire XGBoost classifier. After the random forest classifier is trained, the test set can be segmented to obtain the probability distribution of the image. S13, Image post-processing; S14 employs a surface rendering algorithm. First, it determines the volume elements containing isosurfaces. Then, it uses linear interpolation to find the intersection points between the isosurfaces and the volume elements. Finally, it calculates the normal vector of the isosurfaces to complete the 3D reconstruction of the medical image.

3. The UWB marker point-based orthopedic surgery navigation registration method of claim 2, wherein: Step S13, the specific operations of image post-processing are as follows: S131, according to the cell membrane probability map returned by the xgboost algorithm, an automatic threshold segmentation is performed using the automatic threshold method provided by Fiji; S132, the removal of irrelevant isolated regions is iteratively performed, wherein the removal criterion of the region is based on a series of threshold operations on the region attributes; The region attributes used include the region area, the Euler function, the eccentricity of the ellipse with the same standard second central moment as the region, and the proportion of pixels in both the region and its minimum convex polygon.

4. The orthopedic surgery navigation registration method based on UWB marker points according to claim 1, characterized in that: The specific steps of the step S2, marker point registration, are as follows: S21, marker point coordinate acquisition: The TDOA ranging technology is to obtain the time difference of arrival of signals at two different base stations to the positioning tag through the UWB module, multiply the electromagnetic wave propagation speed c to obtain the distance difference: r i,j = r i - r j = c(t i - t j ) wherein r i,j is the difference between the distance from the marker point to the i-th base station and the distance from the marker point to the j-th base station, r i is the ranging value from the marker point to the i-th base station, i≠j; i, j = 1, 2, …, n; n is the total number of base stations; t is the propagation time of the signal between the base station and the tag; Taking the first base station as the reference base station, the TDOA hyperbolic model can be expressed as: In the formula, r i，1 is the distance difference from the i-th base station to the positioning tag to the reference base station, (x, y) represents the real position coordinates of the marker point, and (x i , y i ) represents the plane coordinates of the i-th base station. S22, coordinate conversion: Let the coordinates of the marker points obtained by TDOA ranging technology in the real-time operation space be {P n}, and the coordinates of the marker points obtained by three-dimensional visualization of the image in the computer space be {I n}, n = 1, 2,..., N, N is the total number of marker points, W is the translation transformation matrix, V is the rotation transformation matrix, and s is the scaling factor. When the transformation matrix S of {P n} and {I n} satisfies the condition , the current transformation matrix S P-1 is the optimal transformation matrix of {P n} and {I n}. The S P-1 can be obtained by using unit quaternion and homogeneous coordinate theory; the specific algorithm process is as follows: The center coordinates of the marker point group in the real-time space and the computer space are obtained, wherein P0 is the center coordinates of the marker point group in the real-time operation space, and I0 is the center coordinates of the marker point group in the computer space; The coordinate transformation of the marker point groupings in the two spaces into relative coordinate values about the center coordinate is then performed, where P' n is the relative coordinate value of the nth marker point in the real-time surgical space, I' n is the relative coordinate value of the nth marker point in the computer space; According to the matrix theory and the quaternion theory, the optimal rotation transformation matrix V can be represented by q (q0, q1, q2, q3); Solve for the scaling factor s to obtain the optimal transformation matrix. From the above formula, we can obtain Since the optimal rotation transformation matrix V is a rigid body transformation, then we have ||V(P′)|| n )||=||I′ n Substituting this into the optimal transformation matrix S, we can obtain the scaling factor s that satisfies the conditions: The optimal translation transformation matrix is solved: W = P n -sV(I n ) The optimal transformation matrix can be solved by the following formula: Using the optimal transformation matrix, the spatial mapping relationship from the real-time operation space to the computer space in the operation navigation system can be solved, and the registration process in the operation process is completed.

5. The UWB marker point-based orthopedic surgery navigation registration method of claim 4, wherein: In step S21, in the marker point coordinate acquisition, the original TDOA value is substituted into the Chan algorithm to calculate the preliminary positioning coordinates of the UWB positioning tag, and then the residual sum of squares is calculated, and the calculation formula is: Then, error elimination is performed, and the residual error is used to measure the closeness between a group of TDOA values and the corresponding positioning results. In the case that the main source of error in the UWB positioning system is the NLOS error, the greater the influence of the NLOS error on a group of measured values, the greater the residual value calculated by the NLOS error; Therefore, the threshold value is set as R ws ≤ Δ, and the TDOA measurement value containing a larger NLOS error is eliminated by comparing with the residual value. When the residual value is greater than the threshold value, it indicates that there is a large deviation between the distance difference of the estimated coordinates to the base station calculated by the Chan algorithm and the TDOA measurement value between the tag and the base station obtained by the sensor measurement. When the NLOS error is the main error of the positioning system, it indicates that the TDOA measurement value contains a large NLOS error, and thus a group of corresponding TDOA measurement values are eliminated; When the residual value is less than or equal to the threshold value, it is considered that the error of the measurement value is small. Finally, the positioning tag position coordinate estimation is performed, the TDOA measurement value with a small error selected is brought into the standard EKF filtering algorithm again, and the final accurate position coordinates of the marker point of the operation instrument can be obtained.

Images