Patella injury first-aid robot protector fitting area segmentation method

Abstract

The present application relates to the technical field of medical robots, and particularly relates to a patella injury first-aid robot protector fitting area segmentation method, comprising the following steps: acquiring a multi-pose knee three-dimensional point cloud, selecting a reference pose to construct a surface model, combining patella injury information to divide a pose function area and label the point cloud, establishing surface mapping of non-reference poses and the reference model, fusing regional labels according to protection priority, generating a target knee segmentation model and outputting the position and label of each surface point. The present application makes the avoidance attribute consistent across poses through multi-pose regional label mapping and protection priority fusion, and combines geodesic distance, curvature and normal protrusion amount for fine zoning, reduces label mismatch and boundary mutation, simultaneously establishes individual muscle activation and normal and tangential deformation relationship, implements zoning compensation and safety gap constraint, detects residual deformation and adapts to continuous morphological changes through limited update, and improves injury protection, fitting adaptability and safety in dynamic use of the protector.

Description

Technical Field

[0001] This invention relates to the field of medical robot technology, specifically to a method for segmenting the fitting area of ​​a patellar injury emergency robot protective gear. Background Technology

[0002] Medical robotics technology is a technological field formed by the interdisciplinary integration of robotics, medical technology, computer technology, automatic control technology, sensing technology, and artificial intelligence technology. It primarily uses mechanical actuators, sensing systems, control systems, and human-computer interaction systems to assist in the operation of medical devices or patients. Applications include surgical assistance, rehabilitation training, medical image processing, and the operation of nursing equipment and medical devices. In medical devices involving high-precision positioning, posture tracking, and repetitive adjustments, medical robots typically combine image information, position and posture information, and mechanical or physiological signals to calculate the target position of the actuator, thereby completing the positioning, adjustment, or assisted operation of the medical device.

[0003] In emergency fixation of patellar injuries, braces need to provide a stable fit in the non-injured areas of the knee while avoiding the injured patella and surrounding areas that may be subjected to pressure. Due to changes in knee joint posture and muscle activation caused by pain and stress, the soft tissues near the patella will experience normal bulging and tangential migration, altering the surface morphology of the knee. Braces models built based on a single posture or moment are prone to localized compression, unsupported areas, or insufficient fit during subsequent use. Existing three-dimensional brace shaping technology typically sets the force fulcrum based on a single surface model, while leg brace control technology mainly adjusts the overall height and opening size. This makes it difficult to integrate the injury protection range under different postures, and also difficult to achieve continuous local boundary delineation and distinguish between muscle activation deformation and continuous morphological changes. Therefore, existing technologies cannot meet the comprehensive requirements of multi-posture adaptability, localized fit, and safe avoidance in emergency fixation of patellar injuries. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for segmenting the fitting area of ​​a robotic protective gear for patellar injury emergency treatment.

[0005] To achieve the above objectives, the present invention provides the following technical solution: This invention provides a method for segmenting the fitting area of ​​a robotic patellar injury emergency brace, comprising the following steps: S1. Obtain three-dimensional point cloud data of the patient's knee in multiple different knee joint postures, select one of the multiple different knee joint postures as the reference posture, take the other knee joint postures as non-reference postures, and then construct a reference knee three-dimensional surface model based on the three-dimensional point cloud data of the knee in the reference posture. S2. Obtain patellar injury information, delineate the injury protection area for each knee joint posture, and then obtain the surface spatial relationship parameters and local surface morphology parameters of each three-dimensional point in the knee three-dimensional point cloud data of each knee joint posture and the injury protection area. Based on the surface spatial relationship parameters and local surface morphology parameters, classify each three-dimensional point into regions to obtain the functional partitions of each knee joint posture and the region labels of each three-dimensional point. The functional partitions include the fitting and fixing area, the transition adjustment area, and the avoidance protection area. S3. Establish the surface mapping relationship between the three-dimensional point cloud data of the knee in each non-reference posture and the reference three-dimensional surface model of the knee. Based on the surface mapping relationship, map the region labels of each non-reference posture to the reference three-dimensional surface model of the knee to obtain the region labels of different knee joint postures. S4. Then set the protection priorities of the avoidance protection zone, transition adjustment zone and fitting fixation zone. According to the reference posture area label and the mapped non-reference posture area labels of the same surface point in the reference knee three-dimensional surface model, the area label with the highest protection priority is recorded as the fusion area label of the surface point to obtain the fusion functional partition. S5. Associate the fusion region label with each surface point in the reference knee 3D surface model to obtain the target knee 3D segmentation model, and output the 3D position and fusion region label of each surface point in the target knee 3D segmentation model.

[0006] Furthermore, the process of classifying each three-dimensional point into regions based on surface spatial relationship parameters and local surface morphology parameters includes: Obtain the geodesic distance from each three-dimensional point to the boundary of the damage protection area, and set a first distance threshold and a second distance threshold, wherein the second distance threshold is greater than the first distance threshold; The three-dimensional points within the damage protection area are divided into avoidance protection areas. For the three-dimensional points outside the damage protection area, the three-dimensional points with a geodesic distance not greater than the first distance threshold are divided into avoidance protection areas, the three-dimensional points with a geodesic distance greater than the first distance threshold but not greater than the second distance threshold are divided into transition adjustment areas, and the three-dimensional points with a geodesic distance greater than the second distance threshold are divided into fitting and fixing areas, thus obtaining the initial area label of each three-dimensional point. Obtain the local curvature and normal protrusion of each 3D point, set curvature threshold and normal protrusion threshold, and when the local curvature of the 3D point exceeds the curvature threshold and the normal protrusion exceeds the normal protrusion threshold, adjust the initial region label of the fixed area to the transition adjustment area or adjust the initial region label of the transition adjustment area to the avoidance protection area to obtain the region label of each 3D point.

[0007] Furthermore, establishing the surface mapping relationship between the knee 3D point cloud data of each non-reference pose and the reference knee 3D surface model includes: The patellar center, patellar contour feature points, and geometric feature points on both sides of the knee joint surface are extracted from the 3D point cloud data of the knee in the reference pose and various non-reference poses. Then, based on the patellar center and patellar contour feature points, the 3D point cloud data of the knee in various non-reference poses is initially registered with the reference 3D surface model of the knee to obtain the initial registration result. Subsequently, the initial registration result is non-rigidly corrected based on the matching relationship of the geometric feature points on both sides of the knee joint surface to obtain the surface mapping relationship.

[0008] Further, the step of recording the region label with the highest protection priority as the fused region label of the surface point based on the reference pose region label and the mapped non-reference pose region labels of the same surface point in the reference knee three-dimensional surface model includes: The protection priority of the avoidance protection zone is higher than that of the transition adjustment zone, and the protection priority of the transition adjustment zone is higher than that of the fixed zone. When the same surface point is labeled as an avoidance protection zone in any knee joint posture, the avoidance protection zone will be used as the fusion region label for that surface point. When the region label of the same surface point in each knee joint posture is not an avoidance protection zone and at least one region label is a transition adjustment zone, the transition adjustment zone is used as the fusion region label of the surface point. When the region label of the same surface point in each knee joint posture is the fitting and fixing area, the fitting and fixing area is used as the fusion region label of that surface point.

