Registration method and readable storage medium

Abstract

The present invention discloses a registration method and a readable storage medium. The registration method includes the following steps: generating an image model point cloud based on a medical imaging image of a subject; registering the image model point cloud to the space where the subject's point cloud is located to obtain a registered image model point cloud; adjusting the registered image model point cloud based on a first displacement relationship between a first data frame of a target object on the subject and its second data frame to obtain a registered real-time point cloud; and registering the image model point cloud to the space where the subject's point cloud is located in real time based on the real-time displacement relationship between the current data frame and the previous data frame of the target object and the registered real-time point cloud. The registration method and readable storage medium provided by the present invention use a face point cloud and a scalp point cloud generated by an MRI image for registration. In the subsequent registration stage, only the displacement of the subject needs to be tracked to achieve real-time registration.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing, and in particular to a registration method and a readable storage medium. Background Art

[0002] In recent years, based on the analysis results of medical images, when treating the human brain in physical space, it is necessary to align the real model of the human brain with the virtual model obtained through medical image reconstruction, so as to achieve real-time navigation during surgery.

[0003] The currently commonly used navigation system is an infrared positioning system based on reflective balls. It requires the subject to wear a locator (for example, a headband with four positioning balls), and then use a positioning stick with reflective balls to perform initial alignment of facial key points.

[0004] It can be seen that the existing registration methods have the following shortcomings: First, since the subject is required to remain still during the initial registration stage, and the initial registration lasts for a long time, it is relatively difficult to implement in actual operation; second, if the subject's head moves during the initial registration process, it will cause registration failure; third, since the locator needs to be used to click on the subject's face to achieve registration, there is a certain risk and the user experience is not good. Summary of the Invention

[0005] In order to solve at least one aspect of the above-mentioned problems and defects in the prior art, the present invention provides a registration method. The technical method is as follows:

[0006] According to one aspect of the present invention, a registration method is provided, the registration method comprising the following steps:

[0007] generating an image model point cloud based on a medical imaging image of a subject;

[0008] Registering the image model point cloud to the space where the subject point cloud is located to obtain a registered image model point cloud;

[0009] adjusting the registered image model point cloud based on a first displacement relationship between the first data frame and the second data frame of the target object on the subject to obtain a registered real-time point cloud;

[0010] Based on the real-time displacement relationship between the current data frame and the previous data frame of the target object and the real-time point cloud after registration, the image model point cloud is registered in real time to the space where the subject's point cloud is located.

[0011] Furthermore, the first displacement relationship and the real-time displacement relationship are both obtained through an infrared positioning system, and the subject point cloud is obtained through a stereo camera.

[0012] Specifically, based on the first displacement relationship between the first data frame and the second data frame of the target object on the subject, the registered image model point cloud is adjusted to obtain the registered real-time point cloud, the acquisition time of the subject point cloud corresponds to the timestamp in the first data frame, and the registered real-time point cloud is the real-time point cloud corresponding to the timestamp in the second data frame.

[0013] Specifically, based on the real-time displacement relationship between the current data frame and the previous data frame of the target object and the real-time point cloud after registration, the image model point cloud is registered in real time to the space where the subject point cloud is located, and the output real-time point cloud after registration is the real-time point cloud corresponding to the timestamp in the current data frame.

[0014] Preferably, the first displacement relationship and the real-time displacement relationship are both displacement matrices corresponding to the real-time position change of the subject.

[0015] Furthermore, in registering the image model point cloud to the space where the subject point cloud is located to obtain a registered image model point cloud, the method of registering the image model point cloud to the space where the subject point cloud is located includes the following steps:

[0016] Preprocessing the image model point cloud and the subject point cloud to obtain a preprocessed subject point cloud and a preprocessed image model point cloud;

[0017] Obtaining a spatial transformation relationship between the preprocessed image model point cloud and the preprocessed subject point cloud by a first registration method;

[0018] Based on the spatial transformation relationship, the preprocessed image model point cloud is registered to the space where the preprocessed subject point cloud is located through a second registration method to obtain the registered image model point cloud.

