Three-dimensional modeling and dynamic adaptation operation method and system for facial contour reshaping

Abstract

This invention discloses a method and system for 3D modeling and dynamic adaptation in facial contouring surgery, specifically relating to the field of computer data processing technology. The method includes acquiring a preoperative 3D basic model of the patient's face at the surgical end and discretizing it to generate the current 3D model. The surgical end establishes observation correspondences based on continuous intraoperative observation data and generates a set of usable regions and event candidate regions. Within the event candidate regions, a local displacement vector field is calculated to determine update trigger events and their affected areas. Based on the set of interactive anchor points and the set of update forbidden zones within the event affected regions, an event constraint set is generated. The current 3D model and the event constraint set are then sent to the cloud to solve for candidate local update results. The surgical end performs consistency checks on the candidate local update results and decides whether to update the current 3D model accordingly, recording traceable information to achieve end-to-cloud collaboration.

Description

Technical Field

[0001] This invention relates to the field of computer data processing technology, and more specifically, to a method and system for three-dimensional modeling and dynamic adaptation of facial contouring. Background Technology

[0002] Currently, in facial contouring surgery, the industry generally hopes to bring the 3D model established before surgery into the surgical procedure, so that doctors can intuitively see the contour changes during the operation and adjust the cutting amount or implant position accordingly. A common practice is to build a relatively detailed 3D model using scan data before surgery, and to continuously collect facial images or depth data during surgery, while continuously performing model alignment and deformation updates on the surgical end. When the computing power of the surgical end is insufficient to complete certain calculations within the required time, the practice is to send the computing tasks and necessary data related to the current update to the cloud server to perform more complex optimization or reconstruction, and then return the update results obtained from the cloud calculation to the surgical end for display and guidance, so as to achieve end-to-cloud collaboration. In actual surgery, surgeons use traction and compression to expose the surgical field, causing significant changes in soft tissue within a short period. Simultaneously, instruments frequently obstruct the view, and blood and reflections can blur or distort local images. Furthermore, the surgical end requires extremely fast feedback, and network fluctuations can lead to incomplete or unstable data at times. Under these circumstances, mainstream continuous update methods repeatedly exhibit the same phenomenon: the model continues to update even when the view is unclear or incorrect, and the update direction is skewed by erroneous data. This deviation gradually spreads from local to overall, manifesting as the contour lines increasingly deviating from the surgeon's confirmed position, and the left-right symmetry being subtly disrupted. Even after the occlusion disappears, it is difficult for the model to automatically return to the correct shape, ultimately resulting in a seemingly smooth but actually deviated result. The technical problem this application aims to solve is how to ensure that every update of the 3D model has verifiable evidence under constraints such as edge-cloud collaboration and intraoperative occlusion and traction, thereby proactively ceasing updates and preventing the spread of deviations when data is unreliable. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method and system for three-dimensional modeling and dynamic adaptation of facial contouring. The method establishes an observation correspondence relationship based on continuous intraoperative observation data at the surgical end and generates a set of available regions and event candidate regions. Within the event candidate regions, a local displacement vector field is calculated to determine update trigger events and their affected regions. Based on the set of interactive anchor points and the set of update forbidden zones within the event affected regions, an event constraint set is generated. The current three-dimensional model and the event constraint set are then sent to the cloud to solve for candidate local update results. The surgical end performs consistency verification on the candidate local update results and determines whether to update the current three-dimensional model accordingly, recording traceable information to address the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a three-dimensional modeling and dynamic adaptation method for facial contouring, comprising: S1. Obtain the patient's preoperative three-dimensional basic model at the surgical end and generate the current three-dimensional model discretely. Receive the set of interactive anchor points and the set of updated restricted areas input by the doctor on the current three-dimensional model and write them into the model state data structure. S2. Collect continuous intraoperative observation data at the surgical end, perform feature tracking on intraoperative observation data at adjacent time points to obtain the observation correspondence, and generate a set of usable regions based on the proportion of continuously trackable feature points in each spatial region, and generate event candidate regions accordingly. S3. Calculate the local displacement vector field based on the observation correspondence within the event candidate area, and determine the updated triggering event and its affected area accordingly. The traction event is determined by keeping the main direction of the local displacement vector field consistent during continuous acquisition. The contact event is determined by keeping the occlusion boundary formed within the event candidate area consistent during continuous acquisition. The doctor confirmation event is determined by the doctor's confirmation input of the contour point or symmetry line. S4. For each update trigger event, generate an event constraint set based on the event influence area, the set of interaction anchor points, and the update restricted area set. The event constraint set includes the freeze constraint that the vertex displacement in the update restricted area set is zero, the maintenance constraint that the interaction anchor point set satisfies the observation correspondence when projected onto the intraoperative observation data before and after the update, and the constraint that the dot product of the displacement vector and the main traction direction in the event influence area is non-negative and the displacement magnitude is not greater than the maximum local displacement magnitude in the available area set under the traction event, or the constraint that the normal component of the displacement vector in the event influence area has the same sign as the normal component of the local displacement vector field under the contact event. S5. The surgical end sends a candidate update request containing the current 3D model and event constraint set to the cloud, and receives the candidate local update result returned by the cloud. The candidate local update result is the set of displacements of vertices within the event-affected area. S6. The surgical end substitutes each candidate local update result into the event constraint set to perform consistency verification to obtain the admission judgment result. The consistency verification includes at least verifying that the vertex displacement in the update forbidden zone set is zero, verifying that the projection of the interactive anchor point set satisfies the observation correspondence, and verifying that the displacement in the event-affected area satisfies the corresponding event constraint. When all consistency verifications are successful, the candidate local update results are applied to the current 3D model to generate the updated current 3D model. Otherwise, the current 3D model remains unchanged and a rejection item flag is generated. S7. Write the update trigger event, event constraint set, candidate local update results and admission judgment results into the update record sequence in chronological order for retrospective display and rollback calculation.

[0005] In a preferred embodiment, S1 includes: S1-1. Receive the preoperative three-dimensional basic model data of the face at the surgical end, perform connected component segmentation on the preoperative three-dimensional basic model data of the face to obtain the maximum connected component, and perform hole detection and hole filling on the maximum connected component to form a closed mesh, and output the closed mesh. S1-2. Calculate the face normal for each facet on the closed mesh and calculate the vertex curvature field for each vertex based on the face normal of its one-ring neighborhood. Perform adaptive resampling on the closed mesh according to the vertex curvature field to generate the current 3D model containing the vertex set and facet topology, and output the current 3D model. S1-3. Receive the set of interactive anchor points and the set of updated restricted areas input by the doctor on the current 3D model. Map the set of interactive anchor points to a set of vertex identifiers and the set of updated restricted areas to a set of face identifiers. Write the current 3D model, the set of vertex identifiers, and the set of face identifiers into the model state data structure and output the model state data structure.

[0006] In a preferred embodiment, S2 includes: Feature point extraction is performed on two adjacent frames of intraoperative observation data at the surgical end, and window matching is performed on each feature point in the next frame of intraoperative observation data to form a feature matching pair. An observation correspondence is generated from the feature matching pair from the coordinates of the feature point in the previous frame to the coordinates of the matching point in the next frame, and the observation correspondence is output. S2-2. Divide the intraoperative observation data into multiple spatial regions according to a preset grid, and in each spatial region, calculate the ratio of the number of feature matching pairs falling into the spatial region to the total number of feature points in the spatial region to generate the sustainable tracking ratio field of the spatial region, and output the sustainable tracking ratio field. S2-3. Write the spatial regions with a sustainable tracking ratio field greater than zero and which generate observation correspondence in multiple consecutive pairs of adjacent frames into the available region set, and write the spatial regions that are adjacent to any available region set spatial regions, contain feature points in the current frame, and do not form feature matching pairs in the observation correspondence into the event candidate region, and output the available region set and the event candidate region.

[0007] In a preferred embodiment, S3 includes: S3-1. At the surgical end, for each spatial region within the event candidate region, calculate the set of local displacement vectors of the spatial region based on the observation correspondence and form a local displacement vector field. Calculate the main direction field and direction dispersion field of the local displacement vector field and write them into the displacement evidence record item. Output the displacement evidence record item. S3-2. At the surgical end, based on the displacement evidence record item, execute the direction consistency checker on the main direction field in multiple consecutive pairs of adjacent frames and execute the dispersion consistency checker based on the direction dispersion field to generate the traction confidence field and traction uncertainty field and write the traction token or pending token, and output the traction token or pending token and the corresponding traction confidence field and traction uncertainty field. S3-3. At the surgical end, for the same spatial region, calculate the occlusion boundary field based on the distribution of feature points in the event candidate region that have not formed an observation correspondence and write it into the occlusion evidence record item. Based on the occlusion evidence record item, execute the boundary consistency checker on the occlusion boundary field in multiple consecutive pairs of adjacent frames to generate the occlusion stability field and write it into the contact token or pending token. Output the contact token or pending token and the corresponding occlusion stability field.

[0008] In a preferred embodiment, S3 further includes: S3-4. Receive the doctor's confirmation input for the contour point or symmetry line at the surgical end and parse the confirmation input into the confirmation type field and confirmation location field to write to the confirmation evidence record item. At the same time, write the confirmation token and associate the confirmation evidence record item with the corresponding spatial area, and output the confirmation token and confirmation evidence record item. S3-5. At the surgical end, execute conflict resolution rules on the traction token, contact token, and confirmation token in the same spatial area to generate a unique event type field and update the event confidence field. The conflict resolution rules include generating a doctor confirmation event and writing an event lock status lock when the confirmation token exists; generating a contact event and writing an event lock status lock when the confirmation token does not exist but the contact token exists; generating a traction event and writing an event lock status lock when neither the confirmation token nor the contact token exists but the traction token exists; and writing a backtracking retest status lock when the pending token exists and reserving the spatial area as an event candidate area. Output the unique event type field, event confidence field, event lock status lock, or backtracking retest status lock. S3-6. At the surgical end, generate the event-affected area based on the unique event type field and the spatial region identifier field, and write the event-affected area, the event confidence field, the displacement evidence record item, the occlusion evidence record item or the confirmation evidence record item into the event record sequence, and output the update trigger event and its event-affected area.