[0009] Furthermore, before associating the fused region labels with each surface point in the reference knee 3D surface model, the method further includes connectivity analysis and boundary continuity correction of the fused functional partitions, specifically including: Set an isolated region area threshold, divide adjacent surface points with the same fusion region label into connected regions, determine connected regions with an area smaller than the isolated region area threshold as isolated regions, and adjust the fusion region label of each surface point in the isolated region to the region label with the highest protection priority in its adjacent regions. The boundary of the region is continuously corrected based on the geodesic distance, local curvature and surface normal vector of the surface points on both sides of the boundary of different functional zones, and the protection priority of the corrected fused region label is not lower than that of the original fused region label.

[0010] Furthermore, it also includes the following steps: A posture change threshold is set. When the knee joint posture change is not greater than the posture change threshold, the three-dimensional point cloud data and surface electromyography (EMG) signals of the knee at multiple acquisition times are acquired simultaneously. The muscle activation index at each acquisition time is obtained based on the surface EMG signals. Then, when constructing the benchmark three-dimensional surface model of the knee, the surface EMG signals are acquired simultaneously, and the benchmark muscle activation index is obtained based on the surface EMG signals. The three-dimensional point cloud data of the knee at each acquisition time is registered with the reference three-dimensional surface model of the knee. The normal change and tangential change of each surface point along the local surface normal and the local surface tangential are obtained. Then, the muscle activation-normal deformation relationship and the muscle activation-tangential deformation relationship are established according to the muscle activation index and the normal and tangential changes, respectively.

[0011] Furthermore, the current muscle activation index is obtained, and the normal morphological compensation amount of each surface point is obtained based on the current muscle activation index, the baseline muscle activation index, and the muscle activation-normal deformation relationship. The tangential deformation compensation parameter of each surface point is obtained based on the current muscle activation index, the baseline muscle activation index, and the muscle activation-tangential deformation relationship. When performing morphological compensation based on the normal morphological compensation amount and the tangential deformation compensation parameter, the fusion region label of each surface point remains unchanged.

[0012] Furthermore, based on the three-dimensional position of each surface point in the reference knee three-dimensional surface model, the corresponding protective gear target position is generated, and each protective gear target position constitutes the protective gear target surface. Then, the minimum safety gap of the avoidance protection zone is set, and the normal shape compensation amount is constrained according to the fusion area label. For the fitting and fixing area, the corresponding protective gear target position is compensated according to the normal shape compensation amount. For the transition adjustment zone, the geodetic distances from the surface point to the boundary of the fixed contact zone and the boundary of the avoidance protection zone are obtained respectively. The ratio of the geodetic distance of the boundary of the avoidance protection zone to the sum of the two geodetic distances is used as the compensation coefficient. The product of the normal shape compensation amount and the compensation coefficient is used as the actual compensation amount. The corresponding protective gear target position is compensated according to the actual compensation amount. For the avoidance protection zone, the corresponding protective gear target position is compensated according to the normal morphology compensation amount, and the distance between the compensated protective gear target surface and the patient's knee surface along the local surface normal is not less than the minimum safety gap.

[0013] Furthermore, it also includes the following steps: Acquire three-dimensional point cloud data of the knee and muscle activation index at multiple detection times; wherein the knee joint posture change at each detection time is not greater than the posture change threshold. Then, the three-dimensional point cloud data of the knee at each detection time is surface registered with the reference three-dimensional surface model of the knee to obtain the actual normal change of each surface point. Based on the muscle activation index and the muscle activation-normal deformation relationship, the predicted normal change is obtained. Then, the difference between the actual normal change and the predicted normal change is calculated, and the difference is used as the residual deformation. Based on the residual deformation of the same surface point at multiple detection times, the region of continuous morphological change is determined. Then, the reference knee three-dimensional surface model is updated based on the region of continuous morphological change, and the fusion region label is updated based on the updated reference knee three-dimensional surface model.

[0014] Furthermore, the step of determining the region of continuous morphological change based on the residual deformation of the same surface point at multiple detection times includes: Set thresholds for residual deformation, consecutive detection count, continuous region area, and single update volume; When the absolute value of the residual deformation of the same surface point at consecutive detection times is greater than or equal to the residual deformation threshold, the number of consecutive detection times is not less than the consecutive detection count threshold, and the sign of the residual deformation remains consistent, the surface point is determined to be a point of continuous morphological change. Based on the adjacency relationship of surface points in the reference knee three-dimensional surface model, the points of continuous morphological change are clustered, and the clustered regions with an area not less than the continuous region area threshold are determined as regions of continuous morphological change. The reference knee 3D surface model is updated along the local surface normal based on the residual deformation of each surface point within the continuous morphological change region, and the single position update amount of each surface point is not greater than the upper limit of the single update amount, and the surface points whose fusion region label was the avoidance protection zone before the update retain the avoidance protection zone fusion region label after the update.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention maps region labels of multiple knee joint postures to the same baseline 3D knee surface model and fuses them according to protection priority. This ensures that surface locations that need to be avoided in different postures maintain avoidance attributes. Only locations where the fit conditions are met in multiple postures are classified as fit fixation zones. At the same time, it combines geodesic distance, local curvature, and normal protrusion for region classification. By using anatomical feature registration, connectivity analysis, and boundary continuity correction, it reduces label mismatch, surface misconnection, and boundary abrupt changes. This provides local fit, continuous transition, and injury avoidance constraints for protective gear, reducing the risk of local compression caused by single posture modeling.

[0016] By simultaneously acquiring three-dimensional point cloud data and surface electromyography signals of the knee, the patient's own muscle activation-normal deformation relationship and muscle activation-tangential deformation relationship are established. Normal compensation, gradual compensation, and minimum safety gap constraints are performed according to functional zones. The target position of the protective gear can be adjusted without changing the damage protection attributes when muscle activation changes. Through residual deformation detection and restricted update mechanism, it can adapt to continuous morphological changes of the knee and maintain damage protection constraints, thereby improving the adaptability and safety of the protective gear under multi-posture and dynamic use conditions. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the overall process of the method steps of the present invention. Detailed Implementation

[0018] 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.

[0019] In the following embodiments, this method is executed by a data processor. The data processor receives knee 3D point cloud data output by a structured light 3D scanning device, knee joint posture data output by a knee joint posture detection device, and surface electromyography (EMG) signals output by a surface EMG acquisition device, and outputs the processing results to the control interface of the emergency robot protective gear. The emergency robot protective gear includes a support unit, a flexible connecting layer, and a sliding connector. The sliding connector includes a limiting groove extending tangentially along the knee surface, a limiting block disposed within the limiting groove, and a miniature linear drive for adjusting the position of the limiting block.

