Specific embodiment
After embodiments of the present invention are described in detail by way of example below in conjunction with attached drawing, of the invention itsHis features, characteristics, and advantages will be more obvious.
Fig. 1 is that a kind of foot's shoe last model provided in an embodiment of the present invention constructs system schematic.As shown in Fig. 1, this isSystem includes follow shot equipment, processing server, image reconstruction system, reconstruct foot's point cloud model processing system, reconstruct footTriangle grid model processing and characteristic point parameter automatically extract simultaneously labeling system, human foot model positioning transformations system, shoe tree dataLibrary building and management system, foot's characteristic parameter driving standard shoe tree anamorphotic system, reconstruct shoe last model detection system, personal shoesLast carving data base management system and reconstruct shoe last model output system.
Fig. 2 is the image reconstruction system workflow schematic diagram that foot's shoe last model shown in Fig. 1 constructs system.Such as Fig. 2 instituteShow, image reconstruction system workflow setting are as follows:
People places object of reference when level is stood, on foot side, and foot is contacted object of reference, is placed on object of reference naturally,Using follow shot equipment around foot shoot one group of two dimensional image or around foot shoot one section of complete video as far as possible everyChange angle is between 10 degree to 45 degree within one second.It is required that mobile device shoots pixel in 200pdi or more.Mobile device passes through netNetwork will shoot gained image graphic and upload to processing server.It is generally difficult to obtain whole foot's letters by a sub-pictureBreath, needs to obtain the image at each position of people's foot, then whole people's foot 3D point cloud number is obtained after being merged by registration algorithmAccording to.
Gray processing processing is carried out for the image of upload, the specific steps are color is divided into R, tri- Color Channels of G, B are rightIn different colorations, brightness, the pixel of saturation degree indicates white brightness with Y, calculates the brightness value of each pixel, makes brightnessValue and its R, G, B value is corresponding, obtains gray level image.And gray processing processing is equally carried out for video, a frame was taken every one secondImage is as reconstruction model reference picture.
The pixel value for belonging to background parts is zeroed by the threshold value that people's foot and background can be reasonably distinguished by setting(or being maximized) retains the pixel value for belonging to people's foot part.Multiple scale detecting angle point technology is used based on gray level image, quicklyIt extracts characteristics of image angle point and finds corresponding relationship in the image of the overlapping of Same Scene, and different images are carried out by referenceScaling on the basis of object size.
First by forming image resolution pyramid, to determine the characteristic point in each image for changing resolution ratio.It is denoised with SIFT algorithm smoothed image, gray level image is used to the N-dimensional space normal distribution equation of Gaussian blurring function:Wherein σ indicates Image Smoothness, that is, normal distyribution function standard deviation, as σ is incremented by, resolution ratioSuccessively decrease, r is blur radius, and blur radius refers to template elements to the distance of template center.By Gaussian blurring function G(x,y,δ)WithOriginal image I(x,y)The scale-space representation formula of the image: L is obtained using convolution algorithm(x,y,δ)=G(x,y,δ)*I(x,y).Image is doneThe Gaussian Blur of different scale, the method that resolution ratio successively reduces down-sampled i.e. are sampled smoothed image, building Gauss goldWord tower and difference of Gaussian pyramid determine the position for indicating image Harris corner characteristics in pyramidal each layer.
By finding the position of each corner characteristics, the angle intensity is shown, according to angle intensity and predetermined minimum intensity threshold pairThan, point is deleted, retest determines whether sum is more than maximum permissible value, if being more than, increases radius and repetitive process again,Alternately, until Angle Position number equals or falls below maximum permissible value.
By using the method for blur gradients, each corner characteristics distribution orientation is given.First angle point is detected under large scale,Then real characteristic angle point relatively precisely determine under smaller scale.
Because Gaussian kernel is unique linear kernel, that is to say, that other will not be introduced by, which being obscured using Gauss collecting image, makes an uproarSound, therefore selected Gaussian kernel just to construct the scale of image.Harris angle point operator is turned with the first derivative of Gaussian kernelBecome scale space operator, for scale space operator to the analysis of image not by image grayscale level, the influence of contrast variation is fullSufficient translation invariance, scale invariability, Euclid's invariance, the characteristics such as affine-invariant features.