[0009] In a preferred embodiment, S4 includes: S4-1. On the surgical end, for each update trigger event, read the event-affected region, the set of interactive anchor points, the set of update restricted areas, the observation correspondence, and the local displacement vector field. Map the event-affected region to the set of event-affected vertices, map the set of interactive anchor points to the set of anchor point vertices, and map the set of update restricted areas to the set of restricted area vertices. At the same time, construct constraint generation record items and write them into the event type field and the region identifier field. Output the set of event-affected vertices, the set of anchor point vertices, the set of restricted area vertices, and the constraint generation record items. S4-2. At the surgical end, using the set of restricted area vertices as the judgment object and the frozen consistency checker as the judgment basis, write a zero displacement freeze constraint to each vertex in the restricted area vertex set and write the freeze constraint identifier field. At the same time, using the set of anchor point vertices as the judgment object and the projection consistency checker as the judgment basis, write a projection preservation constraint to each anchor point in the anchor point vertex set so that the difference between the pixel coordinates of the anchor point vertex projected to the intraoperative observation data before and after the update is zero and write the preservation constraint identifier field. Output the basic constraint set and the basic constraint gating token. S4-3. At the surgical end, the event-affected vertex set is used as the judgment object and the event type field is used as the branch condition to generate event-specific constraints. When the event type field is a traction event, the upper limit of the displacement magnitude is calculated in the available area set based on the local displacement vector field, and the traction main direction field is calculated. The direction consistency checker is used as the judgment basis. A dot product constraint is written to each vertex in the event-affected vertex set so that the dot product of the vertex displacement vector and the traction main direction field is non-negative and the vertex displacement magnitude does not exceed the upper limit of the displacement magnitude. The traction constraint token and traction constraint identifier fields are written. When the event type field is a contact event, the surface normal field is solved based on the event-affected area. The normal consistency checker is used as the judgment basis. A normal sign constraint is written to each vertex in the event-affected vertex set so that the normal component of the vertex displacement vector has the same sign as the normal component of the local displacement vector field. The contact constraint token and contact constraint identifier fields are written. The event-specific constraints and corresponding tokens are output. S4-4. At the surgical end, execute the token merging rule on the basic constraint gating token and the traction constraint token or contact constraint token to generate an event constraint set and write it into the constraint version field and the constraint entry count field. The token merging rule includes writing a state lock for rollback and re-checking when any consistency checker output fails and writing the event-affected vertex set into the list of regions to be rebuilt. When all consistency checkers output pass, write a state lock for constraint locking and associate the event constraint set with the constraint generation record item into the constraint record sequence. Output the event constraint set, state lock and written item.

[0010] In a preferred embodiment, S5 includes: S5-1. The surgical end constructs candidate update request objects based on the current 3D model, available area set, event-affected area and event constraint set, and generates a request summary field. The request summary field includes an event type field, an event-affected area identifier field, a constraint entry count field and an integrity verification value field calculated based on the request object. The request summary field is written into the request record item to generate an external token, and the external token and request object are output. S5-2. The surgical end sends the outgoing token and the request object to the cloud and receives the candidate partial update result and response summary field returned by the cloud. The response summary field includes at least the cloud version identifier field, the solution convergence status field and the result range identifier field. The surgical end performs a consistency checker on the response summary field and the outgoing token to generate a receive token or a rollback re-examination state lock. The consistency checker performs a matching check on the integrity check value field and a coverage consistency check on the result range identifier field and the event impact area identifier field. The candidate partial update result and the receive token or rollback re-examination state lock are output. S5-3. When receiving token generation, the surgical end executes a multi-objective quality scoring function on the candidate local update results to generate a quality scoring field and update the uncertainty field. The multi-objective quality scoring function is calculated based at least on the displacement energy field of the candidate local update results within the update restricted area set, the projection residual field at the interaction anchor point set, and the displacement continuity field at the boundary of the event-affected area. When the quality scoring field does not meet the locally stored threshold packet, a rejection flag is written and the rollback re-examination state lock is set to trigger a downgrade path or retransmission path. When the quality scoring field meets the threshold packet, an acceptance flag is written and the candidate local update results are output.

[0011] In a preferred embodiment, S6 includes: S6-1. Receive candidate local update results and event constraint set at the surgical end, construct verification task record item and write event identifier field, region identifier field, constraint version field and displacement entry count field. At the same time, group the candidate local update results according to the restricted area vertex set, anchor point vertex set and event-affected vertex set to form restricted area displacement field, anchor point displacement field and region displacement field. Output verification task record item and restricted area displacement field, anchor point displacement field and region displacement field. S6-2. Using the restricted area displacement field as the judgment object and the frozen consistency checker as the judgment basis, calculate the magnitude of the displacement vector of each vertex in the restricted area displacement field one by one, write a pass flag for entries with a magnitude equal to zero, write a failure flag for entries with a magnitude not equal to zero, and write a rejection item flag as restricted area failure. At the same time, write the gate token as restricted area token pass or restricted area token failure, and output the gate token and rejection item flag written item. S6-3. Using the anchor point displacement field and the correspondence with the observation as the judgment object and the projection consistency checker as the judgment basis, each anchor point vertex is projected into the intraoperative observation data coordinate system in the updated three-dimensional coordinates to obtain the updated pixel coordinates. The difference between the updated pixel coordinates and the target pixel coordinates given by the observation correspondence is calculated to obtain the anchor point residual vector. For entries where the anchor point residual vector is equal to zero, a pass flag is written. For entries where the anchor point residual vector is not equal to zero, a failure flag is written and a rejection flag is written as the anchor point failure. At the same time, a gating token is written as either the anchor point token pass or the anchor point token failure. The gating token and rejection flag are output as the entries written. S6-4. Using the region displacement field and the event constraint set as the judgment objects and the constraint consistency checker as the judgment basis, substitute each vertex displacement vector in the region displacement field into the freeze constraint, hold constraint and event-specific constraint in the event constraint set to obtain the constraint satisfaction mark sequence. Write a failure flag and a rejection item flag to the entries in the constraint satisfaction mark sequence that do not satisfy the flag, and write a gate token to the region failure flag. Write a pass flag and a gate token to the entries in the constraint satisfaction mark sequence that are all satisfied, and output the gate token and rejection item flag entries. S6-5. At the surgical end, execute the token merging rule on the restricted area token, anchor point token, and region token to generate the admission judgment result and state lock. The token merging rule includes writing a state lock to roll back and re-examine when any token fails, keeping the current 3D model unchanged, and outputting the rejection item flag. When the restricted area token, anchor point token, and region token all pass, write a state lock to update and lock, apply the candidate local update result to the current 3D model to generate the updated current 3D model, and output the admission judgment result as pass. Write the admission judgment result, state lock, and rejection item flag into the verification task record item, and output the admission judgment result and the updated current 3D model or rejection item flag.

[0012] In a preferred embodiment, S7 includes: S7-1. Receive update trigger event, event constraint set, candidate local update result and admission judgment result at the surgical end, generate update record entry and write it into the event type field, event influence area identifier field, constraint version field, displacement entry count field and judgment result field, and output update record entry; S7-2. Calculate the displacement summary field for the candidate local update results and the constraint summary field for the event constraint set. Write the displacement summary field and the constraint summary field into the update record entry and generate the record integrity check value field. Output the update record entry containing the record integrity check value field. S7-3. Append the update record entries to the update record sequence in the order of generation, and when the rollback instruction is received, read the candidate local update results of the target update record entries and perform sign inversion on the candidate local update results to generate a rollback displacement set. Apply the rollback displacement set to the current 3D model to output the current 3D model after rollback.

[0013] In a preferred embodiment, the 3D modeling and dynamic adaptation operating system for facial contouring includes a modeling and annotation module, a tracking and partitioning module, a solution and recognition module, a constraint generation module, a cloud-based solution module, a verification and admission module, and a record rollback module. The modeling and annotation module acquires the patient's preoperative three-dimensional basic model at the surgical end and generates the current three-dimensional model discretely. It receives the set of interactive anchor points and the set of updated restricted areas input by the doctor on the current three-dimensional model and writes them into the model state data structure. The tracking partition module collects continuous intraoperative observation data at the surgical end, performs feature tracking on intraoperative observation data at adjacent time points to obtain the observation correspondence, and generates a set of usable regions based on the proportion of continuously trackable feature points in each spatial region, and generates event candidate regions accordingly. The solution and recognition module calculates the local displacement vector field based on the observation correspondence within the event candidate area, and determines and updates the triggering event and its affected area accordingly. The traction event is determined by keeping the main direction of the local displacement vector field consistent during continuous acquisition, the contact event is determined by keeping the occlusion boundary formed within the event candidate area consistent during continuous acquisition, and the doctor confirmation event is determined by the doctor's confirmation input of the contour point or symmetry line. The constraint generation module is used to generate an event constraint set for each update trigger event based on the event influence area, the set of interaction anchor points, and the set of update restricted areas. The event constraint set includes the freeze constraint that the vertex displacement in the update restricted area set is zero, the preservation constraint that the interaction anchor point set satisfies the observation correspondence when projected onto the intraoperative observation data before and after the update, and the constraint that the dot product of the displacement vector and the main traction direction in the event influence area is non-negative and the displacement magnitude is not greater than the maximum local displacement magnitude in the available area set under the traction event, or the constraint that the normal component of the displacement vector in the event influence area has the same sign as the normal component of the local displacement vector field under the contact event. The cloud-based solution module is used by the surgical end to send a candidate update request containing the current 3D model and the set of event constraints to the cloud, and to receive the candidate local update results returned by the cloud. The candidate local update results are the set of displacements of vertices within the event's influence area. The admission verification module is used by the surgical end to substitute each candidate local update result into the event constraint set to perform consistency verification in order to obtain the admission judgment result. The consistency verification includes at least verifying that the vertex displacement in the update forbidden zone set is zero, verifying that the projection of the interactive anchor point set satisfies the observation correspondence, and verifying that the displacement in the event-affected area satisfies the corresponding event constraint. When all consistency verifications are successful, the candidate local update result is applied to the current 3D model to generate the updated current 3D model; otherwise, the current 3D model remains unchanged and a rejection item flag is generated. The rollback module is used to write the update trigger event, event constraint set, candidate local update results and admission judgment results into the update record sequence in chronological order for retrospective display and rollback calculation.

