A real-time skeleton model reconstruction method and system based on point cloud

Abstract

The application discloses a kind of real-time skeleton model reconstruction method and system based on point cloud, wherein, real-time skeleton model reconstruction method, comprising: S1, the position information of the key point of patient skeleton is collected, and it is roughly registered with standard skeleton model;S2, the point of patient skeleton surface is collected, every collection is set number of points, it is combined with the real-time point cloud of patient skeleton obtained last time to obtain current patient skeleton real-time point cloud, on the basis of S1 rough registration, current patient skeleton real-time point cloud is registered with standard skeleton model;S3, after the point of the surface of the bone of the skeleton of the patient collected satisfies set condition, the reconstruction of the skeleton model of patient is completed according to corresponding registration result.The application can be registered with standard skeleton model in the process of point cloud collection, the three-dimensional reconstruction of the skeleton model of patient can be completed while the surface information of bone is collected, planning and navigation are directly carried out after reconstruction is completed, further reduce the surgical procedure, improve the efficiency of operation.

Description

Technical Field

[0001] This invention relates to the field of medical robot technology, and in particular to a method and system for real-time skeletal model reconstruction based on point clouds. Background Technology

[0002] Currently, when performing joint surgery, especially knee surgery, it is generally necessary to obtain the patient's joint imaging data first, such as CT images or MRI images. Because CT images have high resolution and 3D modeling capabilities, they facilitate preoperative surgical planning by doctors. Therefore, CT images are typically used to obtain imaging information of the patient's bones and other structures. Based on this imaging information, the doctor then formulates a surgical plan.

[0003] During preoperative surgical planning, CT images need to be segmented to obtain a mask of the bones (including the femur, tibia, fibula, etc.). 3D reconstruction is then performed based on the mask results, and preoperative planning is carried out on the reconstruction results, including steps such as osteotomy location, prosthesis type and prosthesis placement.

[0004] Currently, domestic manufacturers use CT scans for preoperative planning and intraoperative navigation. However, traditional knee surgery does not require CT images and has problems such as long preoperative CT segmentation time, high patient costs, high radiation levels, and difficulty in reimbursement.

[0005] Currently, some manufacturers offer solutions based on optical tracking systems that directly collect bone surface information during surgery and perform real-time 3D bone reconstruction using point cloud registration technology. However, this requires applying a complete coat of material to the bone surface before registration. This adds extra waiting time for the surgeon and prevents real-time application and registration. Furthermore, the registration effect cannot be observed in real time during application; any deviations in application are only detected after registration, increasing the cost of error correction. Summary of the Invention

[0006] Purpose of the invention: To address the above-mentioned shortcomings, this invention proposes a real-time skeletal model reconstruction method and system based on point clouds, which greatly reduces the surgical procedure time, improves surgical efficiency, and reduces patient costs compared to CT methods.

[0007] Technical solution:

[0008] A real-time skeleton model reconstruction method based on point clouds includes:

[0009] S1. Collect the location information of key points of the patient's bones and perform coarse registration with the standard bone model;

[0010] S2. Collect points on the surface of the patient's bones. Collect a set number of points and combine them with the real-time point cloud of the patient's bones obtained previously to obtain the current real-time point cloud of the patient's bones. Based on the coarse registration in S1, register the current real-time point cloud of the patient's bones with the standard bone model so that the standard bone model fits the current real-time point cloud of the patient's bones.

[0011] S3. After the points on the patient's bone surface that have been collected meet the set conditions, the patient's bone model is reconstructed based on the corresponding registration results.

[0012] Specifically, the registration includes:

[0013] S21. Using the current real-time point cloud of the patient's skeleton as the source point cloud and the standard skeleton model as the target point cloud, perform fine registration based on the coarse registration in S1.

[0014] S22. Based on the fine registration in S21, perform elastic registration between the current patient's real-time bone point cloud and the standard bone model, so that the standard bone model fits the current patient's real-time bone point cloud.

[0015] More specifically, in S2, the fine registration of the current patient's real-time bone point cloud with the standard bone model specifically involves:

[0016] Based on the previous fine registration, the current real-time point cloud of the patient's skeleton is finely registered with the standard skeleton model. The first registration is based on the coarse registration of S1.

