CN105852971A - Registration navigation method based on skeleton three-dimensional point cloud - Google Patents

Abstract

The invention relates to a registration navigation method based on a skeleton three-dimensional point cloud. The method is characterized by comprising the steps that an X-Y-Z three-dimensional point cloud virtual coordinate system is constructed based on a DICOM data image; meanwhile, equilibrium data collection of the two sides of a skeleton is achieved by a robot through probes, and an X-Y-Z three-dimensional point cloud world coordinate system is established; dynamic rapid registration navigation calculation which comprises rigid registration acceleration and flexible registration acceleration is conducted on the X-Y-Z three-dimensional point cloud virtual coordinate system and the X-Y-Z three-dimensional point cloud world coordinate system; finally, correction is conducted on parameters such as navigation positions and angles. Thus, the robot can achieve X-Y-Z three-dimensional point cloud dynamic registration navigation, the problems that existing registration accuracy is not ideal, and more rays are exposed are solved, the navigation accuracy and efficiency are improved, the registration navigation accuracy of the orthopaedic robot is smaller than or equal to 0.5 mm, ray exposure can be avoided, and the navigation accuracy can be further improved. The registration navigation method is achieved by cooperating with software programming, and the requirement for automatically implementing adjustment is met.

Description

Registration air navigation aid based on skeleton three-dimensional point cloud

Technical field

The present invention relates to a kind of robot operating air navigation aid, particularly relate to a kind of based on skeleton three-dimensional point cloud Registration air navigation aid.

Background technology

From the point of view of development with regard to existing robot for orthopaedic surgery, in addition to robot self structure factor, it is important that A reason be exactly that airmanship remains at some problems, this be due to:

1, organization structure of skeleton is complicated, lacks reliable Anatomical orientation mark.

The profile of vertebrae is irregular, and internal structure is the most more complicated, and it is soft that vertebral surface exists muscle periosteum etc. Tissue stops, is affected by factors such as hemorrhage and range of exposure, it is difficult to be accurately positioned, current orthopaedics hands in addition Art robot is the most not ideal enough to the operation of satisfied high accuracy, as pedicle of cervical vertebra nail insert, nerve decompression etc., shadow Ring the clinical practice of robot for orthopaedic surgery technology.

2, operation range of exposure is limited.

Orthopaedics especially minimally invasive spine surgical is for avoiding too much wound, and operating field is the least, appears region limited, Add accuracy registration difficulty.

3, surgical target vertebra fine motion, impact location and performance accuracy.

Owing to people's vertebrae small volume, bone amount are limited, thereby increases and it is possible to there is the factors such as osteoporosis, be difficult to do Make its holding static to carrying out " rigidly fixing " with sturdy steel nail, impact registration navigation accuracy.

4, picture quality is not ideal enough affects navigation accuracy.

Navigation at present mostly relies on X-Ray image and guides, and not only picture quality is not ideal enough, and precision is difficult to continue Continuous raising, the ray that also add patient and medical personnel exposes.

Meanwhile, existing air navigation aid and end are as follows:

One, X-Ray relies on registration airmanship.

1, X-Ray two-dimensional imaging navigation in art.

Mazor company of Israel in 2006 research and development SpineAssist spinal operation robot, registration technique uses 2D/3D (preoperative 3D iconography and C-arm 2D iconography Registration of Measuring Data in art).After exposure of surgical field, peace Dress patient's tracer, adjusts camera position, registers and calibrate intelligence tool, installs C-arm spike simultaneously Device also connects navigation system.C-arm perspective gathers targeted vertebra anteroposterior position, position, side and the two dimensional image of Double oblique And by image transmitting to navigation system, image transmitting is complete can be used, and need not carry out artificial Point matching Dough-making powder mates.Then perform the operation under two-dimensional virtual is image-guided.

This robot has passed through FDA (FDA (Food and Drug Adminstration)) and CE certification, is more applicable to Fix in Thoracolumbar disk pedicle screw and translaminar facet screw, many in the U.S., Europe, the Middle East etc. Individual countries and regions are widely used.