[0020] This method directly outputs the 3D segmentation results of the knee and the control parameters for the brace, without involving disease diagnosis, surgery, or treatment. The sampling points in the original scan data are called 3D points, and the vertices of the triangular mesh formed after surface reconstruction are called surface points. This indicates the index of the surface point in the reference knee 3D surface model. This indicates the acquisition time sequence number during the deformation relationship establishment phase; the reference knee 3D surface model is denoted as... , No. The three-dimensional positions of the surface points are denoted as follows: The unit surface normal vector is denoted as The upper limits of 3D scanning error, surface mapping error, and support unit execution position error are all obtained through repeated measurements, and the 95th percentile of the absolute value of the repeated measurement error is taken. The standard deviation of the error used for statistical discrimination is calculated separately from the repeated measurement samples.

[0021] Please see Figure 1 This invention provides a method for segmenting the fitting area of ​​a robotic protective gear for patellar injury emergency treatment, comprising the following steps: S1. Obtain three-dimensional point cloud data of the patient's knee in multiple different knee joint postures, select one of the multiple different knee joint postures as the reference posture, take the other knee joint postures as non-reference postures, and then construct a reference knee three-dimensional surface model based on the three-dimensional point cloud data of the knee in the reference posture. Specifically, the data processor first receives at least three knee joint posture parameters input by medical staff. Each knee joint posture is represented by the angle between the longitudinal axis of the thigh segment and the longitudinal axis of the lower leg segment, and is associated with and stored as a posture number, knee flexion angle, posture holding time, and acquisition time. Based on the completeness of point cloud coverage, the size of motion artifacts, and the stability of posture holding, the data processor selects one knee joint posture as the reference posture from multiple knee joint postures, and the remaining postures are used as non-reference postures.

[0022] During each posture maintenance, the structured light 3D scanning device scans the injured knee, extending from the distal thigh to the proximal calf, covering the patella, surrounding soft tissues, and both sides of the knee joint. The acquired 3D point cloud data of the knee includes at least 3D point coordinates, scan confidence, acquisition timestamp, and posture number. The scan confidence threshold is obtained by repeatedly scanning a standard surface. The lowest scan confidence among the valid 3D points that meet the upper limit of 3D scanning error is selected. The data processor deletes 3D points with scan confidence below the threshold and sequentially performs statistical outlier removal, voxel downsampling, and hole inspection. The hole area threshold is set between the maximum non-anatomical notch area formed by repeated scanning and the minimum anatomical depression area to be retained. Only non-anatomical notches with an area lower than the hole area threshold are interpolated using the surface. Notches with an area not lower than the threshold generate a prompt for re-acquisition.

[0023] The data processor performs unit surface normal vector estimation and triangular mesh surface reconstruction on the knee 3D point cloud data of the baseline pose and various non-baseline poses to obtain the knee 3D surface model for each pose. The unit surface normal vector is uniformly pointed outward from the patient's body. The model formed in the baseline pose is denoted as the baseline knee 3D surface model. , of which The three-dimensional positions of the surface points are denoted as follows: The unit surface normal vector is denoted as The data processor also saves the projection position of each 3D point on the corresponding knee 3D surface model, the number of the corresponding triangle, and the centroid coordinates, so that the region label of the 3D point can be transferred to the surface point.

[0024] During the acquisition of 3D point cloud data of the knee in the baseline posture, the surface electromyography (SEMG) acquisition device simultaneously acquires SEMG signals from the rectus femoris, vastus medialis, and vastus lateralis muscles. The three signals are sequentially processed by bandpass filtering from 20 Hz to 450 Hz, power frequency notch filtering from 50 Hz, full-wave rectification, and root mean square (RMS) smoothing. The RMS smoothing window is 200 milliseconds. During the muscle activation calibration phase, three SEMG signals are acquired when the patient is in a safe isometric contraction state. The maximum effective RMS amplitude of each acquisition channel is used as the normalized reference amplitude of that channel. The channel weights for the rectus femoris, vastus medialis, and vastus lateralis muscles are all one-third. The normalized SEMG amplitudes of the three channels are multiplied by their respective channel weights and then summed to obtain the baseline muscle activation index.

[0025] S2. Obtain patellar injury information, delineate the injury protection area for each knee joint posture, and then obtain the surface spatial relationship parameters and local surface morphology parameters of each three-dimensional point in the knee three-dimensional point cloud data of each knee joint posture and the injury protection area. Based on the surface spatial relationship parameters and local surface morphology parameters, classify each three-dimensional point into regions to obtain the functional partitions of each knee joint posture and the region labels of each three-dimensional point. The functional partitions include the fitting and fixation area, the transition adjustment area, and the avoidance protection area. Specifically, patellar injury information includes the patellar injury location and outline already marked in the CT image, as well as the protection margin input by medical staff. The data processor performs three-dimensional segmentation of the patella and the outer surface of the knee in the CT image to form the CT patella surface and the CT knee outer surface, and extracts the patella center, the upper edge of the patella, the lower edge of the patella, the medial edge of the patella, and the lateral edge of the patella.

[0026] During the CT image acquisition and knee 3D point cloud acquisition process, at least three positioning marks that can be recognized by CT images and structured light 3D scanning devices are set on the skin surface of the patient's knee. The data processor uses the positioning marks to obtain the initial rigid transformation from the CT coordinate system to the point cloud coordinate system. Then, the outer surface of the CT knee is registered with the reference knee 3D surface model to obtain the coordinate transformation relationship from the CT coordinate system to the reference point cloud coordinate system.

[0027] The data processor transforms the damage contour in the CT image to a reference point cloud coordinate system, and draws rays from the center of the patella through each contour point on the damage contour to the outer surface of the knee. The intersection of the rays with the outer surface of the knee forms the surface damage contour of the reference posture. Then, the protection margin is expanded along the knee surface to form the damage protection area of ​​the reference posture. For each non-reference posture, the coordinate transformation is completed using positioning marks and patellar surface features. The same damage contour is projected onto the three-dimensional surface of the knee in the corresponding posture, and the same protection margin is expanded along the corresponding surface to obtain the damage protection area of ​​each non-reference posture.

[0028] For any knee joint posture, the data processor uses triangular mesh surface points as graph nodes, mesh edges as connecting edges, and the surface edge length between adjacent surface points as edge weights. Surface points on the boundary of the damage protection zone are used as multi-source starting points. The geodesic distance from each surface point to the boundary of the damage protection zone is obtained through a multi-source shortest path algorithm. The original 3D points obtain their corresponding geodesic distances based on their projection positions on the triangular mesh. The surface spatial relationship parameters of each 3D point include the location identifier of the 3D point inside or outside the damage protection zone and the geodesic distance from the 3D point to the boundary of the damage protection zone.

[0029] A first distance threshold and a second distance threshold are set, with the second distance threshold being greater than the first distance threshold. The first distance threshold is not less than the sum of the damage protection margin, the upper limit of the 3D scanning error, the upper limit of the surface mapping error, and the upper limit of the support unit execution position error. The difference between the second distance threshold and the first distance threshold is not less than the length obtained by dividing the normal height difference between the avoidance protection zone and the bonding and fixing zone by the maximum allowable surface slope of the flexible connection layer. For example, if the damage protection margin is 7 mm, the sum of the upper limit of the 3D scanning error and the upper limit of the surface mapping error is 2 mm, and the upper limit of the support unit execution position error is 1 mm, the first distance threshold can be set to 10 mm; if the normal height difference between the avoidance protection zone and the bonding and fixing zone is 4 mm, and the maximum allowable surface slope of the flexible connection layer is 0.2, the second distance threshold can be set to 30 mm.