For the more position of precise positioning characteristic point and scale, redundancy is removed by being fitted the method for three-dimensional quadratic functionCharacteristic point and unstable skirt response point improve noise resisting ability to enhance matching stability.
It carries out curve fitting to scale space difference of Gaussian function, original position is obtained plus the offset and scale of fittingObtain the exact position of characteristic point.Wherein fitting function are as follows:It asksIt leads and functional equation is allowed to be equal to 0, obtain extreme point deviation amount expression formula are as follows:Corresponding extreme point expression formula are as follows:When extreme point deviation amountWhen, current signature point position must be changed, in a new location repeatedlyInterpolation, if beyond the number of iterations set or exceeding image boundary range, deletes the point until convergence.While in order to avoidD(x)Functional value is too small and wild effect caused by by noise jamming, | D(x)|≤0.03 extreme value point deletion.
After obtaining characteristic point position, local principal curvatures is acquired by Hessian matrix (Hessian matrix), it is assumed that HessianMatrix expression are as follows:The sum of diagonal of a matrix is Tr(H)Its expression formula are as follows: Tr(H)=Dxx+Dyy=alpha+beta,The determinant of matrix H is Det(H), expression formula Det(H)=DxxDyy-(Dxy)2=α β.Curvature threshold is set as γ, is passed throughJudgementIt is whether true, if set up if retain marginal point, otherwise delete, with this judge principal curvatures whetherUnder the threshold value of setting.KD-tree algorithm acceleration search is used in the process.
The Sobel operator for reusing Canny edge detection algorithm calculates the difference Gx of gradient calculating both horizontally and verticallyAnd Gy, calculate gradient-norm and direction:θ=atan2 (Gy,Gx), gradient angle, θ ∈ [- π, π], by its approximationTo four direction, level is respectively represented, vertical and two diagonals (0 °, 45 °, 90 °, 135 °).It can be with ± i π/8 (i=1,3,5,7) divide, fall in the gradient angle in each region to a particular value, represent one of four direction.
The gradient edge more than one pixel usually drawn is wide, but multiple pixels are wide.Therefore gradient map as is alsoIt is very fuzzy.In order to meet only one accurate point width of edge, edge is refined using non-maximum value suppressing method, helps to protectIt stays local maxima gradient and inhibits every other gradient value, only remain position most sharp keen in change of gradient.Algorithm is as follows:
Compare the gradient intensity of current point and the gradient intensity of positive and negative gradient direction point.If the gradient intensity of current point andIt is maximum that the gradient intensity of other equidirectional points, which compares, retains its value.Otherwise inhibit, that is, be set as 0.
Front we gradient direction approximation to level, vertical and two diagonal line four directions, so each pixel rootIt is compared according to itself direction in one of this four direction, decides whether to retain.
Canny algorithm application bivalve value, i.e. a high threshold values and a low valve valve distinguish edge pixel.If edge pictureVegetarian refreshments gradient value is greater than high threshold values, then is considered as strong edge point.If edge gradient value is less than high threshold values, it is greater than low valve valve,Then it is labeled as weak marginal point.Point less than low valve valve is then suppressed.
The 8 connection field pixels for checking a weak marginal point with hysteresis bounds track algorithm again, as long as there is strong edge point to deposit, then this weak marginal point be considered as really edge remain.
The weak edge of all connections is searched for depth-priority-searching method, if one connection weak edge any one point andThe connection of strong edge point, then retain this weak edge, otherwise inhibit this weak edge.
After obtained edge contour, the stabilising direction of partial structurtes is sought using the resulting image gradient of previous step, is usedThe gradient-norm of pixel and direction in cake chart statistics field are segmented by 10 degree, and the maximum region direction of cake chart represents featureThe principal direction of point, in order to enhance matched robustness, Retention area region is greater than the auxiliary direction of conduct of principal direction 80%.SpySign point copies as the characteristic point of segmentation, and direction value is assigned to the characteristic point after duplication respectively.Finally to discrete gradient sideInterpolation fitting processing is carried out to histogram, acquires more accurately deflection angle value.