Meanwhile, document is reported, global its position error precision of thousands of example spinal operations is 1mm.Sukovich Deng application Spine Assist system minimal invasion internal fixation treatment 14 example patient, success rate reaches 93%.96% The position of pedicle screw with plan into nail deviation within 1mm.It is many that Birkenmaier etc. report Europe A kind of new Wicresoft's fusion method percutaneous diagonal fusion is implemented by this system in center.

Although this system improves the accuracy putting nail, but this technology having picture quality poor, operation essence Degree has certain limitations, and corrects longer with the registration time, and framework stationary positioned mode increases operation wound, Parallel-connection structure design limit working range itself, and increase the shortcomings such as ray exposure.

Korea S in 2010 illustrates SPINEBOT V2 spinal operation robot, although named SPINEBOT V2, but but it is the spinal navigation auxiliary robot of a brand-new design, and it is to operating theater instruments and patient position Location also change by being had an X-rayed realization continuously by biplane.Report that this system is inserted for pedicle screw, Range error is (1.38 ± 0.21) mm.

2, three-dimensional C-arm (ISO-C 3D) imaging based navigation in art.

Simens company in 1999 develops three-dimensional C-arm ISO-C 3D in First art in the world, can be Operation rotates 190 ° continuously, gathers 100 width digital point pictures, be automatically performed three-dimensional reconstruction and registration, Immediately guide operation, need not manually carry out a photograph and close dough-making powder according to closing.

ISO-C 3D precision is 0.5 ± 0.48mm.This airmanship realizes accurately with patient anatomy automatically Correspondence, is the most navigation of current domestic application.ISO-C 3D rendering quality is with preoperative CT three-dimensional navigation still Certain gap, limited scanning volume is had at most to relate to 4 vertebras, it is meant that the spinal operation of multi-segmental needs Multi collect image, too increases ray and exposes.

3, O-arm, CT imaging based navigation in art.

Utilize O-arm or CT in art can obtain the 3-D view of better quality, be directly inputted to navigational computer Middle Auto-matching, is current state-of-the-art navigation system, applies in recent years and gradually increase.

Two, non-X-Ray relies on registration airmanship.

1, based on audio signal location technology.

Germany's Voxel-Man invention acquires the audio signal of cutting in art, to instructing operation, and Virtual dentistry, cerebral surgery operation training system are simulated, make operator same during virtual operation Step hears sound.

But, for different osseous tissues, the characteristic audio signal under different parameters is complicated, it is difficult to finely distinguish, Spinal surgery is applied limited.

2, image registration airmanship.

The DICOM file utilizing preoperative CT scan to obtain is input to computer, sets up virtual coordinate system, art In can set up world coordinate system according to image geometry feature or density feature, then by laggard for two co-registration of coordinate systems used Row navigation.

Registrating by finding on object subject to registration the shortest surface distance between two stack features, feature is pressed Its source can be divided into additional feature and internal characteristics.The former registration accuracy is high, but implants because of needs and invade Property label, clinical practice is restricted；The latter uses the characteristic point that vertebral surface is easy to identify, in reality In application, error is bigger.

At present image registration includes a registration, surface adjustment and point cloud registering:

1. registration is put.

Robot obtains after anatomic landmark point coordinate data, with the location comparison of respective point in preoperative image adjusting Whole image, this process is referred to as some registration, has simple and quick feature, but its shortcoming is it is also obvious that cure It is relatively big that green hand's work takes difficulty a little, needs the long period, and CT scan thickness is big, causes vertebra and navigation The anatomic landmark rebuild on CT can not registrate completely, and in some registration process, soft tissue layer may hinder probe With contacting of bone, and osteoporosis can cause the tip of probe to be absorbed in bone surface, affects anatomic landmark point coordinates Data acquisition and registration.During additionally, bone anatomic landmark is inconspicuous or variation is bigger, it is also possible to cause dissection The error check of position and cause bigger registration error.

2. surface adjustment.

Choosing more multiple point in vertebral surface allows computer position, and this process is referred to as surface adjustment.Owing to selecting Point more, surface adjustment precision improve but hour of log-on is relatively long.