[0030] The data processor divides the 3D points within the damage protection area into avoidance protection zones. For 3D points outside the damage protection area, those with a geodesic distance not greater than a first distance threshold are divided into avoidance protection zones, those with a geodesic distance greater than the first distance threshold but not greater than a second distance threshold are divided into transition adjustment zones, and those with a geodesic distance greater than the second distance threshold are divided into fitting and fixing zones, thereby obtaining the initial region labels for each 3D point.

[0031] After obtaining the initial region labels, a local neighborhood is established centered on the projection position of each 3D point on the triangular mesh. The neighborhood radius is eight times the average point spacing of the corresponding point cloud. A local reference surface is fitted using surface points within the neighborhood other than the current 3D point, and the signed distance of the current 3D point relative to this local reference surface along the unit surface normal vector is calculated. The distance towards the protective gear side is taken as the normal projection. A quadratic surface is fitted using surface points within the neighborhood to obtain two local principal curvatures, and half the absolute value of the sum of the two local principal curvatures is taken as the local curvature.

[0032] The curvature threshold is taken as the 90th percentile of the local curvature on the knee surface that is far from the injury protection area, has not undergone hole repair, and is scan-stable. The normal protrusion threshold is taken as three times the standard deviation of the 3D scan repeatability error. For example, when the standard deviation of the 3D scan repeatability error is 0.4 mm, the normal protrusion threshold is set to 1.2 mm. The curvature threshold and the normal protrusion threshold are written into the patient parameter file after the first effective scan, and the same set of thresholds is used for subsequent posture processing and model updates.

[0033] When the local curvature of a 3D point exceeds the curvature threshold and the normal protrusion exceeds the normal protrusion threshold, its initial region label is upgraded by one protection level. Specifically, the fixed-fit area is adjusted to a transition adjustment area, the transition adjustment area is adjusted to an avoidance protection area, and areas already in the avoidance protection area remain unchanged. After adjustment, based on the projection position and centroid coordinates of the 3D point, the region label is transferred to the corresponding surface point of the knee 3D surface model. When multiple labels are obtained for the same surface point, the region label with the highest protection priority is retained, resulting in the functional zoning of each knee joint posture.

[0034] S3. Establish the surface mapping relationship between the three-dimensional point cloud data of the knee in each non-reference posture and the three-dimensional surface model of the reference knee. Based on the surface mapping relationship, map the region labels of each non-reference posture to the three-dimensional surface model of the reference knee to obtain the region labels of different knee joint postures. Specifically, the data processor extracts the patellar center, patellar contour feature points, and bilateral surface geometric feature points from the knee three-dimensional point cloud data of the baseline posture and various non-baseline postures, using the knee three-dimensional surface model reconstructed from the corresponding point cloud data. The patellar contour feature points include the patellar superior edge feature points, inferior edge feature points, medial edge feature points, and lateral edge feature points. The search range of these feature points is limited by the projection position of the CT image, and then the surface curvature extreme points in the corresponding directions are selected from the search range. The bilateral surface geometric feature points of the knee joint include the femoral medial condyle surface curvature extreme points, femoral lateral condyle surface curvature extreme points, medial knee stable protrusions, and lateral knee stable protrusions, and are extracted from the preset anatomical regions on the medial and lateral sides of the knee joint, respectively.

[0035] For any non-reference pose, first make the patella center coincide with the patella center in the reference knee 3D surface model, and then use weighted rigid registration to solve the rotation and translation parameters to minimize the sum of weighted distances between patella contour feature points to obtain the initial registration result. The feature point weight is proportional to the scan confidence and inversely proportional to the position variance of the feature points in the repeated extraction process.

[0036] An embedded deformation map is built on the initial registration results. Deformation nodes are set at geodesic distance intervals along the knee surface of the non-reference pose. The node interval is five to ten times the local average grid side length and less than the minimum width of the transition adjustment zone. Each deformation node is configured with local rotation and translation parameters. Non-rigid correction considers surface point distance, unit surface normal vector consistency, smoothness of adjacent deformation nodes, patellar contour constraints, and geometric feature point constraints on both sides of the knee joint surface. Each constraint term is first normalized according to the initial residual in the calibration point cloud, and then the constraint weight is selected based on the condition of minimizing the surface mapping error of the verification feature points. The constraint weight remains unchanged in each non-reference pose registration of the same patient.

[0037] The surface registration convergence threshold is set to one-tenth of the nominal spatial resolution of the structured light 3D scanning device. Non-rigid correction convergence is determined when the average surface distance change between two adjacent iterations is below this threshold, and the maximum mapping error of the verified feature points does not exceed the upper limit of the surface mapping error. Results that reach the maximum number of iterations but do not simultaneously meet the above conditions are not considered valid mapping results.

[0038] After convergence, each non-reference attitude surface point records the reference model triangle face number it falls into and the three centroid weights to form a surface mapping relationship. The region label of the non-reference attitude surface point is transferred to the three surface points of the corresponding reference model triangle face according to the centroid weight. When the same reference model surface point receives multiple region labels of the same non-reference attitude, the region label with the highest protection priority is retained.

[0039] When the error of feature extraction, surface registration, or label mapping exceeds the upper limit of surface mapping error, the pose is marked as invalid and a re-acquisition command is generated. After receiving the re-acquisitioned 3D point cloud data of the knee, feature extraction and surface registration are re-executed. If the surface mapping error requirement is not met after two consecutive attempts, the pose data will not be used for region label fusion. After all non-reference poses are processed, each surface point in the reference knee 3D surface model has a reference pose region label and a region label mapped from each non-reference pose.

[0040] S4. Then set the protection priorities of the avoidance protection zone, transition adjustment zone and fitting and fixing zone. Based on the reference attitude area label of the same surface point in the reference knee three-dimensional surface model and the mapped non-reference attitude area labels, the area label with the highest protection priority is recorded as the fusion area label of the surface point, and the fusion functional partition is obtained.

[0041] Specifically, the protection priorities of the avoidance protection zone, transition adjustment zone, and fitting fixation zone are set to Level 3, Level 2, and Level 1, respectively. For each surface point in the reference knee 3D surface model, its reference posture region label and the region labels mapped from each non-reference posture are read. When the region label of any knee posture is the avoidance protection zone, the avoidance protection zone is used as the fusion region label of that surface point. When none of the postures belong to the avoidance protection zone but at least one posture belongs to the transition adjustment zone, the transition adjustment zone is used as the fusion region label. Only when the region labels of all postures are the fitting fixation zone is the fitting fixation zone used as the fusion region label. After processing all surface points, the fusion functional partition is obtained.