Each characteristic point establishes a descriptor, with one group of vector description, describes sub- use in characteristic point scale spaceThe gradient information in 8 directions calculated in 4*4 window, total 4*4*8=128 dimensional vector characterization.The size of each subregion and spyIdentical when sign point direction distribution, i.e., there is N_otc pixel in each region, then distributes the square that side length is N_otc for each subregionShape region is sampled.Bilinear interpolation is used again, is determined and is calculated the required image-region of description, reference axis is rotated to beThe direction of characteristic point to ensure rotational invariance, then the sampled point in field is assigned in corresponding subregion, by subregionInterior gradient value is assigned on 8 directions, calculates its weight, then interpolation calculation 8 direction gradients of each seed point.Finally carry outNormalized counts 128 dimension SIFT local descriptions of each characteristic point, the as feature vector of characteristic point.
Euclidean distance by describing son between two characteristic points of searching is closer to, and the Euclidean between other feature point descriptionsApart from farther away characteristic point between as the candidate feature matching pair two images.By each characteristic point matched in obtained imageCoordinate is indicated with homogeneous coordinates, and homogeneous coordinates are standardized expression.
x1,x2∈R3xn, n is pairing sum, x1,x2For matching a pair of of in two imagesPoint.Standardize x1,x2Homogeneous coordinates after:WhereinuxjAnd uyjRespectively indicate xij, yijAverage value, δxjAnd δyjRespectively indicate xij, yijStandard deviation.It is obtained again by homogeneous coordinates hereSecondary coordinates matrix quantificational expression:
Wherein abcdrfghi indicates that rotation and scaling matrix, lmn indicate translation squareBattle array, pqr indicate that projection matrix, s indicate overall conversion matrix.Obtain x1,x2Two shape Interactive matrix, expression formula are as follows:Calculated by calculating and comparing Euclidean distance two shape Interactive matrix by column between difference,And arrange the Euclidean distance between each column vector by descending descending, threshold value point of cut-off is set, Euclidean distance value is greater thanThe matching of point value is truncated to deletion in threshold value, leaves correctly matching pair.
For three-dimensional point cloud model, indicatrix extraction process is exactly that the data point on model is analyzed and countedIt calculates, finds out characteristic point therein, and connect and compose smooth features curve, uneven, shortage topology between point is issued for data pointThe point cloud data of link information carries out Gauss Discrete Mapping to each point in point cloud data with discrete Gauss Map method,Mapping point is subjected to the coagulation type hierarchical clustering based on distance with K-meam algorithm, cluster result and curvature are analyzed,And selected with parameter of the adaptive iterative process to algorithm, optimal feature point set is obtained, the identification feature comprising beforeThe sharp mutable site characteristic point being difficult in point step.Then use the characteristic curve based on PCA principal component analytical method rawLong algorithm, characteristic point is connected, and obtains the characteristic curve of fairing.PCA Principal Component Analysis Algorithm is finally used, to the data obtained sampleDimensionality reduction compression is carried out, storage memory is reduced.
Fig. 3 is that the reconstruct foot point cloud model processing system workflow of the building system of foot's shoe last model shown in Fig. 1 is shownIt is intended to.As shown in figure 3, reconstruct foot's point cloud model processing system workflow are as follows:
By obtaining having compared with precise positioning, direction, feature vector, the point of descriptor by two-dimensional images matching reconstructCloud threedimensional model.But it generally can not be directly as people's foot 3D model data, at this time by the 3D point cloud data that picture reconstruction obtainsObtained point cloud model cannot be directly used to parameter extraction and reconstructed surface, need to carry out some pretreatments also to reduce model notFoot.
Since the 3D point cloud data of acquisition are inevitable that noise (i.e. outlier) exists, simultaneously because human bodyBetween block and will appear cavity, in addition the flatness of point cloud data is also poor.It needs to simplify point cloud model, and is removedDiscrete point, the processing of smooth and holes filling.
Fragment first is carried out to point cloud model with the method for spatial level subdivision, one is calculated for each fragment and simplifies expression,The point cloud model being simplified.
With string high differentiation, the front and back two o'clock of connecting detection point, the distance of technology intermediate data points to string, by this distanceIt is compared with the franchise value being given to, is then abnormal point if more than franchise value, is deleted, so that density height in point cloud model is removed,The noise in the biggish place of Curvature varying.Weighted median filtering method is used again, eliminates burr, smoothing model.