[0017] More specifically, before performing elastic registration between the current patient's real-time bone point cloud and the standard bone model, the method further includes the steps of: meshing the current patient's real-time bone point cloud and the standard bone model, and performing elastic registration between the meshed current patient's real-time bone point cloud and the standard bone model.

[0018] More specifically, the elastic registration is as follows: using the current real-time point cloud of the patient's skeleton as the target point cloud and the standard skeleton model as the source point cloud, elastic registration is performed using the NICP algorithm based on the fine registration.

[0019] Furthermore, the flexible registration specifically refers to:

[0020] (1) Based on the fine registration, the source point cloud and the target point cloud are searched using the nearest neighbor search algorithm;

[0021] (2) Calculate the distance between the corresponding two points in each pair of nearest neighbors, and determine whether each distance is less than the set threshold;

[0022] If it is less than, then the nearest neighbor pair is considered to have been successfully registered, and the positions of the two points remain unchanged.

[0023] Otherwise, the transformation matrix between the two corresponding points is decomposed by Koleski to obtain the corresponding rotation matrix in the pair of nearest neighbors, and the corresponding points in the source point cloud are rotated and the source point cloud is updated, and the iteration count is incremented by 1.

[0024] (3) Determine whether the number of iterations exceeds the set number; if so, end the iteration, obtain the final updated source point cloud, and complete the elastic registration; otherwise, search for the nearest neighbor between the updated source point cloud and the target point cloud, and return to step (2).

[0025] Furthermore, in step (2), after calculating the distance d between the corresponding two points in each pair of nearest neighbors, a distance coefficient k is added to it, and the registration of the pair of nearest neighbors is determined by judging whether kd is less than a set threshold.

[0026] Furthermore, the value of the distance coefficient k gradually decreases with the number of iterations.

[0027] Furthermore, the current real-time point cloud of the patient's skeleton and the mesh of the skeleton model are constructed using the Alpha reconstruction method, the Poisson reconstruction method, or the rolling ball reconstruction method.

[0028] A real-time skeletal model reconstruction system based on point clouds, comprising:

[0029] The point cloud acquisition unit is equipped with a probe tip and at least three non-collinear tracer spheres, which are used to collect point clouds in real time by contacting the bone surface with its probe tip.

[0030] An optical tracking system is used to acquire the position information of the tracer sphere on the point cloud acquisition unit in real time;

[0031] The processing unit acquires the position information of the tracer ball on the point cloud acquisition unit obtained by the optical tracking system, and obtains the position information of the corresponding point on the bone surface based on the position information of its probe tip.

[0032] The processing unit collects a set number of points from the point cloud acquisition unit, combines them with the previously obtained real-time point cloud of the patient's skeleton to obtain the current real-time point cloud of the patient's skeleton, and registers it with a standard skeleton model so that the standard skeleton model fits the current real-time point cloud of the patient's skeleton. After the points on the surface of the patient's skeleton collected by the point cloud acquisition unit meet the set conditions, the reconstruction of the patient's skeleton model is completed according to the registration result.

[0033] Specifically, the point cloud acquisition unit collects key bone points by contacting the patient's bone with its probe tip, and the processing unit obtains the position information of the corresponding key points through the above processing, and performs coarse registration with the standard bone model accordingly.

[0034] The processing unit, based on the previous registration, uses the ICP algorithm to register the current patient's real-time bone point cloud with the standard bone model, wherein the first registration is based on the coarse registration.

[0035] More specifically, after each registration, the processing unit performs elastic registration between the current patient's real-time bone point cloud and the standard bone model using the NICP algorithm, so that the standard bone model fits the current patient's real-time bone point cloud.

[0036] Compared with the prior art, the present invention has the following advantages:

[0037] 1. This invention utilizes an optical tracking system to directly acquire bone surface information during surgery. It can register with a standard bone model in real time during the acquisition process. Through real-time registration, the three-dimensional reconstruction of the patient's bone model can be completed simultaneously with the acquisition of bone surface information. Compared with CT methods, this invention greatly reduces the time of the surgical procedure, improves surgical efficiency, and reduces patient costs when using a surgical robot.