[0042] S5. Associate the fusion region labels with each surface point in the reference knee 3D surface model to obtain the target knee 3D segmentation model, and output the 3D position and fusion region labels of each surface point in the target knee 3D segmentation model.

[0043] Specifically, the data processor, based on the mesh adjacency relationship of the baseline knee 3D surface model, assigns surface points with the same fusion region label and shared mesh edges to the same connected region, and uses the sum of the areas of the triangular faces covered by the connected region as the area of ​​the connected region. The isolated region area threshold is greater than the maximum pseudo-region area formed by scan noise, but less than the minimum controllable support area of ​​a single support unit and the area of ​​the minimum damage protection area and its external avoidance zone in each pose. For example, if the maximum pseudo-region area formed by scan noise is 25 square millimeters, the minimum controllable support area of ​​a single support unit is 100 square millimeters, and the area of ​​the minimum damage protection area and its external avoidance zone is greater than 100 square millimeters, the isolated region area threshold can be set to 50 square millimeters. If no such usable interval exists between the parameters, the scan density is increased and the triangular mesh is reconstructed until the scan pseudo-regions and effective functional zones can be distinguished.

[0044] Connected regions with an area smaller than the isolated region area threshold are identified as isolated regions. The data processor reads the fusion region labels of adjacent grid regions of the isolated region and adjusts the fusion region labels of each surface point within the isolated region to the region labels of the highest protection priority among the adjacent regions. Since the isolated region area threshold is smaller than the area of ​​the minimum damage protection area and its external avoidance zone, the actual damage protection range will not be identified as an isolated region.

[0045] After completing the connectivity analysis, the regional boundaries between different functional zones are extracted. On the initially extracted regional boundaries, the geodesic distance difference, local curvature difference, and unit surface normal angle between adjacent boundary surface points are statistically analyzed. The 90th percentile value of the statistical values ​​of the stable boundary samples is taken as the geodesic distance continuity threshold, curvature continuity threshold, and normal continuity threshold, respectively. Stable boundary samples refer to boundary surface points that do not intersect with the hole repair area, whose local curvature does not exceed the curvature threshold, and whose scan confidence level is not lower than the scan confidence level threshold. The above continuity thresholds remain unchanged in subsequent boundary correction processes.

[0046] When any parameter of an adjacent boundary surface point exceeds the corresponding continuity threshold, a connection path with small changes in geodesic length, local curvature, and unit surface normal vector is searched within the grid neighborhood near the boundary. This connection path is then used to correct the regional boundary. Surface points on the lower protection level side between the old boundary and the corrected boundary are upgraded to a higher protection level. The fusion region label on the higher protection level side remains unchanged. After correction, it is checked whether all surface points within the damage protection area and those with a geodesic distance not greater than the first distance threshold belong to the avoidance protection zone. It is also checked whether the protection priority of each surface point after correction is not lower than the protection priority before correction.

[0047] After the inspection is passed, the 3D position, unit surface normal vector and corrected fusion region label of each surface point are associated and saved. All surface point data, triangular facet index and functional partition boundary together constitute the 3D segmentation model of the target knee. The output data includes at least the 3D position and fusion region label of each surface point.

[0048] S6. Under the condition that the knee joint posture change is not greater than the posture change threshold, simultaneously acquire three-dimensional point cloud data and surface electromyography signals of the knee at multiple acquisition times, and establish muscle activation-normal deformation relationship and muscle activation-tangential deformation relationship. Specifically, the attitude change threshold is set based on the angle measurement error of the knee joint attitude detection device and the knee surface displacement caused by attitude changes. The lower limit of the threshold is three times the standard deviation of the angle measurement error, and the upper limit of the threshold is the maximum angle difference when the normal surface displacement caused by attitude changes does not exceed the upper limit of the three-dimensional scanning error. The attitude change threshold is selected between these two values. For example, under the above conditions, the attitude change threshold can be set to 3 degrees. When the lower limit of the threshold is greater than the upper limit of the threshold, the current data set is marked as unusable for deformation relationship fitting, and an attitude fixation and inertial measurement unit calibration prompt is generated.

[0049] The structured light 3D scanning device, surface electromyography (SEMG) acquisition device, and knee joint posture detection device use the same time reference. The acquisition time of each frame of knee 3D point cloud data is taken at the midpoint of the exposure time interval, and the SEMG signal is extracted 100 milliseconds before and after that moment. The upper limit of the synchronization time error is one-tenth of the length of the root mean square smoothing window. When the root mean square smoothing window is 200 milliseconds, the upper limit of the synchronization time error is 20 milliseconds. Data sets exceeding this time error are not included in the relation fitting.

[0050] Calculate the signal processing method according to step S1. Muscle activation index at each collection time point The 3D point cloud data of the knee at each acquisition time was then surface-registered with the baseline 3D surface model of the knee. Data sets with surface registration errors exceeding the upper limit of surface mapping errors were not included in the deformation relationship fitting.

[0051] The first The coordinate difference between the matching surface position and the three-dimensional position of the reference surface point at each acquisition time is used as the displacement vector. The projection of the displacement vector onto the normal vector of the unit surface is used as the normal change, where it is positive along the external direction and negative along the internal direction. The normal component is subtracted from the displacement vector to obtain the tangential displacement vector located in the tangential plane of the local surface, and the length of the tangential displacement vector is used as the tangential change.

[0052] For each surface point, at least five sets of valid muscle activation index and deformation data were collected, ensuring that the valid data covered at least three different muscle activation levels. A univariate linear regression relationship between the muscle activation index and the normal deformation was fitted using the least squares method to obtain the muscle activation-normal deformation relationship. The same method was used to fit the univariate linear regression relationship between the muscle activation index and the tangential change, thus obtaining the muscle activation-tangential deformation relationship. The independent variable for both relationships is the muscle activation index, and the output is length.

[0053] Data that did not participate in the fitting were extracted from the valid data as validation data. The deformation prediction error was calculated. The allowable value of the deformation relationship prediction error was taken as the 99.7 percentile of the absolute value of the prediction error of the stable posture validation data. When the prediction error of a surface point exceeded the allowable value, the surface point was marked as invalid deformation relationship, and a local re-acquisition prompt was generated. Before a valid deformation relationship was obtained again, the surface point did not participate in muscle activation morphology compensation.

[0054] S7. Obtain the current muscle activation index, calculate the normal morphological compensation amount and tangential deformation compensation parameters of each surface point, and keep the fusion region label of each surface point unchanged. Specifically, the current muscle activation index is obtained according to the signal processing method in step S1, for the first muscle in the benchmark knee three-dimensional surface model. For each surface point, the normal morphological compensation amount is calculated according to the following formula (1): (1) In equation (1), The current muscle activation index, The normal morphological compensation amount is represented by a positive value indicating an increase in compensation in the outward direction relative to the baseline muscle activation state, and a negative value indicating a decrease in compensation in the inward direction. The tangential deformation compensation parameter is calculated according to the following formula (2): (2) In equation (2), This represents the adjustment amount of the required tangential sliding stroke of the flexible connection layer relative to the reference state, and does not represent the active tangential displacement of the support unit. The reference sliding stroke of the sliding connector is obtained through the tensile calibration of the flexible connection layer. The midpoint of the safe sliding range that does not produce plastic deformation and connection separation is taken as the reference sliding stroke. The reference sliding stroke is compared with... The two are added together to obtain the current required sliding stroke. If the current required sliding stroke is lower than the minimum safe sliding stroke, the minimum safe sliding stroke is taken. If the current required sliding stroke exceeds the maximum stroke of the limit groove, insufficient stroke status information is generated. The control parameters of the micro linear drive are generated according to the current required sliding stroke.