By calculating the uniform hole boundary of length of each edge, if it exceeds 2 times of equalization points are away from then taking its midpoint to be added toIn hole boundary.Unified boundary direction, hole boundary is all unified for counterclockwise.Inner and outer boundary is judged, to outer boundaryProfile carries out judgement removal.The concavity and convexity for being closed each vertex in hole boundary is different, and angle calcu-lation mode is different, each according to calculatingFilling point coordinate is calculated in the size of angle, and the illegal judgement point fallen in outside perforated is removed, can be obtainedFilling point coordinate.The iterative cycles process, until new filling point cannot be calculated, then holes filling terminates.
Fig. 4 is the processing of reconstruct foot triangle grid model and characteristic point ginseng that foot's shoe last model shown in Fig. 1 constructs systemNumber automatically extracts and labeling system.As shown in figure 4, the processing of reconstruct foot's triangle grid model and characteristic point parameter automatically extract simultaneouslyLabeling system workflow are as follows:
To treated, point cloud model uses Waston algorithm to carry out triangulation and optimization, by point cloud model triangle griddingChange, increases the topological relation between a cloud, especially in deformation process, triangle mesh model while retaining point cloud data againIt has great advantages.Three-dimensional grid sequence compaction is carried out with the deformation of surface technology based on thin plate spline again, passes through MC algorithmCarry out three-dimensionalreconstruction.MC algorithm uses the tri patch of triangle band connection dispersion, reduces EMS memory occupation；Using reduction tri patchQuantity, reduce time and space complexity；Discrete noise region is filtered using connectivity detection, removes impurity；Finally examineHole is measured, and by filling-up hole to model process of refinement.
Fig. 5 is the human foot model positioning transformations working-flow schematic diagram that foot's shoe last model shown in Fig. 1 constructs system.As shown in figure 5, human foot model positioning transformations working-flow are as follows:
Obtain the model that the non-fluidity triangle gridding of foot is constituted, but the acquisition of the non-fluidity triangle grid model of foot typeIt is the two-dimentional foot type image reconstruction gained original based on stereoscopic vision, the standard shoe tree in human foot model and shoe tree database is allDefined in respective local coordinate, coordinate origin, change in coordinate axis direction is different, and spatial position is entanglement, copes with footModel carries out orientation adjustment according to the shoe last model space coordinate benchmark of system, adapts it to this process identifying system.Figure 11(a)-Figure 11 (c) is foot's feature line feature point schematic diagram, specific implementation step are as follows: extracts the foot of foot's triangle grid modelReturn pulley profile identifies waist position inside and outside sole with foot type sole contour line characteristic trend existing in database, inside and outside sole positionWaist most salient point is respectively C1C2, and waist most salient point is respectively B1B2 inside and outside heel region followed by peripheral point is B, half sole firstToe endpoint is A, is then sequentially connected AC1B1BB2C2, forms complete closed curve, the as preliminary contour line of sole.ThenIt is separately connected C1C2, B1B2, B point makees straight line excessively and line segment B1B2 intersects at point D1, adjustment BD1 and line segment B1B2 corner dimension,Make 96 degree of angle B1D1B, extend BD1 and intersect at point D2 with line segment C1C2,.It crosses half sole the first toe terminal A and is line segment BD2 and prolongThe vertical line of long line simultaneously meets at point A1, then line segment A1B is sole center line, and the Euclidean distance between point A1 and point B is soleLength.Foot's broadside lines of foot's triangle grid model is similarly extracted, it is special with foot type broadside lines existing in databaseSign trend, identification heel position of center line curve are l1, and instep position of center line curve is l2, are sequentially connected curvel1A1Bl2.Make point B as coordinate origin, line segment A1B is in x-axis principal direction, curve l1l and the line segment A1B curve connecting and xzPlane is parallel, and line segment B1B2 and line segment C1C2 are parallel with x/y plane, and triangle grid model is moved integrally adjustment orientation, modelCoordinate system is matched with rectangular coordinate system in space, obtains the coordinate value of individual features point.