[0038] 2. Each registration in this invention can be based on the previous registration, resulting in more accurate registration and thus more precise three-dimensional reconstruction of the patient's bone model.

[0039] 3. This invention provides point cloud completion for skeletal areas where point cloud data is difficult to collect, ensuring the integrity and authenticity of the patient's skeletal point cloud.

[0040] 4. This invention addresses situations where there are mismatches or large errors in some surface points between the patient's skeletal point cloud and the standard skeletal model, or where the standard skeletal model and the patient's skeletal point cloud do not fit well enough, with some points having large deviations. It registers the corresponding points in the patient's skeletal point cloud and the standard skeletal model to ensure the fit between the standard skeletal model and the acquired patient's skeletal point cloud, ensuring the accuracy of the registration and further improving the accuracy of the patient's skeletal model reconstruction. Attached Figure Description

[0041] Figure 1 This is a flowchart of the point cloud-based real-time skeleton model reconstruction method of the present invention;

[0042] Figure 2 This is an example diagram of point cloud completion in an embodiment of the present invention, wherein, Figure 2 (a) is an example image before point cloud completion. Figure 2 (b) is an example image after point cloud completion. Detailed Implementation

[0043] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0044] This invention relates to a real-time skeletal model reconstruction method based on point clouds, such as... Figure 1 As shown, the steps include:

[0045] S1. Collect the location information of key points of the patient's bones and perform coarse registration with the standard bone model;

[0046] Specifically:

[0047] S11. By clicking on the key points of the patient's bones with a probe, the position information of the tracer ball is identified through the optical tracking system, and then the position information of the key points of the patient's bones is obtained.

[0048] In this invention, point cloud information of key points is collected by a point cloud acquisition unit such as a probe. At least three coplanar but non-collinear tracer spheres are set on the probe. The key points of the patient's bones are points with anatomical significance on the patient's bones. For example, three key points are collected on the femur: the intercondylar fossa, the lateral condyle, and the medial condyle; and three key points are collected on the tibia: the medial proximal tibial tangent point, the lateral proximal tibial tangent point, and the tibial tuberosity.

[0049] In this invention, soft tissue may be present at certain key points on the patient's bones. In this case, it is necessary to use a probe to puncture the soft tissue and collect information at the corresponding key points.

[0050] S12. Perform coarse registration between the patient's skeleton and the standard skeleton model;

[0051] Based on the location information of the patient's skeletal key points obtained in S11, and combined with the corresponding points on the standard skeletal model, coarse registration is performed between the patient's skeleton and the standard skeletal model, ensuring that the collected patient's skeletal key points coincide with the corresponding points on the standard skeletal model. In this invention, because only key points are used for registration, the accuracy is not high, and complete overlap is generally impossible. However, this step is only to roughly align the patient's skeleton with the standard skeletal model, providing an initial registration basis for subsequent fine registration. Therefore, the coarse registration in this invention only needs to ensure that the collected patient's skeletal key points coincide as much as possible with the corresponding points on the standard skeletal model, such as meeting set conditions. For example, the sum or average distance between each skeletal key point and its corresponding point on the standard skeletal model is less than a set value. The standard skeletal model uses the average skeletal model of a set number of people within a set age range.

[0052] Specifically, the present invention can use the RANSAC (Random Sample Consensus) algorithm to rigidly register the key points of the patient's skeleton with the corresponding points on the standard skeleton model.

[0053] S2. Collect points on the surface of the patient's bones. Collect a set number of points and combine them with the real-time point cloud of the patient's bones obtained in the previous step to obtain the real-time point cloud of the current patient's bones. Register the real-time point cloud of the current patient's bones with the standard bone model.

[0054] Specifically:

[0055] S21. The point cloud acquisition unit, such as a probe, collects points on the surface of the patient's bone in real time, excluding key points. After collecting a set number of points, the points are combined with the real-time point cloud of the patient's bone obtained in the previous step to obtain the current real-time point cloud of the patient's bone.

[0056] During the data acquisition process, the probe needs to be placed close to the bone surface and kept moving on the bone surface at all times to ensure that the shape of the acquired point cloud is consistent with the actual bone.