[0014] The technical effects and advantages of this invention are as follows: Based on identifying events and generating event constraint sets only within the event candidate region, and allowing updates only after verifying the consistency of each candidate displacement in the cloud at the surgical end, it can relatively suppress the application of unreliable updates when the observation is incomplete or distorted, thereby mitigating bias propagation and stability drift. Based on the feature tracking of adjacent frames to form the observation correspondence and the statistical sustainable tracking ratio according to spatial regions to generate a set of available regions and event candidate regions, displacement calculation and event identification can be limited to the region with observation support, thereby relatively reducing the impact of abnormal observations on the overall update. Based on the continuous consistency of the main direction of local displacement for traction events and the continuous consistency of the occlusion boundary for contact events, the affected area of ​​the events is output. This allows the triggering basis and the affected area to have reproducible values, thereby relatively improving false triggering and range drift. By freezing constraints, the displacement of vertices within the restricted area is limited to zero, and the interaction anchor point projection and observation correspondence are bound by constraints. This enables the formation of constraint links between the restricted area and anchor points during the solution and verification process, thereby relatively reducing the probability of critical areas being erroneously updated. Based on the cloud-based solution of the candidate displacement set within the event-affected area and the consistency verification and admission determination performed by the surgical end, edge-cloud collaboration is achieved. This allows for maintaining a locally controllable update entry while moving heavy computational loads externally, thereby relatively improving update stability under conditions of computing power and network fluctuations. By writing the candidate displacements and admission judgment results of the event constraint set of the update trigger event into the update record sequence in sequence, traceable data can be provided for retrospective display and rollback calculation, thereby relatively shortening the location and recovery path. Attached Figure Description

[0015] Figure 1 This is a flowchart of the present invention.

[0016] Figure 2 This is a schematic diagram of the system modules of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Refer to the instruction manual appendix Figure 1-2 The three-dimensional modeling and dynamic adaptation method for facial contouring of the present invention includes: S1. Obtain the patient's preoperative three-dimensional basic model at the surgical end and generate the current three-dimensional model discretely. Receive the set of interactive anchor points and the set of updated restricted areas input by the doctor on the current three-dimensional model and write them into the model state data structure. This implementation describes how, before the start of dynamic adaptation during surgery, the patient's preoperative 3D facial base model is converted into a unified current 3D model on the surgical end. The doctor's input set of interactive anchor points and update restricted areas are converted into a computable set of vertex identifiers and a set of face identifiers, and then written into the model state data structure. This allows subsequent steps to directly read and reference these sets for constraint generation and consistency verification. The overall principle of this process is to first unify coordinates and extract the facial subject, then repair the input data into a basic mesh capable of geometric operations, subsequently use curvature-driven resampling to form a current 3D model with controlled vertex density and consistent topology, and finally map the doctor's input from the interface operation to vertex and face numbers and solidify them in the model state data structure. This implementation process includes the following steps: S1-1. The surgical end imports the patient's preoperative facial 3D basic model from the preoperative planning software and reads the model's data shape information and coordinate system information. Based on the coordinate system information, it performs coordinate transformation to convert the 3D coordinates to the surgical end model coordinate system. Based on the point cloud Euclidean adjacency relationship or the mesh shared edge adjacency relationship, it constructs the connectivity relationship and traverses to obtain the connected components. The connected component with the largest number of points or faces is selected as the facial main data and output. S1-2. The surgical end generates a basic mesh using facial main data as input. When the facial main data is a point cloud, it performs a nearest neighbor search for each point to obtain a set of nearest neighbor points and performs least squares plane fitting on the set of nearest neighbor points to obtain the normal of the point. Based on the point coordinates and the point normal, it performs triangulation to generate an initial triangular mesh and outputs it. When the facial main data is a triangular mesh, it detects the boundary edges and forms boundary loops. It performs constrained triangulation on each boundary loop to complete hole filling and outputs it. It performs facet intersection detection to locate the intersection region and performs deletion of intersecting facets and local retriangulation on the intersection region to eliminate self-intersection and outputs a basic mesh without holes and without self-intersection. S1-3. The surgical end generates the current 3D model with the basic mesh as input. It calculates the face normal for each face and the vertex normal for each vertex based on the face normal of the one-ring neighborhood. It calculates the vertex curvature field based on the change of the vertex normal in the one-ring neighborhood and writes it into the vertex attribute. It calculates the face target edge length field according to the vertex curvature field of the three vertices of the face and restricts the target edge length between the lower limit and the upper limit of the edge length. It traverses the set of mesh edges and performs edge splitting on edges with an edge length greater than the target edge length and updates the topology of adjacent faces. It performs edge collapse on edges with an edge length less than the target edge length and meets the topology preservation condition and updates the topology of adjacent faces. After each edge splitting or edge collapse, it performs local retriangulation to keep the mesh connectivity consistent. It outputs the current 3D model containing the vertex set and face topology and fixes the vertex numbering rules and face numbering rules. S1-4. The surgical end receives the set of interactive anchor points and the updated restricted area set input by the doctor on the current 3D model. It converts the screen coordinates of the doctor's point or line selection into rays and finds the intersection with the current 3D model face to obtain the intersection point and the hit face. It maps the intersection point to the nearest vertex number of the hit face and forms a list of anchor point vertex numbers. It projects the boundary of the restricted area selected by the doctor onto the current 3D model face set and fills the region through face adjacency traversal to obtain a list of restricted area face numbers and expands it into a list of restricted area vertex numbers. It writes the current 3D model, the list of anchor point vertex numbers, the list of restricted area face numbers, and the list of restricted area vertex numbers into the model state data structure and outputs them. Through the above implementation process, the surgical end obtains a current 3D model with consistent topology and controlled vertex density. The set of interactive anchor points and the set of update restricted areas are written into the model state data structure in the form of vertex numbers and face numbers. This allows subsequent steps to directly construct projection-maintaining constraints on the anchor point vertex number list and displacement-freezing constraints on the restricted area vertex number list, and to perform consistency checks line by line. In practical applications: During the preoperative planning stage of mandibular angle contour adjustment, the surgeon imports the patient's preoperative 3D data, and the surgical end generates the current 3D model. The surgeon selects the left and right mandibular angles on the model to form an anchor point vertex number list and circles the restricted area face number list corresponding to the inferior alveolar nerve's course. The surgical end expands the restricted area face number list into a restricted area vertex number list and writes it together with the current 3D model into the model state data structure. After entering the intraoperative dynamic adaptation stage, the system reads this model state data structure to reference the anchor points and restricted areas to perform subsequent constraint generation and consistency checks, thereby keeping the anchor point positions unchanged and preventing displacement of the restricted area during model updates.

[0019] S2. Collect continuous intraoperative observation data at the surgical end, perform feature tracking on intraoperative observation data at adjacent time points to obtain the observation correspondence, and generate a set of usable regions based on the proportion of continuously trackable feature points in each spatial region, and generate event candidate regions accordingly. This implementation describes how, during the intraoperative dynamic adaptation phase, the surgical end establishes cross-frame observation correspondences for continuously acquired intraoperative observation data. After dividing the observation space into multiple spatial regions, a set of usable regions is generated based on the proportion of continuously trackable feature points within each spatial region. Then, event candidate regions are generated in the adjacent regions of the usable region set, providing directly referenceable region inputs for subsequent local displacement vector field calculation and event recognition updates. The overall principle of this process is to first establish feature matching pairs between adjacent frames to form observation correspondences, then statistically analyze these observation correspondences by spatial region to obtain a continuously trackable proportion field, and finally, based on this proportion field and the adjacency relationship, form two types of region sets while maintaining cross-frame numbering consistency to support subsequent calculations. This implementation process includes the following steps: S2-1. The surgical end collects continuous intraoperative observation data at a fixed sampling frequency and writes each frame of intraoperative observation data into the observation buffer queue. It selects two adjacent frames in the observation buffer queue as the previous frame and the next frame, and performs grayscale normalization and denoising on the two frames to generate a preprocessed frame for feature tracking. Grayscale normalization maps pixel values ​​to a fixed range through linear mapping, and denoising suppresses isolated noise points through local mean filtering or median filtering. The previous frame and the next frame are then output. S2-2. The surgical end performs feature point extraction in the previous frame to obtain a feature point set, and performs window matching on each feature point in the feature point set in the next frame to form a feature matching pair. Feature point extraction is performed by calculating the pixel gradient and generating corner feature points at the local maximum of the gradient magnitude. Window matching is performed by cropping a search window of a fixed size with the feature point as the center and calculating the similarity within the search window and taking the position with the highest similarity as the matching point. At the same time, feature points that fail to match are marked as unmatched and removed from the feature matching pair. The output shows the observation correspondence from the feature point coordinates in the previous frame to the matching point coordinates in the next frame. S2-3. The surgical end divides the observation space of the previous frame into multiple spatial regions according to a preset grid and assigns a spatial region identifier to each spatial region. For each spatial region, it counts the total number of feature points in the spatial region and the number of feature matching pairs that form observation correspondence in the spatial region. It then calculates the sustainable tracking ratio field of the spatial region by dividing the number of feature matching pairs by the total number of feature points. At the same time, it writes the sustainable tracking ratio field and the spatial region identifier into the region statistics table and outputs it. S2-4. The surgical end generates a set of available regions and event candidate regions based on the region statistics table. The spatial region identifiers that have a continuous tracking ratio field greater than zero and have been written into the region statistics table in multiple consecutive pairs of adjacent frames are written into the set of available regions. The spatial region identifiers that share a boundary in the grid with any spatial region identifier in the set of available regions are used as adjacent spatial region identifiers. When the spatial region corresponding to the adjacent spatial region identifier satisfies that the total number of feature points is greater than zero and the number of feature matching pairs is zero, the adjacent spatial region identifier is written into the event candidate region. The set of available regions and event candidate regions are output. Through the above implementation process, the surgical end establishes a directly referential observation correspondence between two adjacent frames of intraoperative observation data. The observation space is divided into a set of available regions and event candidate regions using a sustainable tracking ratio field obtained by spatial region statistics. This allows subsequent steps to calculate a stable local displacement vector field within the set of available regions and to centrally identify abnormal change regions caused by occlusion, traction, or instrument contact within the event candidate regions. This reduces the propagation of biases caused by direct updates in incomplete observation areas. In practical applications: the surgical end acquires surgical field video at a fixed frame rate as intraoperative observation data and divides the image into spatial regions according to a fixed grid. When certain spatial regions can form feature matching pairs in multiple consecutive frames, these spatial regions are written into the set of available regions. When certain spatial regions are adjacent to the set of available regions and have consecutive feature points but cannot form feature matching pairs, these spatial regions are written into the event candidate regions. Subsequent steps further calculate the local displacement vector field within the event candidate regions and identify traction or contact events to drive event constraint generation.