[0055] The fusion region labels in the 3D segmentation model of the target knee are saved as read-only data. The normal morphological compensation amount is only used to adjust the target position of the protective gear, and the tangential deformation compensation parameter is only used to adjust the allowable sliding stroke of the sliding connector. The fusion region labels of each surface point remain unchanged before and after morphological compensation.

[0056] S8. Generate the target surface of the protective gear based on the surface points in the reference knee 3D surface model, and compensate the normal shape according to the fusion region label constraint. Specifically, the data processor reads the pre-calibrated zero position of the support unit and the effective thickness of the flexible connecting layer. The unstressed thickness of the flexible connecting layer and the compression under the predetermined support load are obtained through material compression calibration. The difference between the two is taken as the effective thickness under the predetermined support load. Based on the zero position of the support unit, the effective thickness of the flexible connecting layer and the preset bonding spacing, the data processor calculates the normal spacing after forming the preset bonding state, and takes the normal spacing as the basic bonding distance of the bonding fixing area.

[0057] The basic avoidance distance of the avoidance protection zone is greater than the basic contact distance of the contact fixed zone, and not less than the minimum safety gap. The minimum safety gap is not less than the sum of the net protection gap input by medical staff, the upper limit of three-dimensional scanning error, the upper limit of surface mapping error, the upper limit of support unit execution position error, and the maximum compression of the flexible connection layer under the predetermined support load.

[0058] The data processor offsets each surface point outwards along the unit surface normal vector by the corresponding basic distance to generate the target position of the basic protective gear. The basic fitting distance of the fitting and fixing zone is used, and the basic avoidance distance of the avoidance and protection zone is used. For the transition adjustment zone, the geodesic distance from the surface point to the boundary of the fitting and fixing zone and the boundary of the avoidance and protection zone is obtained. Based on the two geodesic distances, a linear interpolation is performed between the basic fitting distance of the fitting and fixing zone and the basic avoidance distance of the avoidance and protection zone. The target positions of each basic protective gear constitute the target surface of the basic protective gear.

[0059] In the current muscle activation state, each surface point moves along the unit surface normal vector from the reference three-dimensional position by a normal morphological compensation amount, thereby forming the current knee surface for safety clearance inspection. For the fitting fixation area, the target position of the basic protective gear is adjusted according to the complete normal morphological compensation amount. For the transition adjustment area, the geodesic distance from the surface point to the boundary of the fitting fixation area and the geodesic distance to the boundary of the avoidance protection area are obtained, and the compensation coefficient is calculated according to the following formula (3): (3) In equation (3), This represents the compensation coefficient, which is dimensionless and ranges from 0 to 1. It is particularly relevant when the coefficient is close to the boundary of the fixed area. Approaching 1; when nearing the boundary of the avoidance protected area, Approaching 0 The geodetic distance from the surface point to the boundary of the fixed area is given by the following terminology. The distance from the surface point to the boundary of the avoidance protection zone is the geodetic distance; the actual compensation amount in the transition adjustment zone is the normal morphological compensation amount and... The product of gradually decreases from the side of the fixed area to the side of the avoidance protection area.

[0060] For the avoidance protection zone, first adjust the target position of the basic protective gear according to the normal morphological compensation amount, then calculate the distance between the adjusted target position of the protective gear and the current knee surface along the unit surface normal vector. When this distance is less than the minimum safe gap, continue to move the target position of the protective gear outward until the normal distance reaches the minimum safe gap. All the target positions of the protective gear after the area constraint together constitute the target surface of the protective gear.

[0061] The data processor divides the target surface of the protective gear into multiple support unit coverage areas based on the installation position, motion axis, and preset radius of each support unit in the protective gear installation coordinate system. Surface points belonging to "two coverage" areas are assigned to the support unit coverage area with the smaller distance based on their projection distance to the corresponding motion axis. The data processor projects the target position of the protective gear within each support unit coverage area to the motion axis of that support unit and takes the median of the projection values ​​as the target geometric position of that support unit coverage area.

[0062] Subsequently, by combining the zero position of the support unit, the effective thickness of the flexible connecting layer, and the predetermined compression amount, the target geometric position is converted into the target extension amount of the support unit. If the target extension amount exceeds the mechanical stroke of the support unit and the limiting would cause the safety clearance of the avoidance protection zone to be lower than the minimum safety clearance, the support unit is marked as having insufficient stroke, and no execution parameters are generated for that support unit; if no safety clearance risk is generated, the target extension amount is limited to the mechanical stroke range, and the data processor outputs the target surface of the protective gear, the target extension amount of each support unit, and the target sliding stroke of each sliding connector to the control interface of the emergency robot protective gear.

[0063] S9. Acquire three-dimensional point cloud data and muscle activation index of the knee at multiple detection times, and calculate the residual deformation of each surface point; Specifically, during multiple detection moments while the protective gear is in use, three-dimensional point cloud data and surface electromyography (EMG) signals of the knee are acquired simultaneously. The muscle activation index at each detection moment is calculated according to the signal processing method in step S1. Only data with knee joint posture changes not exceeding the posture change threshold, synchronization time error between point cloud and surface EMG signals not exceeding 20 milliseconds, and surface registration error not exceeding the upper limit of surface mapping error are retained.

[0064] The data processor performs surface registration between the 3D point cloud data of the knee at each detection time and the reference 3D surface model of the knee. It projects the displacement between the matched surface position and the reference surface point at each detection time onto the unit surface normal vector to obtain the actual normal change. Then, it substitutes the muscle activation index at each detection time and the reference muscle activation index into the muscle activation-normal deformation relationship, using the difference between the two output values ​​as the predicted normal change. Finally, it subtracts the predicted normal change from the actual normal change to obtain the residual deformation. A positive residual deformation indicates that, after deducting the predicted deformation caused by muscle activation, the knee surface still changes in the outward direction; a negative residual deformation indicates that the knee surface still changes in the inward direction. The residual deformation of each surface point at consecutive detection times is saved in chronological order and input into S10.