[0019] Specifically, the point cloud preprocessing includes the following steps:

[0020] Denoising and downsampling the image model point cloud and the subject point cloud to obtain a downsampled image model point cloud and a downsampled subject point cloud;

[0021] The local surface features of the point clouds in the downsampled image model point cloud and the downsampled subject point cloud are extracted to obtain the model point cloud features and the subject point cloud features.

[0022] Specifically, the first registration method is to match the model point cloud features with the subject point cloud features through the RANSAC algorithm and obtain the translation transformation matrix and rotation transformation matrix for aligning the preprocessed image model point cloud to the space where the preprocessed subject point cloud is located.

[0023] Specifically, the second registration method includes the following steps:

[0024] Optimizing the translation transformation matrix and the rotation transformation matrix by an ICP algorithm to obtain the optimized translation transformation matrix and the optimized rotation transformation matrix;

[0025] Based on the optimized translation transformation matrix and the optimized rotation transformation matrix, the preprocessed image model point cloud is aligned to the space where the preprocessed subject point cloud is located to obtain the aligned image model point cloud.

[0026] According to another aspect of the present invention, a readable storage medium is provided, wherein:

[0027] The readable storage medium stores a program or instruction, and when the program or instruction is executed by the processor, the above-mentioned registration method is performed.

[0028] The registration method and readable storage medium according to the embodiments of the present invention have at least one of the following advantages:

[0029] (1) The registration method and readable storage medium provided by the present invention can achieve the initial positioning of the subject's face by using a stereo camera to capture a face point cloud in the initial registration stage, and then registering it with the scalp point cloud generated by the MRI image. In this way, in the subsequent registration stage, only the subject's displacement needs to be tracked to achieve real-time registration;

[0030] (2) The registration method provided by the present invention realizes the initial registration and real-time registration of the entity point cloud and the model point cloud by using the optical point cloud, thereby ensuring that the target position of the subject obtained in the entity point cloud and the target position planned in the model point cloud can be accurately and in real time coincident;

[0031] (3) The registration method and readable storage medium provided by the present invention achieve the initial registration of the entity point cloud and the model point cloud by using an optical point cloud. Compared with the traditional method of using a positioning rod to register several key points of facial physiological structures with the same position of the model point cloud, it solves the problem that the initial registration stage takes too long and the head movement during the registration process easily leads to registration failure;

[0032] (4) The registration method and readable storage medium provided by the present invention achieve the initial registration of the entity point cloud and the model point cloud by using an optical point cloud. After that, only the infrared positioning system needs to be used to track the subject's displacement to achieve real-time registration, thereby avoiding the danger of using a positioning rod to the subject during the registration process and improving the subject's experience.

[0033] (5) The registration method and readable storage medium provided by the present invention perform model point cloud preprocessing in advance during the optical registration process, thereby reducing the time consumption of the overall registration;

[0034] (6) The registration method and readable storage medium provided by the present invention achieve real-time tracking of the image model point cloud in the real world by using two different cameras to take advantage of their respective strengths. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] These and / or other aspects and advantages of the present invention will become apparent and readily understood from the following description of the preferred embodiments taken in conjunction with the accompanying drawings, in which:

[0036] Figure 1 is a flow chart of a registration method according to one embodiment of the present invention;

[0037] Figure 2 yes Figure 1 Schematic diagram of the registration method shown;

[0038] Figure 3 yes Figure 1 A comparison chart of the registration accuracy of the first registration method and the second registration method is shown;

[0039] Figure 4 This is a schematic diagram of the model point cloud before registration;

[0040] Figure 5 This is a schematic diagram of the subject’s point cloud before registration. DETAILED DESCRIPTION

[0041] The technical solution of the present invention will be further described in detail below through examples and in conjunction with the accompanying drawings. In the specification, the same or similar reference numerals indicate the same or similar components. The following description of the embodiments of the present invention with reference to the accompanying drawings is intended to explain the overall inventive concept of the present invention and should not be construed as limiting the present invention.