[0020] S3. Calculate the local displacement vector field based on the observation correspondence within the event candidate area, and determine the updated triggering event and its affected area accordingly. The traction event is determined by keeping the main direction of the local displacement vector field consistent during continuous acquisition. The contact event is determined by keeping the occlusion boundary formed within the event candidate area consistent during continuous acquisition. The doctor confirmation event is determined by the doctor's confirmation input of the contour point or symmetry line. This implementation describes how, after obtaining the observation correspondence, available region set, and event candidate region, the surgical end calculates a local displacement vector field within the event candidate region. It then utilizes the cross-frame consistency of the local displacement vector field with the occlusion boundary and the doctor's confirmation input to determine the update trigger event and its affected region. The overall principle of this process is as follows: first, the observation correspondence within the event candidate region is converted into displacement vectors and aggregated by spatial region to form a local displacement vector field. Next, the principal direction field is extracted from the local displacement vector field and continuous frame consistency is checked to identify traction events. Simultaneously, the occlusion boundary field is extracted from the event candidate region and continuous frame consistency is checked to identify contact events. Finally, the doctor's confirmation input for contour points or symmetry lines is parsed into a confirmation position field that can be located in the spatial region to identify the doctor's confirmation event, and the event affected region is generated accordingly for subsequent constraint generation. This implementation process includes the following steps: S3-1. The surgical end reads the observation correspondence within the event candidate area and calculates the displacement vector for each feature matching pair. The displacement vector is obtained by subtracting the feature point coordinates from the coordinates of the matching point in the previous frame from the coordinates of the matching point in the next frame. The displacement vectors are then aggregated into a set of local displacement vectors according to the spatial region identifier. The spatial region identifier is determined by the spatial region into which the coordinates of the feature point in the previous frame fall. The set of local displacement vectors indexed by the spatial region identifier is output. S3-2. The surgical end identifies the local displacement vector set corresponding to each spatial region, calculates the local displacement vector field, and generates the main direction field and the direction dispersion field. The local displacement vector field is represented by the set of displacement vectors in the spatial region. The main direction field is obtained by calculating the direction angle of the displacement vector set and counting the direction angle interval with the most frequent occurrences. The direction dispersion field is obtained by averaging the absolute values ​​of the difference between the direction angle of the displacement vector set and the direction angle of the main direction field. The main direction field and the direction dispersion field are written into the displacement evidence record item of the spatial region, and the displacement evidence record item is output. S3-3. The surgical end performs traction consistency verification on displacement evidence records of the same spatial region in multiple consecutive pairs of adjacent frames and determines the traction event. The traction consistency verification compares the main direction field frame by frame in multiple consecutive frames and writes the traction event identifier on the spatial region identifier where the main direction field is the same and the direction dispersion field is the same frame by frame. The set of spatial region identifiers with written traction event identifiers generates the traction event affected area and outputs the traction event and the traction event affected area. S3-4. The surgical end extracts the occlusion boundary field based on the observation correspondence within the event candidate area and performs boundary consistency verification in multiple consecutive pairs of adjacent frames to determine the contact event. The occlusion boundary field is obtained by marking spatial regions that do not form an observation correspondence on the spatial region grid and extracting the outer contour of the marked region. The boundary consistency verification compares the set of spatial region boundaries corresponding to the outer contour frame by frame and writes the contact event identifier when the boundary sets are the same. The set of spatial region identifiers covered by the outer contour written with the contact event identifier is used to generate the contact event affected area. The contact event and the contact event affected area are output. S3-5. The surgical end receives the doctor's confirmation input for the contour point or symmetry line and parses it into a confirmation type field and a confirmation location field. The confirmation location field is the projection coordinate of the model surface position confirmed by the doctor in the current frame. The surgical end maps the confirmation location field to the spatial region grid to obtain the confirmation spatial region identifier and writes it into the doctor confirmation event identifier. It also generates the doctor confirmation event influence area from the confirmation spatial region identifier and outputs the doctor confirmation event and the doctor confirmation event influence area. S3-6. The surgical end performs event aggregation on traction events, contact events, and doctor confirmation events to generate update trigger events and their event influence areas. When a doctor confirmation event exists, the update trigger event is identified as a doctor confirmation event and the influence area of ​​the doctor confirmation event is output. When a doctor confirmation event does not exist but a contact event exists, the update trigger event is identified as a contact event and the influence area of ​​the contact event is output. When neither a doctor confirmation event nor a contact event exists but a traction event exists, the update trigger event is identified as a traction event and the influence area of ​​the traction event is output. The update trigger event and its event influence area are output. Through the above implementation process, the surgical end can convert the observation correspondence into a local displacement vector field within the event candidate region, and identify traction events and contact events by means of continuous frame consistency verification. At the same time, it maps the doctor's confirmation input to a spatial region identifier to form a doctor confirmation event, thereby outputting an update trigger event and its event influence region that can be directly used for subsequent event constraint generation. This avoids the spread of bias caused by directly using abnormal observations during occlusion or traction for model updates. In practical applications: during the mandibular angle exposure stage, the surgical end detects that the main direction fields of multiple adjacent spatial regions remain consistent for multiple consecutive frames and generates a traction event influence region, or detects that the outer contour of the event candidate region remains consistent for multiple consecutive frames and generates a contact event influence region. When the doctor confirms a symmetrical line on the display interface, the surgical end maps the confirmation position field to a confirmation spatial region identifier and generates a doctor confirmation event influence region. Subsequently, the surgical end outputs the determined update trigger event and the corresponding event influence region to the subsequent constraint generation steps to construct frozen constraints, hold constraints, and event-specific constraints.