[0065] S10. Determine the region of continuous morphological change based on the residual deformation, and update the reference knee 3D surface model and fusion region label. Set residual deformation threshold Threshold for consecutive detection times Continuous region area threshold and the maximum amount of data to be updated per session The residual deformation threshold is obtained through stable morphological calibration data. Under the conditions of stable knee joint posture, no continuous change in knee morphology, and known muscle activation index, multiple sets of three-dimensional point cloud data and surface electromyography signals of the knee are continuously acquired. The residual deformation is calculated according to step S9, and the 99.7 percentile value of the absolute value of the residual deformation is taken as the residual deformation threshold. This method of value selection reflects the comprehensive influence of the prediction errors of three-dimensional scanning, surface mapping, and muscle activation-normal deformation relationship. It does not require that the errors are independent of each other. The threshold for the number of consecutive detections... This indicates the minimum number of consecutive detections required to determine if a continuous morphological change has occurred at the same surface point.

[0066] The threshold for the number of consecutive detections is set based on the shortest duration confirmation time and the time interval between adjacent detection times, rounded up to the nearest integer after dividing the shortest duration confirmation time by the detection time interval. For example, if the time interval between adjacent detection times is 30 seconds and the shortest duration confirmation time is 120 seconds, the threshold for the number of consecutive detections can be set to 4. The threshold for the continuous area is not less than the minimum controllable support area of ​​a single support unit, for example, it can be set to 100 square millimeters. The upper limit of the single update amount is the minimum value among the single safe adjustment stroke of the support unit, the maximum surface point displacement allowed to maintain the local continuity of the triangular mesh, and the single morphological correction amount input by medical personnel, for example, it can be set to 1 millimeter.

[0067] A surface point is classified as a point of persistent morphological change when the absolute value of its residual deformation is not less than the residual deformation threshold for at least four consecutive detection times, and the signs of these residual deformations remain consistent. The data processor clusters these persistent morphological change points based on the mesh adjacency relationship of the baseline knee 3D surface model and calculates the area of ​​the triangular faces covered by each cluster. Clusters with an area not less than the continuous region area threshold are classified as persistent morphological change regions; clusters with an area less than the continuous region area threshold retain the detection record, but the baseline knee 3D surface model is not updated in this round.

[0068] For the first region of continuous morphological change Extract the nearest surface point from the given surface points. The residual deformation variables were obtained at each consecutive detection time, and the median was taken after arranging these residual deformation variables numerically. This median is denoted as . ,when When taking 4, It is calculated from the residual deformation of the surface point at the four most recent consecutive detection times; when the effective consecutive detection data is less than At any time, not calculated The three-dimensional position of the surface point is not updated, and then the single position update amount of the surface point is obtained according to the following formula (4), specifically: (4) In equation (4), This refers to the amount of a single location update. This is the maximum amount that can be updated in a single session. Represents a symbolic function. This indicates taking the smaller value. and The units are all millimeters; when When positive, the surface point moves outward along the unit surface normal vector; when... When the value is negative, the surface point moves into the body along the unit surface normal vector, and the distance moved is... The absolute value of . Therefore, the single position update amount of each surface point does not exceed . The edge of the updated region is smoothed using a surrounding mesh neighborhood, ensuring that the displacement of any surface point after smoothing does not exceed [the specified value]. After completing the position update, recalculate the unit surface normal, geodesic distance, local curvature, and normal bulge in the updated region and its surrounding grid neighborhood, and clear the residual deformation sequence that has participated in this update to avoid the same persistent morphological change being repeatedly accumulated.

[0069] The data processor remaps the three-dimensional surface models of the knees in each non-reference posture to the updated reference three-dimensional surface model of the knees, and performs region classification, region label mapping, protection priority fusion, connectivity analysis and boundary continuity correction according to steps S2 to S5. Before the update, the fused region label sets the protection lock mark for the surface points of the avoidance protection zone. When a lower protection level is obtained after reprocessing, the surface point still retains the fused region label of the avoidance protection zone, and the other surface points use the fused region label obtained after reprocessing. Finally, the updated reference three-dimensional surface model of the knees and the fused region label are input into step S8 to generate the updated protective gear target surface, support unit target extension amount and sliding connector target sliding stroke.

[0070] Through the above steps, the three-dimensional point cloud data of the knee, surface electromyography signals, knee joint posture data, and patellar injury area data are sequentially acquired, surface reconstructed, region classified, posture mapped, label fusion, morphological compensation, and model updated in the same data processing chain. Finally, the target surface and control parameters that can be read by the control interface of the emergency robot protective gear are output.

[0071] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for segmenting the fitting area of ​​a robotic protective garment for patellar injury emergency treatment, characterized in that, Includes the following steps: S1. Obtain three-dimensional point cloud data of the patient's knee in multiple different knee joint postures, select one of the multiple different knee joint postures as the reference posture, take the other knee joint postures as non-reference postures, and then construct a reference knee three-dimensional surface model based on the three-dimensional point cloud data of the knee in the reference posture. S2. Obtain patellar injury information, delineate the injury protection area for each knee joint posture, and then obtain the surface spatial relationship parameters and local surface morphology parameters of each three-dimensional point in the knee three-dimensional point cloud data of each knee joint posture and the injury protection area. Based on the surface spatial relationship parameters and local surface morphology parameters, classify each three-dimensional point into regions to obtain the functional partitions of each knee joint posture and the region labels of each three-dimensional point. The functional partitions include the fitting and fixing area, the transition adjustment area, and the avoidance protection area. S3. Establish the surface mapping relationship between the three-dimensional point cloud data of the knee in each non-reference posture and the reference three-dimensional surface model of the knee. Based on the surface mapping relationship, map the region labels of each non-reference posture to the reference three-dimensional surface model of the knee to obtain the region labels of different knee joint postures. S4. Then set the protection priorities of the avoidance protection zone, transition adjustment zone and fitting fixation zone. According to the reference posture area label and the mapped non-reference posture area labels of the same surface point in the reference knee three-dimensional surface model, the area label with the highest protection priority is recorded as the fusion area label of the surface point to obtain the fusion functional partition. S5. Associate the fusion region label with each surface point in the reference knee 3D surface model to obtain the target knee 3D segmentation model, and output the 3D position and fusion region label of each surface point in the target knee 3D segmentation model.

2. The method for segmenting the fitting area of ​​the patellar injury emergency robot brace according to claim 1, characterized in that, The method of classifying each three-dimensional point into regions based on surface spatial relationship parameters and local surface morphology parameters includes: Obtain the geodesic distance from each three-dimensional point to the boundary of the damage protection area, and set a first distance threshold and a second distance threshold, wherein the second distance threshold is greater than the first distance threshold; The three-dimensional points within the damage protection area are divided into avoidance protection areas. For the three-dimensional points outside the damage protection area, the three-dimensional points with a geodesic distance not greater than the first distance threshold are divided into avoidance protection areas, the three-dimensional points with a geodesic distance greater than the first distance threshold but not greater than the second distance threshold are divided into transition adjustment areas, and the three-dimensional points with a geodesic distance greater than the second distance threshold are divided into fitting and fixing areas, thus obtaining the initial area label of each three-dimensional point. Obtain the local curvature and normal protrusion of each 3D point, set curvature threshold and normal protrusion threshold, and when the local curvature of the 3D point exceeds the curvature threshold and the normal protrusion exceeds the normal protrusion threshold, adjust the initial region label of the fixed area to the transition adjustment area or adjust the initial region label of the transition adjustment area to the avoidance protection area to obtain the region label of each 3D point.