[0042] Currently, a common registration method uses a combination of a locator and a positioning rod to achieve full-process real-time registration. The specific process is as follows: First, the subject wears a locator with four reflective balls on their head to position the subject. Then, the positioning rod is used to spatially locate the subject's brain to establish the relative position between the subject's brain and the locator. During the spatial positioning process, several points on the subject's face are clicked and positioned. During this process, the subject is required to remain stable and not move.

[0043] Therefore, the present invention provides a registration method that can at least avoid the dangers brought to the subject by using a positioning rod during the registration process, while solving the problems of the initial registration stage taking too long and the head movement that easily causes registration failure during the registration process.

[0044] See also Figure 1, which shows a registration method according to an embodiment of the present invention. The registration method comprises the following steps:

[0045] Step S1 generates an image model point cloud based on the medical imaging image of the subject;

[0046] Step S2 registers the image model point cloud to the space where the subject point cloud is located to obtain a registered image model point cloud;

[0047] Step S3: adjusting the registered image model point cloud based on a first displacement relationship between the first data frame and the second data frame of the target object on the subject to obtain a registered real-time point cloud;

[0048] Step S4 iterates step S3, and based on the real-time displacement relationship between the current data frame and the previous data frame of the target object and the real-time point cloud after registration, the image model point cloud is registered in real time to the space where the subject point cloud is located.

[0049] The medical imaging image of the subject may be an MRI image, preferably a T1 modality MRI image, or may be a CT imaging image, an ultrasound imaging image, or the like.

[0050] The following example is described using a T1 modality MRI image of the subject's brain as an example of a medical imaging image of the subject. The principles of images formed by other imaging methods are the same as those of this embodiment and will not be described in detail here.

[0051] First, an infrared positioning system (e.g., an infrared camera) and a stereo camera (e.g., an optical camera) are used to simultaneously perform real-time shooting, with the infrared camera tracking the subject to obtain the subject's real-time world coordinates, and the optical camera obtaining the subject's point cloud (e.g., the subject's facial point cloud); second, the subject's MRI image is converted into an image model point cloud.

[0052] A stereo camera is a device that uses stereo imaging technology to perform stereo imaging, including structured light stereo cameras, TOF 3D cameras, binocular stereo vision cameras, etc. Those skilled in the art can choose one according to actual needs, as long as they can obtain the subject's point cloud.

[0053] Specifically, the steps of converting the MRI image of the subject into the image model point cloud are as follows: normalizing the MRI image of the subject into a standardized image, then performing boundary extraction on the standardized image (for example, performing boundary extraction through morphological processing and ray algorithm) to obtain an initial boundary image, and then performing noise reduction processing on the initial boundary image (for example, performing noise reduction processing through statistical filtering, radius filtering, etc.) to obtain a boundary point cloud image, that is, the image model point cloud.

[0054] In this example, the terms "first" and "second" are used for descriptive purposes only and should not be construed as limiting the order of the method. Those skilled in the art can adjust the order of the shooting and image conversion steps according to actual needs, or even perform both simultaneously.

[0055] In step S2, the method for registering the image model point cloud to the space where the subject point cloud is located includes the following steps:

[0056] Preprocessing the image model point cloud and the subject point cloud to obtain a preprocessed subject point cloud and a preprocessed image model point cloud;

[0057] Obtaining a spatial transformation relationship between the preprocessed image model point cloud and the preprocessed subject point cloud by a first registration method;

[0058] Based on the spatial transformation relationship, the preprocessed image model point cloud is registered to the space where the preprocessed subject point cloud is located through a second registration method to obtain the registered image model point cloud.

[0059] In one example, the point cloud preprocessing includes the following steps:

[0060] Downsampling the image model point cloud and the subject point cloud to obtain a downsampled image model point cloud and a downsampled subject point cloud;

[0061] The local surface features of the point clouds in the downsampled image model point cloud and the downsampled subject point cloud are extracted to obtain the model point cloud features and the subject point cloud features.