[0021] S4. For each update trigger event, generate an event constraint set based on the event influence area, the set of interaction anchor points, and the update restricted area set. The event constraint set includes the freeze constraint that the vertex displacement in the update restricted area set is zero, the maintenance constraint that the interaction anchor point set satisfies the observation correspondence when projected onto the intraoperative observation data before and after the update, and the constraint that the dot product of the displacement vector and the main traction direction in the event influence area is non-negative and the displacement magnitude is not greater than the maximum local displacement magnitude in the available area set under the traction event, or the constraint that the normal component of the displacement vector in the event influence area has the same sign as the normal component of the local displacement vector field under the contact event. This implementation describes how, after obtaining the update trigger event and its affected region, the surgical end reads the set of interactive anchor points and the set of update restricted areas from the model state data structure, maps the affected region to the vertex set of the current 3D model, and generates a set of event constraints that can be directly used for cloud-based solving and local consistency verification. The overall principle of this process is to uniformly express constraints as calculation conditions for the displacement of the current 3D model vertices. First, freeze constraints on the update restricted area set and maintain constraints on the set of interactive anchor points are constructed. Then, event-specific constraints are generated according to the update trigger event type and merged with the basic constraints. Finally, an event constraint set with a clear target, clear calculation items, and clear value rules is output. This implementation process includes the following steps: S4-1. For each update trigger event, the surgical end reads the event-affected region, the set of interactive anchor points, the set of update restricted areas, the observation correspondence, the local displacement vector field, and the set of available regions. It then maps the event-affected region to the set of event-affected vertices. When the event-affected region is represented by a set of spatial region identifiers, the pixel range corresponding to the spatial region identifiers is projected onto the current 3D model to obtain the set of hit faces. The vertex numbers of the hit face sets are then combined to obtain the event-affected vertex set. When the event-affected region is represented by the model surface position, the vertex numbers of the face to which the position belongs and the vertex numbers of its one-ring neighboring face are used to form the event-affected vertex set. At the same time, the set of interactive anchor points is converted into a list of anchor point vertex numbers, and the set of update restricted areas is converted into a list of restricted area vertex numbers. The event-affected vertex set, the list of anchor point vertex numbers, and the list of restricted area vertex numbers are then output. S4-2. The surgical end generates freeze constraints based on the list of restricted area vertex numbers. The freeze constraint defines the displacement vector component of each vertex in the restricted area vertex number list as zero and writes the constraint into the freeze constraint entry of the event constraint set. The surgical end generates hold constraints based on the list of anchor point vertex numbers. The hold constraints define the difference vector between the projected pixel coordinates before and after the update of each anchor point vertex in the anchor point vertex number list as zero. The projected pixel coordinates before the update are obtained by transforming the three-dimensional coordinates of the anchor point vertex before the update to the observation coordinate system through the camera extrinsic parameters and projecting them onto the pixel plane through the camera intrinsic parameters. The projected pixel coordinates after the update are obtained by projecting the three-dimensional coordinates of the anchor point vertex after the update onto the pixel plane with the same extrinsic and intrinsic parameters. The camera extrinsic and intrinsic parameters are calibrated by the surgical end before the operation and written into the model state data structure and read in this step. The difference vector is zero by calculating the difference vector and determining that each component is equal to zero during the consistency check. At the same time, the hold constraint is written into the hold constraint entry of the event constraint set. The output is a basic constraint set containing the freeze constraint entry and the hold constraint entry. S4-3. The surgical end generates event-specific constraints based on the update trigger event type and merges and outputs the event constraint set. When the update trigger event is a traction event, the surgical end reads the traction principal direction from the local displacement vector field and normalizes it to obtain the traction direction vector. It also calculates the upper limit of the displacement magnitude of the local displacement vector field within the available region set. The upper limit of the displacement magnitude is obtained by calculating the magnitude of all displacement vectors within the available region set and taking the largest value. The surgical end defines a traction dot product constraint and a traction magnitude constraint for each vertex in the event-affected vertex set. The traction dot product constraint defines that the dot product of the vertex's displacement vector and the traction direction vector is non-negative. The traction magnitude constraint defines that the magnitude of the vertex's displacement vector does not exceed the upper limit of the displacement magnitude, and the traction dot product is approximately... The tension and traction mode length constraints are written into the tension constraint entry of the event constraint set. When the update trigger event is a contact event, the surgical end calculates the surface normal vector for each vertex in the event-affected vertex set and reads the corresponding displacement vector from the local displacement vector field. The projection value of the displacement vector on the surface normal vector direction is used as the normal component, and its sign is used as the normal component sign. The surface normal vector is obtained by weighted summation of the normals of the face to which the vertex belongs and then normalized. The surgical end defines the contact same-sign constraint as the normal component sign of the vertex displacement vector being the same as the normal component sign of the local displacement vector field, and writes the contact same-sign constraint into the contact constraint entry of the event constraint set. The surgical end merges the basic constraint set with the tension constraint entry or the contact constraint entry into the event constraint set and outputs it. Through the above implementation process, the surgical end transforms the updated restricted area set and interactive anchor point set into freeze constraints and projection-preserving constraints on vertex displacements, and transforms traction events or contact events into event-specific constraints on the set of vertices affected by the event. This ensures that the event constraint set has a clear target, clear computational terms, and clear value rules, thus allowing it to be directly used for cloud-based candidate update solutions and local consistency checks, avoiding unexecutable situations caused by unclear constraint targets or ambiguous value methods. In practical applications: when the surgical end identifies traction events and outputs the event-affected area during the mandibular angle exposure stage, this step maps the event-affected area to the mandible. The event-affected vertex set around the corner is processed. At the same time, the list of anchor point vertex numbers pre-selected by the doctor and the list of restricted area vertex numbers are read. The freeze constraint limits the displacement of the restricted area vertices to zero, the hold constraint limits the projection difference vector of the anchor point vertices to zero, and the pull constraint limits the dot product of the event-affected vertex displacement and the pull direction vector to non-negative and limits the displacement magnitude to within the upper limit of the displacement magnitude. Then the event constraint set is sent to the cloud to solve the candidate local update results and is used for consistency verification at the surgical end, thereby ensuring that the restricted area does not shift, the anchor points remain aligned, and the displacement direction and magnitude of the pull-affected area meet the geometric conditions of the pull event.

[0022] S5. The surgical end sends a candidate update request containing the current 3D model and event constraint set to the cloud, and receives the candidate local update result returned by the cloud. The candidate local update result is the set of displacements of vertices within the event-affected area. This implementation describes how, after the event constraint set has been generated, the surgical end organizes candidate update requests and sends them to the cloud; how the cloud solves for candidate local update results based on the current 3D model and event constraint set in the request; and how the surgical end receives the candidate local update results and establishes a one-to-one correspondence between them and the vertices within the event's influence region, thereby providing a directly applicable displacement set for subsequent consistency verification and admission determination. The overall principle of this process is that the surgical end only sends geometric data and constraint entries related to the event's influence region; the cloud solves for the displacement set of the vertices within the event's influence region under the constraint conditions and returns it; the surgical end writes the returned displacement set into a local result structure and outputs it based on the vertex number table. This implementation process includes the following steps: S5-1. The surgical end reads the current 3D model, the event constraint set, and the event-affected vertex set, and constructs a candidate update request. The candidate update request includes a vertex number list corresponding to the event-affected vertex set, a vertex 3D coordinate list corresponding to the vertex number list, a one-ring neighborhood patch topology list corresponding to the vertex number list, and a constraint entry list of the event constraint set. The constraint entry list is represented by the constraint type field and the action object number list field. The freeze constraint entry records the restricted area vertex number list and the constraint item with a displacement value of zero. The hold constraint entry records the anchor point vertex number list and the constraint item with a projection difference vector value of zero. The pull constraint entry records the event-affected vertex number list and the constraint item with a dot product that is non-negative and a displacement magnitude that does not exceed the upper limit of the displacement magnitude. The contact constraint entry records the event-affected vertex number list and the constraint item with the same sign as the normal component. The surgical end writes the candidate update request into the pending queue and outputs it. S5-2. The surgical end retrieves candidate update requests from the queue to be sent and sends them to the cloud. After receiving the candidate update requests, the cloud reconstructs the local mesh of the event-affected vertex set in its memory and constructs a solution problem with the displacement vector of the event-affected vertex set as the variable to be solved. The cloud converts the freeze constraint entries into the equality condition that the corresponding vertex displacement vector component is equal to zero, converts the hold constraint entries into the equality condition that the projection difference vector before and after the corresponding anchor vertex update is equal to zero, and converts the pull constraint entries into the inequality condition that the dot product of the corresponding vertex displacement vector and the pull direction vector is non-negative and the displacement magnitude does not exceed the upper limit of the displacement magnitude. Alternatively, the contact constraint entries are converted into the condition that the sign of the normal component of the corresponding vertex displacement vector is the same as the sign of the normal component of the local displacement vector field. Under the condition of satisfying the equality and inequality conditions, the cloud solves the displacement vector of each vertex in the event-affected vertex set and forms the candidate local update result, and outputs the candidate local update result. S5-3. The cloud encapsulates the candidate local update results into a list of displacement vectors in the order of the vertex number list and returns it to the surgical end. After receiving the list of displacement vectors, the surgical end establishes a one-to-one mapping relationship between vertex numbers and displacement vectors according to the vertex number list and generates a candidate local update result structure. The candidate local update result structure contains a list of vertex numbers and a corresponding list of displacement vectors and is associated with the set of vertices affected by the event. The surgical end outputs the candidate local update results. Through the above implementation process, the surgical end limits the candidate update requests to local mesh data and event constraint set entries in the event-affected area. This allows the cloud to solve for and return the displacement set of vertices within the event-affected area under constraints. The surgical end establishes a correspondence between the displacement set and the vertex set based on the vertex number order, enabling subsequent consistency checks to substitute frozen constraints, hold constraints, and event-specific constraints one by one to complete the admission determination. In practical applications: after the surgical end identifies the traction event in the mandibular angle region and generates the event constraint set, it sends the vertex number list, vertex 3D coordinate list, and local patch topology list of the event-affected vertex set around the mandibular angle, along with the frozen constraint entries, hold constraint entries, and traction constraint entries, to the cloud. The cloud solves for the displacement vector of each event-affected vertex based on the above entries and returns the displacement vector list in vertex number order. The surgical end writes the displacement vector list into the candidate local update result structure and outputs it to the consistency check step to determine whether to allow the displacement set to be applied to the current 3D model.