[0057] In this invention, the collected point cloud needs to be displayed at a corresponding location on the bone so that the doctor knows the location where the point cloud has been collected. In joint replacement surgery or related operations, such as knee replacement surgery, the density of the collected points needs to be increased in the areas related to the corresponding operation, such as the medial and lateral femoral condyles, the tibial plateau, and other key areas related to the osteotomy plane planning operation, so as to plan the osteotomy plane more accurately.

[0058] Therefore, in this invention, the point cloud collection of these key areas must meet the following conditions:

[0059] Indexes corresponding to points in different key regions are pre-set on the standard skeletal model. During the point cloud acquisition process, such as when a probe is used to acquire the point cloud of the patient's bone surface excluding key points, a nearest neighbor search algorithm, such as the KDTree algorithm, is used to search for nearest neighbors based on the coarse registration in S1. This searches for the points in the standard skeletal model that are closest to the points in the point cloud of the patient's bone surface. The number of points located in key regions is recorded. If the number of points located in key regions reaches a set condition, the acquisition of points in the corresponding region on the patient's bone surface is considered complete. In this invention, the set condition is that the ratio between the number of points located in key regions and the number of points in the corresponding key regions on the standard skeletal model is greater than a set value, such as 0.9.

[0060] Some key areas are located at the boundary between soft tissue and bone or within soft tissue, making them impossible to collect samples from, such as... Figure 2 As shown in (a), point cloud completion is required. The specific point cloud completion is as follows:

[0061] The point cloud boundary of the patient's bone surface is calculated. Based on the point cloud boundary, the distance between each key point collected in S1 and each boundary point is calculated. An interval parameter between each key point and each boundary point is then set. The point cloud model is then completed using this interval parameter, and finally stitched together with the actual point cloud to form the final point cloud data, i.e., the real-time point cloud of the patient's bone. This invention can use algorithms such as latitude and longitude scanning, mesh generation, normal estimation, and alpha shapes to calculate the point cloud boundary of the patient's bone surface. Preferably, this invention uses a boundary extraction method based on normal estimation of point cloud data.

[0062] Taking the medial and lateral condyles as an example, after acquiring the medial and lateral condyle points, point cloud completion is needed in the area from the key points of the medial and lateral condyles to the middle of the point cloud on the bone surface. Based on the calculated point cloud boundary of the patient's bone surface, the distance from the medial and lateral condyle points to the point cloud boundary points is calculated. An interval parameter is then set accordingly. New points are generated in the area from the medial and lateral condyle points to the point cloud boundary points based on the interval parameter, and then stitched together with the actual point cloud to form the final point cloud of the patient's bone. Figure 2 As shown in (b).

[0063] Similarly, during point cloud completion, if the ratio between the number of points to be completed in the corresponding area to be completed and the number of points in the corresponding area on the standard skeleton model is greater than a set value, then the points in that area are considered to have been completed.

[0064] In this invention, the quantity is set to 100.

[0065] S22. Based on the previous registration, perform fine registration between the real-time point cloud of the current patient's skeleton obtained in S21 and the standard skeleton model;

[0066] Specifically, using the real-time point cloud of the current patient skeleton obtained in S21 as the source point cloud and the standard skeleton model as the target point cloud, fine registration is performed using the ICP algorithm based on the previous registration. The first registration is based on the coarse registration in S1, and the previous registration is the fine registration between the real-time point cloud of the patient skeleton obtained in the previous registration and the standard skeleton model.

[0067] In this invention, because the real-time point cloud of the current patient's skeleton is initially small and differs greatly from the standard skeleton model, if the registration is performed using the standard skeleton model as the source point cloud and the real-time point cloud of the current patient's skeleton as the target point cloud, as in the conventional ICP algorithm, registration may fail. Therefore, this invention uses the real-time point cloud of the current patient's skeleton as the source point cloud and the standard skeleton model as the target point cloud, thereby ensuring the accuracy of registration.

[0068] In this step, the real-time point cloud of the patient's skeleton is registered with the standard skeleton model. This process involves the registration of the patient's skeleton point cloud and the standard skeleton model as a whole. Without changing the shape of the patient's skeleton point cloud, the registration is made to align and fit them as closely as possible. Compared with the aforementioned coarse registration, the registration accuracy is higher, and the real-time point cloud of the patient's skeleton and the standard skeleton model fit each other better than the aforementioned coarse registration.