[0062] For example, the model point cloud was downsampled to 10mm resolution using a voxel grid using Open3D, where the side length of the voxel grid was set to 10mm. Similarly, the subject point cloud was downsampled to 10mm resolution using a voxel grid using Open3D.

[0063] Afterwards, the normal vectors of each point in the downsampled image model point cloud are calculated, and the positional relationship between the normal vectors of all points in the k-neighborhood is calculated using the FPFH algorithm based on the normal vectors of each point in the image model point cloud. This results in a simplified point feature histogram, i.e., the FPFH feature of the image model point cloud. Similarly, the normal vectors of each point in the sampled subject point cloud are calculated, and the FPFH feature of the subject point cloud is calculated based on the normal vectors of each point in the subject point cloud using the FPFH algorithm.

[0064] Preferably, the method for calculating the normal vector is to use a KDtree search algorithm that mixes KNN and radius search to obtain the nearest neighbor, and then estimate the plane based on least squares fitting to obtain the normal vector of the point.

[0065] Of course, those skilled in the art will understand that deep learning algorithms, such as PointNet, PerfectMatch, and FCGF, can also be used to extract local surface features, such as curvature and normal vectors, from the downsampled image model point cloud and the downsampled subject point cloud.

[0066] Those skilled in the art can also understand that point cloud preprocessing also includes performing image segmentation on the downsampled subject point cloud to remove outliers generated by the stereo camera, and obtaining the segmented subject point cloud, and then calculating the normal vectors of each point of the downsampled image model point cloud and the segmented subject point cloud, and extracting their respective FPFH features based on the FPFH algorithm.

[0067] Of course, those skilled in the art may also use, for example, a k-nearest neighbor point cloud filtering method to filter the downsampled subject point cloud to remove outliers generated by the stereo camera. This example is merely illustrative, and those skilled in the art may also use other existing filtering and noise reduction methods to reduce or eliminate noise in the point cloud.

[0068] In one example, the first registration method is to match the FPFH features of the model point cloud with the FPFH features of the subject point cloud through a random sampling consensus (RANSAC) algorithm and obtain a translation transformation matrix and a rotation transformation matrix for aligning the preprocessed image model point cloud to the space where the preprocessed subject point cloud is located.

[0069] For example, the RANSAC algorithm is used to determine the matching relationship between the FPFH features of the model point cloud and the FPFH features of the subject point cloud. This involves randomly extracting N samples from the dataset of the model point cloud FPFH features and the subject point cloud FPFH features. The transformation matrices for aligning the model point cloud FPFH features to the subject point cloud FPFH features, namely the translation and rotation matrices, are then calculated. Subsequently, based on the projection error between the model point cloud FPFH features and the space containing the subject point cloud FPFH features, mismatched points are removed, completing the coarse registration of the model point cloud and the subject point cloud.

[0070] Of course, those skilled in the art can also understand that the first registration method can also be the maximum likelihood random sampling algorithm (MLESAC) or the progressive consistent sampling algorithm (PROSAC) to match the two point clouds in the feature space to obtain the relationship between the corresponding points in the two point clouds, and then calculate the translation transformation matrix and rotation transformation matrix of the preprocessed image model point cloud and the preprocessed subject point cloud based on the relationship between the corresponding points.

[0071] In one example, the second registration method includes the following steps:

[0072] Optimizing the translation transformation matrix and the rotation transformation matrix by an ICP algorithm to obtain the optimized translation transformation matrix and the optimized rotation transformation matrix;

[0073] Based on the optimized translation transformation matrix and the optimized rotation transformation matrix, the preprocessed image model point cloud is aligned to the space where the preprocessed subject point cloud is located to obtain the aligned image model point cloud.