Half sole width C 1C2 followed by width B1B2M, the long A1B of foot, the instep height (midpoint l2 z are obtained by spatial valueCoordinate value), the first toe height, the data such as ankle height.Thus in standard shoe tree number library, by manually selecting with height, shoesMoney type (sport footwear, playshoes, high-heeled shoes, sandals etc.), gender, the options such as area, then the foot three reconstructed by automatic identificationCharacteristic point characteristic curve data on the grid model of angle select user's foot Model Matching degree with reconstruct according to standard foot last carving differenceHighest shoe tree group, the standard shoe tree that system Auto-matching is most adapted to.Because completely the same shoe tree is not with foot type sizeThe shoe tree of the most comfortable interferes proper motion in order to avoid shoes are tightly attached to people's foot, needs to reserve some surpluses of putting in shoes with afterTolerance, general last carving enclose=bottoms-feel difference.Exist only between foot and shoe tree and puts surplus and dimension difference, it could preferablyIt locks foot and does not have extruding sense of discomfort again.
After selection obtains matched standard shoe tree, although human foot model has pressed shoe tree spatial coordinate system in step beforeOrientation adjustment is carried out, but there are still spatial position dislocation in local location, need to carry out space measurement Data Position to foot last carving modelAlignment, including adjustment foot type heel and shoe tree followed by be aligned, adjustment foot type is integrally evenly distributed in last carving body the right and left, adjustment footBottom touchdown point and last carving bottom land point contact, but the step for during do not change pair of the foot type model integrally between each positionIt should be related to, benchmark is transformed to whole unified rotation translation.
For the human foot model after realignment, the long i.e. line segment A1B length of foot is being intercepted respectively along positive direction of the x-axis (with millimeterFor unit) 25% and 68% point be set as point F1, F2, then choose that the long point abscissa of 68% foot increases 15mm and abscissa is reduced respectivelyThe two o'clock of 20mm is set to point F3, F4, draws one respectively perpendicular to the auxiliary of x-axis in tri- data points of F1F3F4 of interceptionAlignment line and the auxiliary rotational circle for being parallel to xz plane are helped, and in auxiliary line position respectively by foot's triangle grid model, soleContour line and foot broadside lines are divided into four parts, by the step for the contour line that obtains and foot's triangle grid model by notForm four groups with position, the auxiliary line of same position and auxiliary circle formed into a group, the step for secondary alignmentLine is parallel with z-axis and intersects with x-axis, assists the intersection point of the center of circle of rotational circle co-located secondary alignment line and x-axis.
According to the standard shoe tree that matching obtains, and the back height value (in inches) of selection, previous step is setThe secondary alignment line set and auxiliary rotational circle position, carry out profile line segment rotation and translation by certain pivot rule, and rotation is arrangedThe minimum and maximum limit value of gyration.Being located at F3 point rotation angle is α, is β in F4 point rotation angle, rotates angle in F1 pointFor γ, F3 is region 1 along positive direction of the x-axis curved portion, and F3 is region 2 along positive direction of the x-axis intersection along negative direction of the x-axis and F4,F1 is region 3 along positive direction of the x-axis intersection along negative direction of the x-axis and F1, and F1 is region 4 along negative direction of the x-axis part.First by region1 using point F3 as rotary middle point auxiliary rotational circle on rotated counterclockwise by angle α (degree),α∈[2,8].Then region 3 is rotated clockwise into angle beta (degree) on auxiliary rotational circle using point F4 as rotary middle point,
Then by region 4 using point F2 as rotary middle point auxiliary rotational circle on rotated counterclockwise by angle γ (degree),
The region rotated is turned inward into line segmentCarry out translation make different zones line segment between original tie point as close as possible to using the crosspoint between different zones inner curve as point of cut-offCurve is trimmed, the curve being retained in region, then will be smoothly connected between each curved section with second order parameter continuity,Similarly the curved surface in different zones is trimmed and is smoothly connected.
Fig. 6 is foot's characteristic parameter driving standard shoe tree anamorphotic system work that foot's shoe last model shown in Fig. 1 constructs systemMake flow diagram.As shown in fig. 6, foot's characteristic parameter drives standard shoe tree anamorphotic system workflow are as follows:
Horizontal standing human foot model is converted into band with high human foot model, and obtains band with the profile of high human foot modelCurve.Purpose for human foot model deformation is primarily to make foot type and shoe tree camber ratio more consistent, convenient for foot last carving modelAnalysis of Human Comfort and data inspection correction, while the exposed human foot model laid flat being allowed to be adjusted to closer to making when wearing shoesType radian.