[0023] S6. The surgical end substitutes each candidate local update result into the event constraint set to perform consistency verification to obtain the admission judgment result. The consistency verification includes at least verifying that the vertex displacement in the update forbidden zone set is zero, verifying that the projection of the interactive anchor point set satisfies the observation correspondence, and verifying that the displacement in the event-affected area satisfies the corresponding event constraint. When all consistency verifications are successful, the candidate local update results are applied to the current 3D model to generate the updated current 3D model. Otherwise, the current 3D model remains unchanged and a rejection item flag is generated. This implementation describes how, after receiving candidate local update results from the cloud, the surgical end performs consistency checks by substituting each candidate local update result into the event constraint set, and generates an admission decision result accordingly to determine whether to apply the candidate local update results to the current 3D model. The overall principle of this process is to uniformly treat the event constraint set as computable verification items for vertex displacement, anchor point projection, and event-specific constraints. The surgical end checks each candidate local update result in groups according to constraint type, forming a sequence of verification results. Then, based on the sequence of verification results, it generates an admission decision result and rejection item identifiers, and completes model updates and data structure write-back upon successful admission. This implementation process includes the following steps: S6-1. The surgical end reads the candidate local update results, event constraint set, current 3D model, update forbidden zone set, interaction anchor point set, observation correspondence and event-affected vertex set, and establishes a one-to-one mapping table from vertex number to displacement vector. The mapping table is generated by pairing the vertex number list and displacement vector list in the candidate local update results with the same sequence number. At the same time, the verification result table is initialized and the frozen constraints, hold constraints and event-specific constraints are written into the list of items to be verified respectively. The mapping table and the verification result table are output. S6-2. The surgical end performs consistency verification on the frozen constraints. Specifically, it reads the displacement vectors in the mapping table one by one from the list of restricted vertex numbers corresponding to the updated restricted set and calculates the three-dimensional component values ​​of the displacement vectors. It determines whether all three-dimensional components of the displacement vector are equal to zero and writes the determination result into the frozen verification field of the verification result table. When there is any restricted vertex whose displacement vector three-dimensional component is not equal to zero, it writes a rejection item mark as restricted area failure and writes a frozen verification failure mark. It outputs the frozen verification field and the restricted area failure rejection item mark or the frozen verification pass mark. S6-3. The surgical end performs consistency verification on the constraint. Specifically, it reads the 3D coordinates of the anchor vertex in the current 3D model before the update and reads the displacement vector of the anchor vertex from the mapping table to calculate the updated 3D coordinates. The 3D coordinates before and after the update are transformed to the observation coordinate system through the camera extrinsic parameters and projected to the pixel plane through the camera intrinsic parameters to obtain the pixel coordinates before and after the update. Then, the target pixel coordinates corresponding to the anchor vertex are read from the observation correspondence and the difference vector between the pixel coordinates before and after the update and the difference vector between the pixel coordinates after and the target pixel coordinates are calculated respectively. It is determined whether the two components of the difference vector between the pixel coordinates after the update and the target pixel coordinates are both equal to zero and the determination result is written to the anchor verification field of the verification result table. When there is any anchor point whose difference vector component is not equal to zero, the rejection item is written as the anchor point failure and the anchor point verification failure mark is written. The anchor point verification field and the anchor point failure rejection item mark or the anchor point verification pass mark are output. S6-4. The surgical end performs consistency checks on event-specific constraints. Specifically, it reads the displacement vector in the mapping table for each vertex number in the set of vertexes affected by the event and performs corresponding checks according to the type of update triggering event. When the update triggering event is a traction event, it reads the traction main direction vector and calculates the dot product of the displacement vector and the traction main direction vector and determines whether the dot product is non-negative. At the same time, it calculates the magnitude of the displacement vector and determines whether the magnitude does not exceed the upper limit of the displacement magnitude and writes the determination result into the event verification field of the verification result table. When the update triggering event is a contact event, it calculates the surface normal vector of the vertex and calculates the projection value of the displacement vector in the direction of the surface normal vector as the normal component and takes its sign. Then, it reads the sign of the normal component at the corresponding position from the local displacement vector field and determines whether the two signs are the same and writes the determination result into the event verification field of the verification result table. When any vertex affected by the event does not meet the corresponding event-specific constraint, it writes a rejection item mark as an event failure and writes an event verification failure mark. It outputs the event verification field and the event failure rejection item mark or the event verification pass mark. S6-5. The surgical end generates an admission judgment result based on the verification result table and performs an update action. When the freeze verification field, anchor point verification field, and event verification field are all written with the pass flag, the admission judgment result is generated as pass, and the candidate local update result is applied to the current 3D model. The application method is to read the 3D coordinates before the update from the current 3D model for each vertex number in the candidate local update result, add them to the vertex displacement vector to obtain the updated 3D coordinates, and write them back to the coordinate field of the vertex to generate the updated current 3D model. When any verification field is written with the failure flag, the admission judgment result is generated as rejection, the current 3D model remains unchanged, and the rejection item identifier is output. At the same time, the admission judgment result and the rejection item identifier are written to the verification record entry of this update and output. Through the above implementation process, the surgical end substitutes the candidate local update results into the event constraint set using a mapping method from vertex number to displacement vector. It then performs computable consistency checks on the update forbidden zone set, the interaction anchor point set, and the event-affected vertex set. If the check passes, the coordinates of the current 3D model are updated; if the check fails, the current 3D model remains unchanged, and a rejection flag is output. This ensures that the intraoperative dynamic adaptation has a reproducible admission decision process and avoids the application of candidate updates that do not meet the constraints, thus preventing the spread of bias. In practical application: after the cloud returns the candidate local update results for the mandibular angle region, the surgical end first checks the corresponding region of the inferior alveolar nerve projection. The system checks each vertex in the restricted area list to see if its displacement vector component is zero. Then, it calculates the updated projected pixel coordinates for each of the pre-selected mandibular angle anchor point vertices and compares them with the target pixel coordinates in the observation correspondence. Subsequently, it calculates the dot product of the displacement vector and the main traction direction vector for each vertex set affected by the mandibular angle event and calculates whether the displacement modulus satisfies the constraint. When all three types of checks pass, the displacement vectors are added to the corresponding vertex coordinates to generate the updated current 3D model and enter the next round of observation acquisition and event recognition process. When any check fails, the corresponding rejection item flag is output and the current 3D model remains unchanged to wait for subsequent frames to be solved again.

[0024] S7. Write the update trigger event, event constraint set, candidate local update results and admission judgment results into the update record sequence in chronological order for retrospective display and rollback calculation. This implementation describes how, after each candidate local update result admission determination, the surgical end writes the update trigger event, event constraint set, candidate local update result, and admission determination result into the update record sequence, and performs backtracking display and rollback calculation based on the update record sequence when needed. The overall principle of this process is to use the frame sequence number as the time sequence carrier, encapsulate the event type, event influence area identifier, constraint entry, and displacement entry related to each update into update record entries according to a fixed field structure, and append them to the update record sequence. At the same time, during backtracking, events and constraints are replayed according to the update record entries, and during rollback, the opposite displacement is performed to restore the model coordinates according to the displacement entries of the update record entries. This implementation process includes the following steps: S7-1. After generating the admission determination result, the surgical end reads the update trigger event, the event affected area, the event constraint set, the candidate local update result and the admission determination result, and reads the current frame number as the time sequence field. It generates update record entries and writes them into the event type field, the event affected area identifier field, the constraint version field, the vertex number list field, the displacement vector list field and the determination result field. The vertex number list field and the displacement vector list field are taken from the candidate local update result. The update record entries are then output. S7-2. The surgical end generates a constraint summary field for the event constraint set and writes it into the update record entry. The constraint summary field contains a list of restricted area vertex numbers for frozen constraint entries, a list of anchor point vertex numbers for preserved constraint entries, and a list of event-affected vertex numbers for event-specific constraint entries, as well as their constraint type fields. The surgical end calculates a record check value field for the update record entry and writes it into the update record entry. The record check value field is obtained by concatenating the event type field, event-affected area identifier field, vertex number list field, displacement vector list field, and judgment result field in a predetermined order and then calculating the hash value. The update record entry containing the record check value field is output. S7-3. The surgical end appends the update record entries to the update record sequence in ascending order of frame number. When receiving the rollback instruction, it reads the target update record entry according to the frame number and outputs its event type field, event influence area identifier field and constraint summary field for rollback display. When receiving the rollback instruction, it reads the vertex number list field and displacement vector list field of the target update record entry according to the frame number and performs a negative operation on each displacement vector in the displacement vector list field to generate a rollback displacement vector list. Then, it adds the rollback displacement vector list to the corresponding vertex coordinates of the current 3D model one by one according to the vertex number list field to output the current 3D model after rollback. Through the above implementation process, the surgical end establishes a time sequence based on frame numbers and encapsulates the events, constraints, displacements, and judgment results related to each update into structured update record entries, which are then appended to the update record sequence. This allows for retrospective display based on update record entries and rollback calculations performed by negating the displacements, thus enabling reproducible traceability and recovery when inconsistencies arise during surgery or updates need to be revoked. In practical applications: when the surgical end completes a candidate update driven by a traction event in the mandibular angle region and obtains an admission judgment result of "pass," the surgical end writes the frame number, traction event type, event-affected area identifier, frozen constraints and retained constraints summary, as well as the vertex number list and displacement vector list of the candidate local update results into update record entries and appends them to the update record sequence. When the doctor triggers a retrospective command on the display interface, the system reads the corresponding record according to the frame number and displays the event type and constraint summary of this update. When the doctor triggers a rollback command on the display interface, the system reads the same record, negates the displacement vectors, and adds them one by one to the corresponding vertex coordinates to restore the current 3D model before this update.

[0025] Furthermore, it also includes: a 3D modeling and dynamic adaptation operating system for facial contouring, including a modeling and annotation module, a tracking and partitioning module, a solution and recognition module, a constraint generation module, a cloud-based solution module, a verification and admission module, and a record rollback module. The modeling and annotation module acquires the patient's preoperative three-dimensional basic model at the surgical end and generates the current three-dimensional model discretely. It receives the set of interactive anchor points and the set of updated restricted areas input by the doctor on the current three-dimensional model and writes them into the model state data structure. The tracking partition module collects continuous intraoperative observation data at the surgical end, performs feature tracking on intraoperative observation data at adjacent time points to obtain the observation correspondence, and generates a set of usable regions based on the proportion of continuously trackable feature points in each spatial region, and generates event candidate regions accordingly. The solution and recognition module calculates the local displacement vector field based on the observation correspondence within the event candidate area, and determines and updates the triggering event and its affected area accordingly. The traction event is determined by keeping the main direction of the local displacement vector field consistent during continuous acquisition, the contact event is determined by keeping the occlusion boundary formed within the event candidate area consistent during continuous acquisition, and the doctor confirmation event is determined by the doctor's confirmation input of the contour point or symmetry line. The constraint generation module is used to generate an event constraint set for each update trigger event based on the event influence area, the set of interaction anchor points, and the set of update restricted areas. The event constraint set includes the freeze constraint that the vertex displacement in the update restricted area set is zero, the preservation constraint that the interaction anchor point set satisfies the observation correspondence when projected onto the intraoperative observation data before and after the update, and the constraint that the dot product of the displacement vector and the main traction direction in the event influence area is non-negative and the displacement magnitude is not greater than the maximum local displacement magnitude in the available area set under the traction event, or the constraint that the normal component of the displacement vector in the event influence area has the same sign as the normal component of the local displacement vector field under the contact event. The cloud-based solution module is used by the surgical end to send a candidate update request containing the current 3D model and the set of event constraints to the cloud, and to receive the candidate local update results returned by the cloud. The candidate local update results are the set of displacements of vertices within the event's influence area. The admission verification module is used by the surgical end to substitute each candidate local update result into the event constraint set to perform consistency verification in order to obtain the admission judgment result. The consistency verification includes at least verifying that the vertex displacement in the update forbidden zone set is zero, verifying that the projection of the interactive anchor point set satisfies the observation correspondence, and verifying that the displacement in the event-affected area satisfies the corresponding event constraint. When all consistency verifications are successful, the candidate local update result is applied to the current 3D model to generate the updated current 3D model; otherwise, the current 3D model remains unchanged and a rejection item flag is generated. The rollback module is used to write the update trigger event, event constraint set, candidate local update results and admission judgment results into the update record sequence in chronological order for retrospective display and rollback calculation.