[0074] For example, the image model point cloud is registered to the subject point cloud based on the registration matrix M1 (i.e., the translation transformation matrix and the rotation transformation matrix) obtained by coarse registration to obtain the coarsely registered image model point cloud. The image model point cloud after coarse registration is then calculated using the Point2Plane ICP algorithm to calculate the registration root mean square error between the image model point cloud and the coarsely registered image model point cloud and the new registration matrix M2 (new translation transformation matrix and new rotation transformation matrix). The new registration matrix M2 is then iterated on the registration matrix M1 obtained by coarse registration, and the current registration root mean square error is calculated again. When the root mean square error is less than 1 mm, the iteration is stopped, and the registration matrix M2 is the optimized translation transformation matrix and the optimized rotation transformation matrix. When the root mean square error between the point cloud obtained by the transformation based on the registration matrix M2 and the model point cloud is still greater than 1 mm, the registration matrix M3 obtained by the Point2Plane ICP algorithm based on the registration matrix M2 is iterated on the registration matrix M2 until the root mean square error between the new point cloud and the model point cloud is less than 1 mm, and the iteration is stopped.

[0075] The image model point cloud S after registration world The expression is:

[0076]

[0077] Where M2 represents the optimized translation transformation matrix and the optimized rotation transformation matrix, C i Represents the model point cloud, S mri Represents an image model point cloud.

[0078] like Figure 3 As shown, the error between the subject point cloud and the model point cloud obtained by the first registration method (i.e., coarse registration) is less than 8 mm. After the second registration method (i.e., fine registration), the error between the two is reduced to 1 mm. Figure 4 is the model point cloud before registration, and its nose tip coordinates are (-38.0234, -4.60849, -588.799). Figure 5 The point cloud of the subject before registration has the coordinates of the nose tip as (-38, -4, -590). The coordinate distance between the two points before registration is 1.35 mm. After coarse and fine registration, the coordinate distance between the two points is 1.35 mm. Figure 3As shown, the registration accuracy between the two (i.e., the coordinate distance between the two points) is 1 mm.

[0079] After the second registration algorithm completes the precise registration, the optical camera stops working. At this time, the infrared camera continues to work at 30FPS, outputting the displacement matrix between the current frame and the previous frame in real time to capture the subject's head movement.

[0080] Then, the subject's point cloud position is adjusted in real time according to the subject's head position motion information to obtain a real-time subject point cloud.

[0081] For example, an infrared camera is used to track the subject's locator, and the real-time displacement matrix between the two frames is calculated based on the coordinates of the locator in the current frame and the coordinates of the locator in the previous frame. The aligned image model point cloud is then adjusted based on the real-time displacement matrix to obtain the aligned real-time point cloud of the current frame.

[0082] Since the image model point cloud and the subject point cloud are accurately registered in the early stage, in the subsequent real-time registration, it is only necessary to realize the real-time registration of the image model point cloud and the subject point cloud based on the accurately registered image model point cloud and the real-time displacement matrix.

[0083] For example, after the image model point cloud and the subject point cloud are registered using the first and second registration methods, a registered image model point cloud is obtained. Since the subject point cloud contains the point cloud acquisition time, the displacement matrix at the corresponding time can be found from the data set of the infrared camera tracking locator (e.g., a reflective ball, infrared locator) based on this time. For example, the infrared camera captures data frames with timestamps, and based on the timestamp of each frame, the displacement matrix corresponding to the subject point cloud acquisition time can be obtained. This design can further improve the accuracy of the registration.

[0084] For example, the acquisition time of the subject's point cloud is consistent with the acquisition time of the locator tracking data of the third frame of the infrared camera (i.e., the first data frame) or is within the preset error range, and the displacement relationship between the fourth frame and the third frame is the first displacement matrix. Then the real-time point cloud of the fourth frame (i.e., the second data frame) after registration is the registered image model point cloud. The registered real-time point cloud of the fourth frame is obtained based on the first displacement matrix. At this time, the drawing time of the real-time point cloud of the fourth frame after registration is the time indicated on the timestamp of the fourth frame.