[0069] S23. Based on the fine registration in S22, perform flexible registration between the two.

[0070] In this invention, after the fine registration in S22, there may be some mismatches or large errors, or the standard bone model and the real-time point cloud of the patient's bones may not fit well, with some points having large deviations. Therefore, this invention performs elastic registration between the real-time point cloud of the patient's bones and the standard bone model based on the fine registration in S2, so that the standard bone model fits the real-time point cloud of the patient's bones, and then the standard bone model is transformed into the shape of the actual patient's bones by adjustment.

[0071] Specifically, this invention employs the NICP algorithm to elastically register the real-time point cloud of the patient's skeleton with a standard skeleton model. In this invention, the real-time point cloud of the patient's skeleton is used as the target point cloud, and the standard skeleton model is used as the source point cloud.

[0072] (1) Based on the S22 fine registration, the nearest neighbor search algorithm is used to search for the nearest neighbor points of the real-time point cloud of the patient's bones and the standard bone model;

[0073] (2) Calculate the distance between the corresponding two points in each pair of nearest neighbors, and determine whether each distance is less than the set threshold;

[0074] If it is less than, then the nearest neighbor pair is considered to have been successfully registered, and the positions of the two points remain unchanged.

[0075] Otherwise, the transformation matrix between corresponding points is obtained by Koleski decomposition to obtain the corresponding rotation matrix in the pair of nearest neighbors, and the corresponding points in the source point cloud are rotated and the source point cloud is updated, and the iteration count is incremented by 1; wherein, in the first iteration, the transformation matrix between corresponding points is the transformation matrix obtained by S22 fine registration;

[0076] (3) Determine whether the number of iterations exceeds the set number P; if so, end the iteration, obtain the final updated source point cloud, and complete the elastic registration; otherwise, search for the nearest neighbor between the updated source point cloud and the target point cloud, and return to step (2).

[0077] In this invention, in step (2), to prevent over-registration between the real-time point cloud of the patient's skeleton and the standard skeleton model, after calculating the distance d between corresponding points in each pair of nearest neighbors, a distance coefficient k is added to it. Whether the pair of nearest neighbors is successfully registered is determined by whether kd is less than a set threshold. Furthermore, the value of the distance coefficient k gradually decreases with the number of iterations, such as by decreasing proportionally or by a fixed step size.

[0078] S3. After the points on the patient's bone surface that have been collected meet the set conditions, the reconstruction of the patient's bone model is completed based on the corresponding elastic registration results.

[0079] After the points collected from the patient's bone surface meet the set conditions, that is, after obtaining the final point cloud of the patient's bone, the reconstruction of the patient's bone model is completed according to the corresponding elastic registration obtained above; wherein, the points collected from the patient's bone surface meet the set conditions by being greater than a set number or by the ratio between the number of points collected and the number of surface points of the standard bone model being greater than a set value.

[0080] Furthermore, before elastic registration, the real-time point cloud of the patient's skeleton can be meshed with the standard skeleton model, or the two can be directly elastically registered.

[0081] In this invention, the real-time point cloud of the patient's bones and the standard bone model can be meshed using methods such as Alpha reconstruction, Poisson reconstruction, and rolling ball reconstruction. The points in the real-time point cloud of the patient's bones and the standard bone model are iterated until the entire point cloud is converted into a mesh model. Preferably, this invention uses rolling ball reconstruction to mesh the real-time point cloud of the patient's bones and the standard bone model.

[0082] This invention also provides a real-time standard skeleton model reconstruction system based on point clouds, comprising:

[0083] The point cloud acquisition unit, specifically a probe, is equipped with a probe tip and at least three non-collinear tracer spheres; the point cloud acquisition unit collects point clouds by contacting the bone surface with its probe tip.

[0084] An optical tracking system acquires the position information of the tracer sphere on the point cloud acquisition unit in real time;

[0085] The processing unit acquires the position information of the tracer ball on the point cloud acquisition unit obtained by the optical tracking system. Thus, when the point cloud acquisition unit is acquiring point cloud data, the position information of the corresponding detection tip can be obtained through the position information of the tracer ball, and then the position information of the corresponding point on the bone surface can be obtained.