In sum, domestic and international existing robot for orthopaedic surgery airmanship, though X-Ray rely on or , all there is the bottleneck that navigation accuracy is not ideal enough, dynamic registration is difficult and ask in non-X-Ray dependent form airmanship Topic.

Summary of the invention

For solving above-mentioned technical problem, it is an object of the invention to provide a kind of based on skeleton three-dimensional point cloud join Quasi-air navigation aid.

The registration air navigation aid based on skeleton three-dimensional point cloud of the present invention, it comprises the following steps:

Step one, based on DICOM data image, constitutes X, Y, Z three-dimensional point cloud virtual coordinate system.

Step 2, sets up X, Y, Z three-dimensional point cloud world coordinate system.

Step 3, at X, Y, Z three-dimensional point cloud virtual coordinate system and the dynamic rapid registering of world coordinate system Navigation calculates, and accelerates to accelerate with flexible registration including for Rigid Registration.

Step 4, is modified parameters such as the position navigated and angles.

Further, above-mentioned registration air navigation aid based on skeleton three-dimensional point cloud, wherein, described step In one, by DICOM data, to vertebra profile rapid extraction, i.e. by spiral CT machine, to osseous tissue Scanning, uses two dimensional image dividing method, extracts the vertebra profile in CT image, and extraction can be measured by bilateral Structure both sides vertebral surface coordinate points data, take the set of its line midpoint, set up X, Y, Z three-dimensional midpoint Cloud virtual coordinate system.

Further, above-mentioned registration air navigation aid based on skeleton three-dimensional point cloud, wherein, described in build The enforcement step of vertical X, Y, Z three-dimensional point cloud virtual coordinate system is,

A1, to key figure layer, carries out segmentation and extracts with contour line；

A2, extracts the contour line of neighbor map layer；

A3, travels through all sequences image, obtains overall profile data；

A4, screening target skeleton such as vertebra profile on can bilateral measurement structure such as spinous process, articular process, calculating Obtain vertebra point cloud coordinate data.

Further, above-mentioned registration air navigation aid based on skeleton three-dimensional point cloud, wherein, described two Dimension image partition method include domain method, region-growing method, edge detection method, method based on fuzzy partition, One or more in drivewheel skeleton pattern extraction method combine.

Further, above-mentioned registration air navigation aid based on skeleton three-dimensional point cloud, wherein, described step In rapid two, by probe at the vertebra rear structure bilateral such as spinous process, articular process, carry out vertebra coordinate data and adopt Collection, takes the midpoint of 2 lines.

Further, above-mentioned registration air navigation aid based on skeleton three-dimensional point cloud, wherein, described pin Accelerating Rigid Registration, use Revised ICP algorithm, processing procedure is,

B1, rough registration, obtain characteristic of correspondence point；

B2, initializes, and the rotation transition matrix R peace transfer that registration based on point calculates between corresponding point is changed Matrix T, order

$s=1,0\≤\mathrm{\ω}\≤1,{\mathrm{\σ}}^2=\frac{1}{DNM}\underset{n=1}{\overset{N}{\Σ}}\underset{m=1}{\overset{M}{\Σ}}{||x_n-y_m||}^2;$

B3, utilizes fast Gaussian transform, calculates matrix P1, P^T1, PX, then calculate R, t, s and σ²Parameter Value, obtains transforming function transformation function,

T (Y)=sYR^T+1t^T；

B4, is judged whether to reach convergence, without returning to step 2. iteration by setting probability P.