[0085] The registered real-time point cloud of the fifth frame (i.e., the second data frame) is then obtained based on the registered real-time point cloud of the fourth frame and the real-time displacement relationship between the fourth and fifth frames. The time at which the registered real-time point cloud of the fifth frame is drawn is the time indicated by the timestamp of the fifth frame. This process is analogous and will not be repeated here. This achieves real-time registration of the image model point cloud and the subject point cloud.

[0086] That is to say, the first displacement matrix and the real-time displacement matrix output by the infrared camera can obtain the real-time position changes of the subject, such as head movement transformation. Therefore, based on the first displacement matrix and the corresponding real-time displacement matrix, the real-time registration point cloud of the subject in each frame can be obtained, and then the real-time registration of the model point cloud and the subject point cloud can be realized.

[0087] The "first data frame" and "second data frame" mentioned here are only used for purpose description and should not be understood as a description or limitation on the order of frames, nor should they be understood as a description or limitation on the number of frames, that is, the second data frame can be one frame or multiple different frames.

[0088] In one example, a readable storage medium is provided according to another embodiment of the present invention. The "readable storage medium" of an embodiment of the present invention refers to any medium that participates in providing a program or instruction to a processor for execution. The medium can take a variety of forms, including but not limited to non-volatile media, volatile media, and transmission media. Non-volatile media include, for example, optical or magnetic disks, such as storage devices. Volatile media include dynamic memory, such as main memory. Transmission media include coaxial cables, copper wires, and optical fibers, including wires comprising buses. Transmission media can also take the form of sound waves or light waves, such as those generated during radio frequency (RF) and infrared (IR) data communications. Common forms of readable storage media include, for example, floppy disks, flexible disks, hard disks, magnetic tape, any other magnetic medium, CD-ROMs, DVDs, any other optical medium, punch cards, paper tape, any other physical medium with a pattern of holes, RAM, PROMs and EPROMs, FLASH-EPROMs, any other memory chips or cartridges, carrier waves as described below, or any other medium from which a computer can read. The readable storage medium stores a program or instruction, and when the program or instruction is executed by the processor, the above-mentioned registration method is executed.

[0089] In one example, the readable storage medium is arranged on a server in the form of a memory, such as a cloud server. The cloud server is also provided with a processor, which executes the program or instruction stored in the memory. The processor may be a central processing unit (CPU). In one example, the cloud server may be a virtual server formed by mapping a physical server through virtualization technology. Among them, there may be one or more physical servers. When there are multiple physical servers, the cloud server may be a virtual server formed by mapping a server cluster through virtualization technology. The virtual server may also be one or more. In one example, the cloud server can be provided to users through a cloud platform. The registration method and the readable storage medium according to an embodiment of the present invention have at least one of the following advantages:

[0090] (1) The registration method and readable storage medium provided by the present invention can, in the initial registration stage, use a three-dimensional optical lens to capture a facial point cloud, and then register it with the scalp point cloud generated by the MRI image, thereby achieving the initial positioning of the subject's face. In this way, in the subsequent registration stage, only the subject's displacement needs to be tracked to achieve real-time registration;

[0091] (2) The registration method provided by the present invention realizes the initial registration and real-time registration of the entity point cloud and the model point cloud by using the optical point cloud, thereby ensuring that the target position of the subject obtained in the entity point cloud and the target position planned in the model point cloud can be accurately and in real time coincident;

[0092] (3) The registration method and readable storage medium provided by the present invention achieve the initial registration of the entity point cloud and the model point cloud by using an optical point cloud. Compared with the traditional method of using a positioning rod to register several key points of facial physiological structures with the same position of the model point cloud, it solves the problem that the initial registration stage is too time-consuming and the registration failure is easily caused by head movement during the registration;

[0093] (4) The registration method and readable storage medium provided by the present invention achieve the initial registration of the entity point cloud and the model point cloud by using an optical point cloud. After that, only the infrared positioning system needs to be used to track the subject's displacement to achieve real-time registration, thereby avoiding the danger of using a positioning rod to the subject during the registration process and improving the subject's experience.