Each position degree of enclosing line is imported, band is utilized into minimum two with the contour curve of high human foot model and standard shoe last modelMultiplication and conventional method matched curve, using the contour line of human foot model as the contour line parameter of benchmark adjustment criteria last carving model,One circle contour line is divided into 360 points, wherein closing on 3 points of polar radius by parameter spacing between comparative silhouette line and degree of enclosing lineIt is respectively as follows: ρi+1,ρi,ρi-1, intermediate polar radius value is
If ρiCorresponding polar radius is ρ after fairingi, the triangular apex obtained after fairing is ρ, with minimizing energy method public affairsFormula combination polar radius value seeks the polar radius point after Fair Fitting:Wherein θIn the triangle constituted for 3 points, while meeting least square method:Utilize weight factorControlling curve fairness and Approximation with human foot model parameter, generate final shoe tree control line.
Last surface is generated according to the new shoe tree control line of generation, it will system mouth line and last carving using back arc line and last carving linebackReturn pulley profile is divided into two parts respectively, by adjacent contours connection, forms three four borderline regions.It imports in previous step and adjustsWhole good new degree of enclosing line, is trimmed to three parts for each position degree of enclosing line last carving return pulley profile and last carving lineback.Last carving bottom is used firstContour line and last carving bottom point degree of enclosing line, constitute the last carving bottom surface for having the direction z radian.System mouth line and last carving lineback intersection are chosen againDegree of enclosing line last carving return pulley profile is divided into two parts, then by last carving lineback, back arc line, last carving return pulley profile, system mouth line and shouldThe adjacent contours connection of position degree of enclosing line forms five four borderline regions.Then using in curved surface of last region boundary line and encloseIt spends line and constitutes various pieces curved surface.
Connection based on surface boundary continuity and fairness is carried out to the curved surface that above-mentioned steps obtain, makes adjacent curved surface sideBoundary line all has same points and order, and adjacent curved surface is made to meet second order continuity in boundary, and singular point or extra is not presentInflection point, Curvature varying is smaller between curved surface.Make to be mutually linked between curved surface simultaneously, the curvature portion of extension is repaired along intersecting lensIt cuts.Curved surface ideal not enough for fairing effect, by regulate and control curved surface control point, make control point correspond and uniformlyDistribution, and then the more fairing of the curved surface of last of each region is stitched together.Resulting each position curved-surface structure line is parameterizedMode stores, and convenient artificial change is corrected while reducing calculation amount caused by adjustment next time local curve.
The closing that the curved surface being smoothly connected and last carving bottom surface and system mouth face are stitched into a hypostazation is whole, thus obtainsShoe last model based on foot's picture reconstruction model, and the shoe tree master pattern for adjusting, can be according to choosing different shoesClass, shoes money, by the parameter corresponding relationship between different type standard shoe tree, carry out corresponding tune with height on the shoe last model of reconstructIt is whole.Change for different shoe last styles can recognize the shoe last and last carving status secant profile, generally toe of script shoe last modelDegree of enclosing line divides shoe last model, then identifies that target shoe last model boundary connects contour line, by the last carving body model and target after segmentationShoe last model carries out elastic deformation connection, removes seam using three-dimensional smoothing algorithm, is smoothly connected.The curved surface that will be smoothly connectedThe closing for being stitched into a hypostazation with last carving bottom surface and system mouth face is whole, to obtain the customized shoe-last mould of different shoe last stylesType.
Fig. 7 is the reconstruct shoe last model detection system workflow schematic diagram that foot's shoe last model shown in Fig. 1 constructs system.As shown in fig. 7, reconstruct shoe last model detection system workflow are as follows:
Calculating human foot model measures the distance and angle between the shoe last model corresponding points of building, and maximum detecting distance is arranged,Maximum angle is compared the distance obtained and angle is calculated with the allowable error value of setting, and measurement obtains maximum positive normalDeviation and maximum negative Norma l deviation, evaluate local error distribution consistency degree, if the maximum value being calculated is than the permission of settingIt is worth small, shows that shoe tree generated meets the requirements, otherwise just needs to modify to the shoe tree of generation, be met the requirements until generatingShoe tree.