[0026] Working principle: At the surgical end, the preoperative 3D basic model is first processed into the current 3D model, and the doctor's pre-selected interaction anchor points and update restricted areas are written into the model state. During the operation, observation data is continuously collected, and feature tracking is performed between adjacent frames to obtain the observation correspondence. Then, the proportion of continuously trackable feature points is statistically analyzed according to spatial regions to generate a set of usable regions and obtain event candidate regions accordingly. The system only calculates the local displacement vector field within the event candidate regions and identifies and updates the triggering events and the affected regions. Traction events are determined by the continuity and consistency of the principal displacement direction, and contact events are determined by the continuity and consistency of the occlusion boundary. The doctor confirms the event... The event is determined by the doctor's confirmation input of the contour points or symmetry lines; then, based on the event's affected area, three types of constraints are superimposed to generate an event constraint set, including the freeze constraint where the displacement of the restricted area vertex is zero, the maintenance constraint where the anchor point projection satisfies the observation correspondence, and the event-specific constraint corresponding to traction or contact; the surgical end sends the current 3D model and the event constraint set to the cloud to solve for the candidate local update displacement set. The surgical end substitutes each constraint into the consistency check. If it passes, the update is applied; otherwise, it is rejected and a rejection item is given. At the same time, the event, constraint, candidate displacement and judgment result are written into the update record sequence in order for backtracking and rollback. In scenarios like jaw angle contouring, the traction device pulls on soft tissue, and instrument occlusion and reflections cause some areas to stabilize and others to break. The system first classifies stable areas into usable areas and areas close to usable areas but with breaks into event candidate areas. Then, it calculates the displacement vector field in the event candidate areas: if the main displacement direction remains consistent across consecutive frames, it is judged as a traction event and the affected area is determined; if the occlusion boundary remains consistent across consecutive frames, it is judged as a contact event and the affected area is determined. If the doctor confirms a symmetry line or a contour point on the interface, the system generates the affected area of ​​the doctor-confirmed event based on the confirmed position. Then, the system freezes the restricted area, locks the anchor point projection, and superimposes traction or contact constraints. These constraints are sent to the cloud to solve for candidate displacements. The surgical end only updates the model when the restricted area has not shifted, the anchor point projection still corresponds to the observation, and the event area satisfies the event-specific constraints; otherwise, it directly rejects the model and retains the current state. At the same time, it can be recorded to review or roll back and restore the model.

[0027] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for three-dimensional modeling and dynamic adaptation of facial contouring, characterized in that, include: S1. Obtain the patient's preoperative three-dimensional basic model at the surgical end and generate the current three-dimensional model discretely. Receive the set of interactive anchor points and the set of updated restricted areas input by the doctor on the current three-dimensional model and write them into the model state data structure. S2. Collect continuous intraoperative observation data at the surgical end, perform feature tracking on intraoperative observation data at adjacent time points to obtain the observation correspondence, and generate a set of usable regions based on the proportion of continuously trackable feature points in each spatial region, and generate event candidate regions accordingly. S3. Calculate the local displacement vector field based on the observation correspondence within the event candidate area, and determine the updated triggering event and its affected area accordingly. The traction event is determined by keeping the main direction of the local displacement vector field consistent during continuous acquisition. The contact event is determined by keeping the occlusion boundary formed within the event candidate area consistent during continuous acquisition. The doctor confirmation event is determined by the doctor's confirmation input of the contour point or symmetry line. S4. For each update trigger event, generate an event constraint set based on the event influence area, the set of interaction anchor points, and the update restricted area set. The event constraint set includes the freeze constraint that the vertex displacement in the update restricted area set is zero, the maintenance constraint that the interaction anchor point set satisfies the observation correspondence when projected onto the intraoperative observation data before and after the update, and the constraint that the dot product of the displacement vector and the main traction direction in the event influence area is non-negative and the displacement magnitude is not greater than the maximum local displacement magnitude in the available area set under the traction event, or the constraint that the normal component of the displacement vector in the event influence area has the same sign as the normal component of the local displacement vector field under the contact event. S5. The surgical end sends a candidate update request containing the current 3D model and event constraint set to the cloud, and receives the candidate local update result returned by the cloud. The candidate local update result is the set of displacements of vertices within the event-affected area. S6. The surgical end substitutes each candidate local update result into the event constraint set to perform consistency verification to obtain the admission judgment result. The consistency verification includes at least verifying that the vertex displacement in the update forbidden zone set is zero, verifying that the projection of the interactive anchor point set satisfies the observation correspondence, and verifying that the displacement in the event-affected area satisfies the corresponding event constraint. When all consistency verifications are successful, the candidate local update results are applied to the current 3D model to generate the updated current 3D model. Otherwise, the current 3D model remains unchanged and a rejection item flag is generated. S7. Write the update trigger event, event constraint set, candidate local update results and admission judgment results into the update record sequence in chronological order for retrospective display and rollback calculation.

2. The three-dimensional modeling and dynamic adaptation method for facial contouring according to claim 1, characterized in that: S1 includes: S1-1. Receive the preoperative three-dimensional basic model data of the face at the surgical end, perform connected component segmentation on the preoperative three-dimensional basic model data of the face to obtain the maximum connected component, and perform hole detection and hole filling on the maximum connected component to form a closed mesh, and output the closed mesh. S1-2. Calculate the face normal for each facet on the closed mesh and calculate the vertex curvature field for each vertex based on the face normal of its one-ring neighborhood. Perform adaptive resampling on the closed mesh according to the vertex curvature field to generate the current 3D model containing the vertex set and facet topology, and output the current 3D model. S1-3. Receive the set of interactive anchor points and the set of updated restricted areas input by the doctor on the current 3D model. Map the set of interactive anchor points to a set of vertex identifiers and the set of updated restricted areas to a set of face identifiers. Write the current 3D model, the set of vertex identifiers, and the set of face identifiers into the model state data structure and output the model state data structure.

3. The three-dimensional modeling and dynamic adaptation method for facial contouring according to claim 2, characterized in that: S2 includes: Feature point extraction is performed on two adjacent frames of intraoperative observation data at the surgical end, and window matching is performed on each feature point in the next frame of intraoperative observation data to form a feature matching pair. An observation correspondence is generated from the feature matching pair from the coordinates of the feature point in the previous frame to the coordinates of the matching point in the next frame, and the observation correspondence is output. S2-2. Divide the intraoperative observation data into multiple spatial regions according to a preset grid, and in each spatial region, calculate the ratio of the number of feature matching pairs falling into the spatial region to the total number of feature points in the spatial region to generate the sustainable tracking ratio field of the spatial region, and output the sustainable tracking ratio field. S2-3. Write the spatial regions with a sustainable tracking ratio field greater than zero and which generate observation correspondence in multiple consecutive pairs of adjacent frames into the available region set, and write the spatial regions that are adjacent to any available region set spatial regions, contain feature points in the current frame, and do not form feature matching pairs in the observation correspondence into the event candidate region, and output the available region set and the event candidate region.

4. The three-dimensional modeling and dynamic adaptation method for facial contouring according to claim 3, characterized in that: S3 includes: S3-1. At the surgical end, for each spatial region within the event candidate region, calculate the set of local displacement vectors of the spatial region based on the observation correspondence and form a local displacement vector field. Calculate the main direction field and direction dispersion field of the local displacement vector field and write them into the displacement evidence record item. Output the displacement evidence record item. S3-2. At the surgical end, based on the displacement evidence record item, execute the direction consistency checker on the main direction field in multiple consecutive pairs of adjacent frames and execute the dispersion consistency checker based on the direction dispersion field to generate the traction confidence field and traction uncertainty field and write the traction token or pending token, and output the traction token or pending token and the corresponding traction confidence field and traction uncertainty field. S3-3. At the surgical end, for the same spatial region, calculate the occlusion boundary field based on the distribution of feature points in the event candidate region that have not formed an observation correspondence and write it into the occlusion evidence record item. Based on the occlusion evidence record item, execute the boundary consistency checker on the occlusion boundary field in multiple consecutive pairs of adjacent frames to generate the occlusion stability field and write it into the contact token or pending token. Output the contact token or pending token and the corresponding occlusion stability field.

5. The three-dimensional modeling and dynamic adaptation method for facial contouring according to claim 4, characterized in that: S3 also includes: S3-4. Receive the doctor's confirmation input for the contour point or symmetry line at the surgical end and parse the confirmation input into the confirmation type field and confirmation location field to write to the confirmation evidence record item. At the same time, write the confirmation token and associate the confirmation evidence record item with the corresponding spatial area, and output the confirmation token and confirmation evidence record item. S3-5. At the surgical end, execute conflict resolution rules on the traction token, contact token, and confirmation token in the same spatial area to generate a unique event type field and update the event confidence field. The conflict resolution rules include generating a doctor confirmation event and writing an event lock status lock when the confirmation token exists; generating a contact event and writing an event lock status lock when the confirmation token does not exist but the contact token exists; generating a traction event and writing an event lock status lock when neither the confirmation token nor the contact token exists but the traction token exists; and writing a backtracking retest status lock when the pending token exists and reserving the spatial area as an event candidate area. Output the unique event type field, event confidence field, event lock status lock, or backtracking retest status lock. S3-6. At the surgical end, generate the event-affected area based on the unique event type field and the spatial region identifier field, and write the event-affected area, the event confidence field, the displacement evidence record item, the occlusion evidence record item or the confirmation evidence record item into the event record sequence, and output the update trigger event and its event-affected area.