Further, above-mentioned registration air navigation aid based on skeleton three-dimensional point cloud, wherein, described pin Accelerating flexibility registration, use the tracking method for registering theoretical based on continuity, the process of acceleration is,

C1, rough registration, obtain characteristic of correspondence point, initialize other parameter, order

${\mathrm{\σ}}^2=\frac{1}{DNM}\underset{n=1}{\overset{N}{\Σ}}\underset{m=1}{\overset{M}{\Σ}}{||x_n-y_m||}^2,0\≤\mathrm{\ω}\≤1,\mathrm{\β}>0;$

C2, utilizes fast Gaussian transform to calculate matrix P1, P^T1, PX, by low-rank matrix method of approximation, Approximate matrix G；

C3, accelerates to solve W=(G+ λ σ according to low-rank matrix method of approximation²d^-1(P1))-¹(d^-1(P1) PX-Y), and According to N_p=1^TP1 obtains N_PValue；

C4, obtains transforming function transformation function, T=T (Y, W)=Y+GW；

C5, is judged whether to reach convergence by setting probability P, otherwise returns step 2 and calculate.

Yet further, above-mentioned registration air navigation aid based on skeleton three-dimensional point cloud, wherein, described step In rapid four, use following steps,

D1, establishes ambiguous model, obtains the uncertain region of targeted vertebra fine motion；

D2, the frequency response template of structure uncertainty plant, selection reference object, select a class frequency point, According to targeted vertebra frequency corresponding data at selected Frequency point, generate a corresponding group objects template；

D3, calculates, draws border, according to object template and navigation request, on Nicholas figure, generates Each plan boundary of reference object open-loop frequency response, the common factor taking different boundary obtains compound frequency domain border；

D4, designs reforming process, on Nicholas, adjusts the open loop frequency response curve of reference object, Obtain SERVO CONTROL rule, according to the frequency response requirement of closed loop system, adjust prefilter.

By such scheme, the present invention at least has the advantage that

1, robot can be allowed to realize the navigation of X, Y, Z three-dimensional point cloud dynamic registration, solve registration at present The difficult problem that precision is undesirable, improves navigation accuracy and efficiency, makes skeleton robot registrate navigation accuracy≤0.5mm, It is avoided that ray exposes.

2, robot can be allowed to pass through probe, it is achieved zero visual area bilateral equalization data gathers, and improves further and leads Boat accuracy.

3, software programming can be coordinated to be realized the method that the present invention provides, meet automatization and implement to adjust Whole needs.

4, provide the parameters revision such as effective navigation position and angle, optimize navigation accuracy.

Described above is only the general introduction of technical solution of the present invention, in order to better understand the technology of the present invention Means, and can being practiced according to the content of description, below with presently preferred embodiments of the present invention and coordinate attached After figure describes in detail such as.

Accompanying drawing explanation

Fig. 1 is lumbar X, Y, Z three-dimensional point cloud distribution schematic diagram.

Fig. 2 is for setting up the schematic diagram of bone surface coordinate points.

Detailed description of the invention

Below in conjunction with the accompanying drawings and embodiment, the detailed description of the invention of the present invention is described in further detail.With Lower embodiment is used for illustrating the present invention, but is not limited to the scope of the present invention.

Registration air navigation aid based on skeleton three-dimensional point cloud, its unusual part is to comprise the following steps:

First, as it is shown in figure 1, based on DICOM data image, constitute X, Y, Z three-dimensional point cloud Virtual coordinate system, can pass through DICOM data, to vertebra profile rapid extraction.Specifically, by double Source Flash (SOMATOM Definition Flash) 64 row's spiral CT machines, scan osseous tissue, use two Dimension image partition method, extracts the vertebra profile in CT image, and extraction can bilateral measurement structure both sides vertebra Surface coordinate point data, takes the set of its line midpoint, sets up X, Y, Z three-dimensional point cloud virtual coordinate system. Meanwhile, two dimensional image dividing method include domain method, region-growing method, edge detection method, based on fuzzy partition Method, one or more in drivewheel skeleton pattern extraction method combine.

From the point of view of further, the enforcement step setting up X, Y, Z three-dimensional point cloud virtual coordinate system is as follows:

A1, to key figure layer, carries out segmentation and extracts with contour line.

A2, extracts the contour line of neighbor map layer.

A3, travels through all sequences image, obtains overall profile data.

A4, screening profile on spinous process, joint highlight bilateral can measurement structure, calculate obtain vertebra midpoint Cloud coordinate data.