3. The method for segmenting the fitting area of ​​the patellar injury emergency robot brace according to claim 1, characterized in that, The establishment of the surface mapping relationship between the knee 3D point cloud data of each non-reference posture and the reference knee 3D surface model includes: The patellar center, patellar contour feature points, and geometric feature points on both sides of the knee joint surface are extracted from the 3D point cloud data of the knee in the reference pose and various non-reference poses. Then, based on the patellar center and patellar contour feature points, the 3D point cloud data of the knee in various non-reference poses is initially registered with the reference 3D surface model of the knee to obtain the initial registration result. Subsequently, the initial registration result is non-rigidly corrected based on the matching relationship of the geometric feature points on both sides of the knee joint surface to obtain the surface mapping relationship.

4. The method for segmenting the fitting area of ​​the patellar injury emergency robot brace according to claim 1, characterized in that, The step of recording the region label with the highest protection priority as the fused region label of the surface point based on the reference pose region label of the same surface point in the reference knee 3D surface model and the mapped non-reference pose region labels includes: The protection priority of the avoidance protection zone is higher than that of the transition adjustment zone, and the protection priority of the transition adjustment zone is higher than that of the fixed zone. When the same surface point is labeled as an avoidance protection zone in any knee joint posture, the avoidance protection zone will be used as the fusion region label for that surface point. When the region label of the same surface point in each knee joint posture is not an avoidance protection zone and at least one region label is a transition adjustment zone, the transition adjustment zone is used as the fusion region label of the surface point. When the region label of the same surface point in each knee joint posture is the fitting and fixing area, the fitting and fixing area is used as the fusion region label of that surface point.

5. The method for segmenting the fitting area of ​​the patellar injury emergency robot brace according to claim 4, characterized in that, Before associating the fused region labels with each surface point in the reference knee 3D surface model, the method further includes connectivity analysis and boundary continuity correction of the fused functional partitions, specifically including: Set an isolated region area threshold, divide adjacent surface points with the same fusion region label into connected regions, determine connected regions with an area smaller than the isolated region area threshold as isolated regions, and adjust the fusion region label of each surface point in the isolated region to the region label with the highest protection priority in its adjacent regions. The boundary of the region is continuously corrected based on the geodesic distance, local curvature and surface normal vector of the surface points on both sides of the boundary of different functional zones, and the protection priority of the corrected fused region label is not lower than that of the original fused region label.

6. The method for segmenting the fitting area of ​​the patellar injury emergency robot protective gear according to claim 1, characterized in that, It also includes the following steps: A posture change threshold is set. When the knee joint posture change is not greater than the posture change threshold, the three-dimensional point cloud data and surface electromyography (EMG) signals of the knee at multiple acquisition times are acquired simultaneously. The muscle activation index at each acquisition time is obtained based on the surface EMG signals. Then, when constructing the benchmark three-dimensional surface model of the knee, the surface EMG signals are acquired simultaneously, and the benchmark muscle activation index is obtained based on the surface EMG signals. The three-dimensional point cloud data of the knee at each acquisition time is registered with the reference three-dimensional surface model of the knee. The normal change and tangential change of each surface point along the local surface normal and the local surface tangential are obtained. Then, the muscle activation-normal deformation relationship and the muscle activation-tangential deformation relationship are established according to the muscle activation index and the normal and tangential changes, respectively.

7. The method for segmenting the fitting area of ​​the patellar injury emergency robot brace according to claim 6, characterized in that, Obtain the current muscle activation index, and based on the current muscle activation index, the baseline muscle activation index, and the muscle activation-normal deformation relationship, obtain the normal morphological compensation amount for each surface point. Also, based on the current muscle activation index, the baseline muscle activation index, and the muscle activation-tangential deformation relationship, obtain the tangential deformation compensation parameters for each surface point. When performing morphological compensation based on the normal morphological compensation amount and the tangential deformation compensation parameters, keep the fusion region label of each surface point unchanged.

8. The method for segmenting the fitting area of ​​the patellar injury emergency robot brace according to claim 7, characterized in that, Based on the three-dimensional position of each surface point in the reference knee three-dimensional surface model, the corresponding protective gear target position is generated. Each protective gear target position constitutes the protective gear target surface. Then, the minimum safety gap of the avoidance protection zone is set. The normal shape compensation amount is constrained according to the fusion area label. For the fitting and fixing area, the corresponding protective gear target position is compensated according to the normal shape compensation amount. For the transition adjustment zone, the geodetic distances from the surface point to the boundary of the fixed contact zone and the boundary of the avoidance protection zone are obtained respectively. The ratio of the geodetic distance of the boundary of the avoidance protection zone to the sum of the two geodetic distances is used as the compensation coefficient. The product of the normal shape compensation amount and the compensation coefficient is used as the actual compensation amount. The corresponding protective gear target position is compensated according to the actual compensation amount. For the avoidance protection zone, the corresponding protective gear target position is compensated according to the normal morphology compensation amount, and the distance between the compensated protective gear target surface and the patient's knee surface along the local surface normal is not less than the minimum safety gap.

9. The method for segmenting the fitting area of ​​the patellar injury emergency robot brace according to claim 6, characterized in that, It also includes the following steps: Acquire three-dimensional point cloud data of the knee and muscle activation index at multiple detection times; wherein the knee joint posture change at each detection time is not greater than the posture change threshold. Then, the three-dimensional point cloud data of the knee at each detection time is surface registered with the reference three-dimensional surface model of the knee to obtain the actual normal change of each surface point. Based on the muscle activation index and the muscle activation-normal deformation relationship, the predicted normal change is obtained. Then, the difference between the actual normal change and the predicted normal change is calculated, and the difference is used as the residual deformation. Based on the residual deformation of the same surface point at multiple detection times, the region of continuous morphological change is determined. Then, the reference knee three-dimensional surface model is updated based on the region of continuous morphological change, and the fusion region label is updated based on the updated reference knee three-dimensional surface model.

10. The method for segmenting the fitting area of ​​the patellar injury emergency robot brace according to claim 9, characterized in that, The method of determining regions of continuous morphological change based on the residual deformation of the same surface point at multiple detection times includes: Set thresholds for residual deformation, consecutive detection count, continuous region area, and single update volume; When the absolute value of the residual deformation of the same surface point at consecutive detection times is greater than or equal to the residual deformation threshold, the number of consecutive detection times is not less than the consecutive detection count threshold, and the sign of the residual deformation remains consistent, the surface point is determined to be a point of continuous morphological change. Based on the adjacency relationship of surface points in the reference knee three-dimensional surface model, the points of continuous morphological change are clustered, and the clustered regions with an area not less than the continuous region area threshold are determined as regions of continuous morphological change. The reference knee 3D surface model is updated along the local surface normal based on the residual deformation of each surface point within the continuous morphological change region, and the single position update amount of each surface point is not greater than the upper limit of the single update amount, and the surface points whose fusion region label was the avoidance protection zone before the update retain the avoidance protection zone fusion region label after the update.

Images