6. The three-dimensional modeling and dynamic adaptation method for facial contouring according to claim 5, characterized in that: S4 includes: S4-1. On the surgical end, for each update trigger event, read the event-affected region, the set of interactive anchor points, the set of update restricted areas, the observation correspondence, and the local displacement vector field. Map the event-affected region to the set of event-affected vertices, map the set of interactive anchor points to the set of anchor point vertices, and map the set of update restricted areas to the set of restricted area vertices. At the same time, construct constraint generation record items and write them into the event type field and the region identifier field. Output the set of event-affected vertices, the set of anchor point vertices, the set of restricted area vertices, and the constraint generation record items. S4-2. At the surgical end, using the set of restricted area vertices as the judgment object and the frozen consistency checker as the judgment basis, write a zero displacement freeze constraint to each vertex in the restricted area vertex set and write the freeze constraint identifier field. At the same time, using the set of anchor point vertices as the judgment object and the projection consistency checker as the judgment basis, write a projection preservation constraint to each anchor point in the anchor point vertex set so that the difference between the pixel coordinates of the anchor point vertex projected to the intraoperative observation data before and after the update is zero and write the preservation constraint identifier field. Output the basic constraint set and the basic constraint gating token. S4-3. At the surgical end, the event-affected vertex set is used as the judgment object and the event type field is used as the branch condition to generate event-specific constraints. When the event type field is a traction event, the upper limit of the displacement magnitude is calculated in the available area set based on the local displacement vector field, and the traction main direction field is calculated. The direction consistency checker is used as the judgment basis. A dot product constraint is written to each vertex in the event-affected vertex set so that the dot product of the vertex displacement vector and the traction main direction field is non-negative and the vertex displacement magnitude does not exceed the upper limit of the displacement magnitude. The traction constraint token and traction constraint identifier fields are written. When the event type field is a contact event, the surface normal field is solved based on the event-affected area. The normal consistency checker is used as the judgment basis. A normal sign constraint is written to each vertex in the event-affected vertex set so that the normal component of the vertex displacement vector has the same sign as the normal component of the local displacement vector field. The contact constraint token and contact constraint identifier fields are written. The event-specific constraints and corresponding tokens are output. S4-4. At the surgical end, execute the token merging rule on the basic constraint gating token and the traction constraint token or contact constraint token to generate an event constraint set and write it into the constraint version field and the constraint entry count field. The token merging rule includes writing a state lock for rollback and re-checking when any consistency checker output fails and writing the event-affected vertex set into the list of regions to be rebuilt. When all consistency checkers output pass, write a state lock for constraint locking and associate the event constraint set with the constraint generation record item into the constraint record sequence. Output the event constraint set, state lock and written item.

7. The three-dimensional modeling and dynamic adaptation method for facial contouring according to claim 6, characterized in that: S5 includes: S5-1. The surgical end constructs candidate update request objects based on the current 3D model, available area set, event-affected area and event constraint set, and generates a request summary field. The request summary field includes an event type field, an event-affected area identifier field, a constraint entry count field and an integrity verification value field calculated based on the request object. The request summary field is written into the request record item to generate an external token, and the external token and request object are output. S5-2. The surgical end sends the outgoing token and the request object to the cloud and receives the candidate partial update result and response summary field returned by the cloud. The response summary field includes at least the cloud version identifier field, the solution convergence status field and the result range identifier field. The surgical end performs a consistency checker on the response summary field and the outgoing token to generate a receive token or a rollback re-examination state lock. The consistency checker performs a matching check on the integrity check value field and a coverage consistency check on the result range identifier field and the event impact area identifier field. The candidate partial update result and the receive token or rollback re-examination state lock are output. S5-3. When receiving token generation, the surgical end executes a multi-objective quality scoring function on the candidate local update results to generate a quality scoring field and update the uncertainty field. The multi-objective quality scoring function is calculated based at least on the displacement energy field of the candidate local update results within the update restricted area set, the projection residual field at the interaction anchor point set, and the displacement continuity field at the boundary of the event-affected area. When the quality scoring field does not meet the locally stored threshold packet, a rejection flag is written and the rollback re-examination state lock is set to trigger a downgrade path or retransmission path. When the quality scoring field meets the threshold packet, an acceptance flag is written and the candidate local update results are output.

8. The three-dimensional modeling and dynamic adaptation method for facial contouring according to claim 7, characterized in that: S6 includes: S6-1. Receive candidate local update results and event constraint set at the surgical end, construct verification task record item and write event identifier field, region identifier field, constraint version field and displacement entry count field. At the same time, group the candidate local update results according to the restricted area vertex set, anchor point vertex set and event-affected vertex set to form restricted area displacement field, anchor point displacement field and region displacement field. Output verification task record item and restricted area displacement field, anchor point displacement field and region displacement field. S6-2. Using the restricted area displacement field as the judgment object and the frozen consistency checker as the judgment basis, calculate the magnitude of the displacement vector of each vertex in the restricted area displacement field one by one, write a pass flag for entries with a magnitude equal to zero, write a failure flag for entries with a magnitude not equal to zero, and write a rejection item flag as restricted area failure. At the same time, write the gate token as restricted area token pass or restricted area token failure, and output the gate token and rejection item flag written item. S6-3. Using the anchor point displacement field and the correspondence with the observation as the judgment object and the projection consistency checker as the judgment basis, each anchor point vertex is projected into the intraoperative observation data coordinate system in the updated three-dimensional coordinates to obtain the updated pixel coordinates. The difference between the updated pixel coordinates and the target pixel coordinates given by the observation correspondence is calculated to obtain the anchor point residual vector. For entries where the anchor point residual vector is equal to zero, a pass flag is written. For entries where the anchor point residual vector is not equal to zero, a failure flag is written and a rejection flag is written as the anchor point failure. At the same time, a gating token is written as either the anchor point token pass or the anchor point token failure. The gating token and rejection flag are output as the entries written. S6-4. Using the region displacement field and the event constraint set as the judgment objects and the constraint consistency checker as the judgment basis, substitute each vertex displacement vector in the region displacement field into the freeze constraint, hold constraint and event-specific constraint in the event constraint set to obtain the constraint satisfaction mark sequence. Write a failure flag and a rejection item flag to the entries in the constraint satisfaction mark sequence that do not satisfy the flag, and write a gate token to the region failure flag. Write a pass flag and a gate token to the entries in the constraint satisfaction mark sequence that are all satisfied, and output the gate token and rejection item flag entries. S6-5. At the surgical end, execute the token merging rule on the restricted area token, anchor point token, and region token to generate the admission judgment result and state lock. The token merging rule includes writing a state lock to roll back and re-examine when any token fails, keeping the current 3D model unchanged, and outputting the rejection item flag. When the restricted area token, anchor point token, and region token all pass, write a state lock to update and lock, apply the candidate local update result to the current 3D model to generate the updated current 3D model, and output the admission judgment result as pass. Write the admission judgment result, state lock, and rejection item flag into the verification task record item, and output the admission judgment result and the updated current 3D model or rejection item flag.

9. The three-dimensional modeling and dynamic adaptation method for facial contouring according to claim 8, characterized in that: S7 includes: S7-1. Receive update trigger event, event constraint set, candidate local update result and admission judgment result at the surgical end, generate update record entry and write it into the event type field, event influence area identifier field, constraint version field, displacement entry count field and judgment result field, and output update record entry; S7-2. Calculate the displacement summary field for the candidate local update results and the constraint summary field for the event constraint set. Write the displacement summary field and the constraint summary field into the update record entry and generate the record integrity check value field. Output the update record entry containing the record integrity check value field. S7-3. Append the update record entries to the update record sequence in the order of generation, and when the rollback instruction is received, read the candidate local update results of the target update record entries and perform sign inversion on the candidate local update results to generate a rollback displacement set. Apply the rollback displacement set to the current 3D model to output the current 3D model after rollback.

10. A 3D modeling and dynamic adaptation operating system for facial contouring, including a modeling and annotation module, a tracking and partitioning module, a solution and recognition module, a constraint generation module, a cloud-based solution module, a verification and admission module, and a record rollback module, characterized in that: The modeling and annotation module acquires the patient's preoperative three-dimensional basic model at the surgical end and generates the current three-dimensional model discretely. It receives the set of interactive anchor points and the set of updated restricted areas input by the doctor on the current three-dimensional model and writes them into the model state data structure. The tracking partition module collects continuous intraoperative observation data at the surgical end, performs feature tracking on intraoperative observation data at adjacent time points to obtain the observation correspondence, and generates a set of usable regions based on the proportion of continuously trackable feature points in each spatial region, and generates event candidate regions accordingly. The solution and recognition module calculates the local displacement vector field based on the observation correspondence within the event candidate area, and determines and updates the triggering event and its affected area accordingly. The traction event is determined by keeping the main direction of the local displacement vector field consistent during continuous acquisition, the contact event is determined by keeping the occlusion boundary formed within the event candidate area consistent during continuous acquisition, and the doctor confirmation event is determined by the doctor's confirmation input of the contour point or symmetry line. The constraint generation module is used to generate an event constraint set for each update trigger event based on the event influence area, the set of interaction anchor points, and the set of update restricted areas. The event constraint set includes the freeze constraint that the vertex displacement in the update restricted area set is zero, the preservation constraint that the interaction anchor point set satisfies the observation correspondence when projected onto the intraoperative observation data before and after the update, and the constraint that the dot product of the displacement vector and the main traction direction in the event influence area is non-negative and the displacement magnitude is not greater than the maximum local displacement magnitude in the available area set under the traction event, or the constraint that the normal component of the displacement vector in the event influence area has the same sign as the normal component of the local displacement vector field under the contact event. The cloud-based solution module is used by the surgical end to send a candidate update request containing the current 3D model and the set of event constraints to the cloud, and to receive the candidate local update results returned by the cloud. The candidate local update results are the set of displacements of vertices within the event's influence area. The admission verification module is used by the surgical end to substitute each candidate local update result into the event constraint set to perform consistency verification in order to obtain the admission judgment result. The consistency verification includes at least verifying that the vertex displacement in the update forbidden zone set is zero, verifying that the projection of the interactive anchor point set satisfies the observation correspondence, and verifying that the displacement in the event-affected area satisfies the corresponding event constraint. When all consistency verifications are successful, the candidate local update result is applied to the current 3D model to generate the updated current 3D model; otherwise, the current 3D model remains unchanged and a rejection item flag is generated. The rollback module is used to write the update trigger event, event constraint set, candidate local update results and admission judgment results into the update record sequence in chronological order for retrospective display and rollback calculation.

Images