Category of model integration is loaded into master pattern library and is stored in personal shoe tree database, the pipe based on shoe tree databaseReason mode and data analysis, timely update and safeguard personal shoe tree database, and can be by different brands in shoe tree databaseDifferent model shoes money shoe last model parameter is analyzed with the collocation degree of personal similar shoes money shoe last model parameter, obtains comfort levelReport, is shown by interactive interface, is facilitated user to choose the shoes being more comfortably more suitable for and is carried out.
Fig. 8 is the reconstruct shoe last model output system workflow schematic diagram that foot's shoe last model shown in Fig. 1 constructs system.As shown in figure 8, reconstruct shoe last model output system workflow are as follows:
The foot type shoe last model completed for building can be converted directly into non-fluidity grid model format, utilize 3 D-printingMachine carries out rapid shaping.Thimble can also be added in shoe last and last carving heel point position by system, and model group is combined into entity,Numerical Control Simulation processing is carried out, is exported to meet the file format of CAM processing and obtaining NC numerical control code.It can also be by extracting model wheelProfile obtains the contour line in three-view diagram direction, and output printing or laser engraving are shoe tree snap-gauge and cardboard, for model check and correction.
Fig. 9 is the personal shoe tree data base management system workflow signal that foot's shoe last model shown in Fig. 1 constructs systemFigure.As shown in figure 9, personal shoe tree data base management system workflow are as follows:
Maintenance for personal shoe tree database, main method are as follows: file parameters maintenance is carried out to existing model, on basisWhen foot's reconstruction model and constant basic shoe last model, increases or modification model parameter updates existing according to Rational Parameters analysisThere is model parameter, cancels that original model is old or wrong parameter.It can be carried out by combination decision tree algorithm in filtration parameterFeature selecting and direction selection, filtering are cancelled model parameter, are no longer stored in individual subscriber shoe tree database.Group is used simultaneouslyDecision Tree algorithms are closed, many huge trees are generated to objective attribute target attribute parameter, then basis finds the statistical result of each attributeThe biggish character subset of information content.If the character subset of some often changes, and some character subset difference in changeBe worth very little, then using the character subset infrequently changed as fixed character, more multioperation when increasing every time or modifying model parameterThe character subset group often changed is left in space for, reduces the operational analysis time.
Figure 10 is that foot's shoe last model shown in Fig. 1 constructs the shoe tree database sharing of system and management system workflow is shownIt is intended to.As shown in Figure 10, shoe tree database sharing and management system workflow are as follows:
Building and management for shoe tree database construct shoe tree data base administration by way of parametrization managementSystem.Figure 12 (a)-Figure 12 (b) is shoe tree characteristic curve schematic diagram, construction method are as follows: by manually selecting with height, shoes money type(sport footwear, playshoes, high-heeled shoes, sandals etc.), gender, the options such as area, then the foot's triangle gridding reconstructed by automatic identificationCharacteristic point characteristic curve data (substantially degree of enclosing, the data such as foot length) on model, the standard shoe tree that system Auto-matching is most adapted to.
The wherein foundation of shoe last model database is managed in a manner of race's table, and uses combination decision tree algorithm(Random Forests) carries out feature selecting and constructs effective classifier, by the shoe last with same or similar structure styleWith different footwear standard shoe last models, general shoe tree position is established as male parent, then each parameter is controlled on its basisSystem, which generates, derives from type shoe tree, and classification integration is loaded into master pattern library.