[0094] (5) The registration method and readable storage medium provided by the present invention perform model point cloud preprocessing in advance during the optical registration process, thereby reducing the time consumption of the overall registration;

[0095] (6) The registration method and readable storage medium provided by the present invention achieve real-time tracking of the image model point cloud in the real world by using two different cameras to take advantage of their respective strengths.

[0096] Although some embodiments of the present general inventive concept have been shown and described, it will be appreciated by those skilled in the art that changes may be made to these embodiments without departing from the principles and spirit of the present general inventive concept, the scope of which is defined in the claims and their equivalents.

Claims

1. A registration method, comprising the following steps: generating an image model point cloud based on a medical imaging image of a subject; Registering the image model point cloud to the space where the subject point cloud is located to obtain a registered image model point cloud; adjusting the registered image model point cloud based on a first displacement relationship between a first data frame of a target object on the subject and a second data frame thereof to obtain a registered real-time point cloud; Based on the real-time displacement relationship between the current data frame and the previous data frame of the target object and the real-time point cloud after registration, the image model point cloud is registered in real time to the space where the subject's point cloud is located.

2. The registration method according to claim 1, wherein: The first displacement relationship and the real-time displacement relationship are both obtained through an infrared positioning system, and the subject point cloud is obtained through a stereo camera.

3. The registration method according to claim 1, wherein: Based on a first displacement relationship between a first data frame and a second data frame of a target object on a subject, the registered image model point cloud is adjusted to obtain a registered real-time point cloud, wherein the acquisition time of the subject point cloud corresponds to the timestamp in the first data frame, and the registered real-time point cloud is the real-time point cloud corresponding to the timestamp in the second data frame.

4. The registration method according to claim 3, wherein: Based on the real-time displacement relationship between the current data frame and the previous data frame of the target object and the real-time point cloud after registration, the image model point cloud is registered in real time to the space where the subject's point cloud is located. The output real-time point cloud after registration is the real-time point cloud corresponding to the timestamp in the current data frame.

5. The registration method according to any one of claims 1 to 4, wherein: The first displacement relationship and the real-time displacement relationship are both displacement matrices corresponding to the real-time position change of the subject.

6. The registration method according to any one of claims 1 to 4, wherein: In registering the image model point cloud to the space where the subject point cloud is located to obtain a registered image model point cloud, the method of registering the image model point cloud to the space where the subject point cloud is located includes the following steps: Preprocessing the image model point cloud and the subject point cloud to obtain a preprocessed subject point cloud and a preprocessed image model point cloud; Obtaining a spatial transformation relationship between the preprocessed image model point cloud and the preprocessed subject point cloud by a first registration method; Based on the spatial transformation relationship, the preprocessed image model point cloud is registered to the space where the preprocessed subject point cloud is located through a second registration method to obtain the registered image model point cloud.

7. The registration method according to claim 6, wherein: The point cloud preprocessing includes the following steps: Performing denoising and downsampling processing on the image model point cloud and the subject point cloud to obtain a downsampled image model point cloud and a downsampled subject point cloud; The local surface features of the point clouds in the downsampled image model point cloud and the downsampled subject point cloud are extracted to obtain the model point cloud features and the subject point cloud features.

8. The registration method according to claim 7, wherein: The first registration method is to match the model point cloud features with the subject point cloud features through the RANSAC algorithm and obtain the translation transformation matrix and rotation transformation matrix for aligning the preprocessed image model point cloud to the space where the preprocessed subject point cloud is located.

9. The registration method according to claim 8, wherein: The second registration method comprises the following steps: Optimizing the translation transformation matrix and the rotation transformation matrix by an ICP algorithm to obtain the optimized translation transformation matrix and the optimized rotation transformation matrix; Based on the optimized translation transformation matrix and the optimized rotation transformation matrix, the preprocessed image model point cloud is aligned to the space where the preprocessed subject point cloud is located to obtain the aligned image model point cloud.

10. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by a processor, the registration method according to any one of claims 1 to 9 is executed.

Images