[0086] The processing unit collects a set number of points from the point cloud acquisition unit, combines them with the obtained point cloud to obtain a real-time point cloud of the patient's skeleton, and performs fine registration with a standard skeleton model. After the points on the surface of the patient's skeleton collected by the point cloud acquisition unit meet the set conditions, the reconstruction of the patient's standard skeleton model is completed according to the registration result.

[0087] Specifically, the processing unit acquires the position information of the tracer ball on the point cloud acquisition unit obtained by the optical tracking system. Thus, when the point cloud acquisition unit performs point cloud acquisition, the position information of the corresponding detection tip can be obtained through the position information of the tracer ball, thereby obtaining the position information of the corresponding point on the bone surface, and thus obtaining the real-time point cloud of the patient's bones.

[0088] The point cloud acquisition unit first collects key points of the patient's bones by contacting the probe tip with the key points of the bones. The processing unit obtains the position information of the corresponding key points through the above processing and performs coarse registration with the standard bone model accordingly. Then, the point cloud acquisition unit collects points on the bone surface. After collecting a set number of points, it combines them with the previously collected point cloud to obtain the real-time point cloud of the patient's bones. The standard bone model adopts the average standard bone model of a set number of people within a set age range.

[0089] The processing unit performs fine registration of the real-time point cloud of the patient's skeleton with the standard skeleton model using the ICP algorithm, based on the previous registration. The first registration is based on the coarse registration. On the basis of the fine registration, the processing unit performs elastic registration of the real-time point cloud of the patient's skeleton with the standard skeleton model using the NICP algorithm, so that the real-time point cloud of the patient's skeleton fits the standard skeleton model. Finally, after the points on the surface of the patient's skeleton meet the set conditions, the reconstruction of the patient's skeleton model is completed based on the elastic registration results.

[0090] In this invention, both the ICP algorithm and the NICP algorithm are consistent with those in the aforementioned standard skeletal model reconstruction method.

[0091] This invention eliminates the need for CT image acquisition when reconstructing a standard patient bone model. It allows for direct intraoperative acquisition of bone surface information via an optical tracking system, with real-time registration with the standard bone model during acquisition. This real-time registration enables simultaneous 3D reconstruction of the patient's bone model and bone surface information acquisition. After reconstruction, planning and navigation can be directly performed, further reducing the surgical procedure and improving efficiency. Furthermore, this invention reduces the patient's radiation exposure time, significantly lowering medical costs. Each registration in this invention is based on the previous one, resulting in more accurate registration and thus more precise 3D reconstruction of the patient's bone model. Meanwhile, during the acquisition of real-time point clouds of the patient's skeleton, this invention performs point cloud completion for bone areas where point clouds are difficult to acquire, ensuring the integrity and authenticity of the patient's skeleton point cloud. After registration, for situations where there are mismatches or large errors between the real-time point cloud of the patient's skeleton and the standard skeleton model, or where the standard skeleton model and the real-time point cloud of the patient's skeleton do not fit well enough, with some points having large deviations, elastic registration is performed. That is, corresponding points in the standard skeleton model and the real-time point cloud of the patient's skeleton are registered to ensure the fit between the standard skeleton model and the acquired real-time point cloud of the patient's skeleton, ensuring the accuracy of registration and further improving the accuracy of the patient's skeleton model reconstruction.