The mode in standard shoe tree model data source has: 1. pairs of existing shoes type demands are arranged and are divided, by different regionsThe last size standard of country is by Data Integration, and according to each positional parameter data are manually entered, it is reasonable that system calculates data automaticallyProperty, corresponding professional parameter is calculated automatically, generates shoe tree in the automatic Parametric modeling procedure algorithm of internal systemNurbs surface model, dimensioning, detection model is normative, obtains the calculating analysis report of Model Matching, category of model integrationIt is loaded into master pattern library；2. picture reconstruction material object shoe tree mode, obtains shoe tree point cloud model, pre-processes and extract feature point featureLine) key parameter is obtained, it is automatic to calculate data reasonability, it is generated in the automatic Parametric modeling procedure algorithm of internal systemShoe tree nurbs surface model, dimensioning, detection model is normative, obtains the calculating analysis report of Model Matching, category of modelIntegration is loaded into master pattern library；3. shooting people foot image, foot's point cloud model is obtained from image reconstruction, point cloud model is convertedFor triangle grid model, characteristic point characteristic curve is pre-processed and extracted, model and feature line feature point are integrally deformed to obtainBand wears model with high simulation, obtains key parameter, automatic to calculate data reasonability, automatically selects most fitting shoes in systemsLast carving, foot last carving position local directed complete set, and adjustment obtains target shoe tree, dimensioning, detection model rule on script standard shoe last modelPlasticity, obtains the calculating analysis report of Model Matching, and category of model integration is loaded into master pattern library；4. by shoe tree cardboard or cardPlate is aligned each cardboard or snap-gauge position, obtains shoe tree side profile, and respectively degree of enclosing profile and sole profile, contour curve is loaded intoParametric modeling system, inputs shoe tree size and regional population, system calculate data reasonability automatically, is calculated automatically correspondingProfessional parameter, in the automatic Parametric modeling procedure algorithm of internal system generate shoe tree nurbs surface model, label rulerVery little, detection model is normative, obtains the calculating analysis report of Model Matching, and category of model integration is loaded into master pattern library；5. byShoes image reconstruction mode in kind obtains shoes point cloud model, and pretreatment measures and inputs corresponding position cladding thickness, in origin cloud mouldCladding thickness is uniformly removed on the basis of type, and the point cloud model of removal cladding thickness is pre-processed and extracts feature point featureLine obtains key parameter, automatic to calculate data reasonability, generates shoes in the automatic Parametric modeling procedure algorithm of internal systemLast carving nurbs surface model, dimensioning, detection model is normative, obtains the calculating analysis report of Model Matching, category of model is wholeIt closes and is loaded into the methods of master pattern library.
In order to enrich shoe tree database data while not increase input step and processing inside, not to same moulding shoe treeWith size model construction, using the scale value for calculating target shoe last model and existing size shoe last model, multidimensional scaling, which calculates, to be repairedChange and numbers storage.The grade of this step put operation is realized by three-dimension varying matrix, keep with stilt degree it is constant in the case whereModel overall dimensions degree of enclosing is allowed to reach scaling standard.Three-dimension varying matrix includes: spin matrix, the geometric transformation square of equal proportionBattle array, translation matrix, whole projection matrix etc..Building for different shoes money shoe last models, using in existing shoes money customized shoe-lastOn the basis of, with different shoes money shoe trees by corresponding relationship, customized shoe-last, corresponding adjustment local size structure and scaling are docked,Style can be adjusted according to standard pattern model splicing.
Dimensioning content includes length of last sole degree, back height, preceding to lift up height, half sole touchdown point followed by touchdown point, footBack height, mouth width degree of uniting, open height of uniting, mouth length of uniting, toe degree of enclosing, sole degree of enclosing, foot waist dimension, instep degree of enclosing, ankle encloseDegree etc..
Foot last carving Analysis of Human Comfort and inspection generate cross-section comparison's number at each position according to the foot last carving model matchedAccording in conjunction with shoe tree comfort level rule, the mean square deviation of otherness, makes analysis between calculating last carving critical size and foot critical sizeWith inspection.
Using Parametric designing, the pass between size and size is established with the method for mathematical relationship or logical relationSystem can shorten the time of modification model.Unified model type is sorted out respectively and is managed.Detection model is normative simultaneously, obtainsThe calculating analysis report of Model Matching facilitates modification and correction, improves shoe tree database data confidence level.
The method that the embodiment of the present invention drives shoe tree parametric reconstruction by using characteristic point passes through controlling feature line and controlThe parameters relationship of point is made to complete the parametric modeling of shoe tree, the control parameter by modifying shoe tree can complete new shoes modelBuilding.And to people's foot, the extracting method of shoes and shoe tree characteristic parameter is studied, using the characteristic parameter Controlling model simplifiedGeneration, and complete the rational inspection of model parameter, the inspection etc. of reconstruction model matching.Tune is stored in a manner of parametrizationIt is generated with shoe last model data, while with driving parameter model, model modification is facilitated to update and analysis management.
It should be noted that above-described embodiment is only used to illustrate structure and its working effect of the invention, and it is not used asIt limits the scope of the invention.One of ordinary skilled in the art is right without prejudice to thinking of the present invention and structureThe adjustment or optimization that above-described embodiment carries out, should regard as the claims in the present invention and be covered.