[0092] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details of the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations (such as quantity, shape, position, etc.) can be made to the technical solution of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A method for real-time skeletal model reconstruction based on point clouds, characterized in that, include: S1. Collect the location information of key points of the patient's bones and perform coarse registration with the standard bone model; S2. Collect points on the surface of the patient's bones. Collect a set number of points and combine them with the real-time point cloud of the patient's bones obtained in the previous step to obtain the current real-time point cloud of the patient's bones. Based on the coarse registration in S1, use the current real-time point cloud of the patient's bones as the source point cloud and the standard bone model as the target point cloud for fine registration. Based on precise registration, elastic registration is performed between the current real-time point cloud of the patient's skeleton and the standard skeleton model, so that the standard skeleton model fits the current real-time point cloud of the patient's skeleton. Specifically: (1) Based on the fine registration, the source point cloud and the target point cloud are searched using the nearest neighbor search algorithm; (2) Calculate the distance between the corresponding two points in each pair of nearest neighbors, and determine whether each distance is less than the set threshold; If it is less than, then the nearest neighbor pair is considered to have been successfully registered, and the positions of the two points remain unchanged. Otherwise, the transformation matrix between the two corresponding points is decomposed by Koleski to obtain the corresponding rotation matrix in the nearest neighbor pair, and the corresponding points in the source point cloud are rotated and the source point cloud is updated, with the iteration count incremented by 1. (3) Determine if the number of iterations exceeds the set number; If yes, the iteration ends, the final updated source point cloud is obtained, and the elastic registration is completed; otherwise, the nearest neighbor between the updated source point cloud and the target point cloud is searched, and the process returns to step (2). S3. After the points on the patient's bone surface that have been collected meet the set conditions, the patient's bone model is reconstructed based on the corresponding registration results.

2. The real-time skeletal model reconstruction method according to claim 1, characterized in that, The precise registration specifically refers to: Based on the previous fine registration, the current real-time point cloud of the patient's skeleton is finely registered with the standard skeleton model. The first registration is based on the coarse registration of S1.

3. The real-time skeletal model reconstruction method according to claim 1, characterized in that, Before performing elastic registration between the current patient's real-time bone point cloud and the standard bone model, the method further includes the following steps: meshing the current patient's real-time bone point cloud and the standard bone model, and then performing elastic registration between the meshed current patient's real-time bone point cloud and the standard bone model.

4. The real-time skeletal model reconstruction method according to claim 1, characterized in that, In step (2), after calculating the distance d between the corresponding two points in each pair of nearest neighbors, a distance coefficient k is added to it, and the registration of the pair of nearest neighbors is determined by judging whether kd is less than a set threshold.

5. The real-time skeletal model reconstruction method according to claim 4, characterized in that, The value of the distance coefficient k gradually decreases with the number of iterations.

6. The real-time skeletal model reconstruction method according to claim 3, characterized in that, The current real-time point cloud of the patient's skeleton and the mesh of the skeleton model are constructed using the Alpha reconstruction method, Poisson reconstruction method, or rolling ball reconstruction method.

7. A real-time skeletal model reconstruction system based on point clouds, characterized in that, include: The point cloud acquisition unit is equipped with a probe tip and at least three non-collinear tracer spheres, which are used to collect point clouds in real time by contacting the bone surface with its probe tip. An optical tracking system is used to acquire the position information of the tracer sphere on the point cloud acquisition unit in real time; The processing unit acquires the position information of the tracer ball on the point cloud acquisition unit obtained by the optical tracking system, and obtains the position information of the corresponding point on the bone surface based on the position information of its probe tip. The processing unit collects a set number of points from the point cloud acquisition unit, combines them with the previously obtained real-time point cloud of the patient's skeleton to obtain the current real-time point cloud of the patient's skeleton, and registers it with a standard skeleton model so that the standard skeleton model fits the current real-time point cloud of the patient's skeleton. After the points on the surface of the patient's skeleton collected by the point cloud acquisition unit meet the set conditions, the reconstruction of the patient's skeleton model is completed according to the corresponding registration result, thereby implementing the real-time skeleton model reconstruction method as described in any one of claims 1 to 6.

8. The real-time skeletal model reconstruction system according to claim 7, characterized in that, The point cloud acquisition unit collects key points of the bones by contacting the key points of the patient's bones with its probe tip. The processing unit obtains the position information of the corresponding key points through the above processing and performs coarse registration with the standard bone model accordingly. The processing unit, based on the previous registration, uses the ICP algorithm to register the current patient's real-time bone point cloud with the standard bone model, wherein the first registration is based on the coarse registration.

9. The real-time skeletal model reconstruction system according to claim 8, characterized in that, After each registration, the processing unit performs elastic registration between the current patient's real-time bone point cloud and the standard bone model using the NICP algorithm, so that the standard bone model fits the current patient's real-time bone point cloud.

Images