During this period, in order to provide preferably data acquisition, sweep parameter is as follows: voltage 120kV, electric current 140-156mA, DFOV12.0Cm, Scan slice thickness 0.625mm, Pitch0.8, rebuild thickness 0.625 Mm, its surface resolution is 512 × 512, and sweep spacing is 0.625mm.

Afterwards, X, Y, Z three-dimensional point cloud world coordinate system is set up.Reality is implemented when, can lead to Cross probe at the vertebra rear structure bilateral such as spinous process, articular process, carry out vertebra coordinate data collection, take 2 points The midpoint of line.

As in figure 2 it is shown, setting up bone surface coordinate points is F1, F2, the midpoint of its line is P, actual acquisition Coordinate points be P1, P2, such P1 F1, P2 F2 just there occurs measurement error.Control probe measurement Pressure so that it is bilateral keep as, make probe thrust degree of depth F1P1 and the F2P2 mono-of bilateral bone surface Sample, i.e. F1P1=F2P2, and the midpoint of F1F2 line and P1P2 line is still P, the two midpoint is sat Mark data are identical.Thus eliminate vertebra coordinate points measurement error, improve the measurement essence of vertebra coordinate data Degree, sets up for high-precision three-dimensional point cloud world coordinate system and provides the foundation.

Thus, from the point of view of actual enforcement, the method using the present invention, zero visual area probe bilateral can be realized equal Weighing apparatus data acquisition technology, probe can be in vertebra rear structure non-surgical (i.e. zero art such as spinous process, articular process Wild) percutaneous measures, and expands the acquisition range of vertebra data.Additionally, due to probe be spinous process, The vertebra rear structures such as articular process gather vertebra coordinate data, avoid PELD lateral surgical sclerotin completely and lack The region damaged, solves spinal column dypass Minimally Invasive Surgery coordinate data and gathers a difficult problem, can be at various vertebral column minimally invasives Art formula is well set up X, Y, Z three-dimensional point cloud world coordinate system.

Afterwards, lead at X, Y, Z three-dimensional point cloud virtual coordinate system and the dynamic rapid registering of world coordinate system Boat calculates, and accelerates to accelerate with flexible registration including for Rigid Registration.

In point cloud registration Algorithm, ICP algorithm is widely used because of simple, but it is prone to be absorbed in office Portion's maximum.ICP algorithm heavy dependence initial registration position, it requires two point clouds (each point cloud bag Include multiple spinous process, the sub-point cloud of articular process) initial position must be enough near, and when exist noise spot, May cause during exterior point registrating unsuccessfully.

In order to overcome ICP algorithm for the limitation of initial position, and reduce the complexity of calculating, the present invention A kind of improvement ICP method for registering based on probability is proposed, to solve for noise spot, exterior point or missing point Robustness problem and computational efficiency problem.

Specifically, in the present invention as, the registration of two point cloud set is regarded a kind of Multilayer networks Problem, one of them point cloud set X_N×D=(x₁,…x_N)^TRegard data point as, in another Point converges conjunction Y_N×D=(y₁,…y_M)^TRegard the barycenter of gauss hybrid models as, then intend using maximum likelihood Method of estimation registrates barycenter and the data point set of gauss hybrid models.

Registrating for rigidity point cloud, we define gauss hybrid models centroid transformation vector T and are:

T(y_m；R, t, s)=sRy_m+t (1)

In above formula, as long as parameter R can be obtained, t, s, then Rigid Registration problem is just readily solved.Firstly the need of Determining the object function needed for registration, point cloud data collection Y is as the center of gravity of gauss hybrid models, and midpoint is converged X is as the data point formed by gauss hybrid models, then corresponding gauss hybrid models probability density function is:

$p\left(x\right)=\underset{m=1}{\overset{M\mathrm{+1}}{\Σ}}P\left(m\right)p\left(x\right|m)---\left(2\right)$

It addition, in order to noise spot and exterior point present in point cloud data are described, it is simple to follow-up data processes, The present invention will increase extra being uniformly distributed in mixed modelThen mix probability density Model is:

$p\left(x\right)=\mathrm{\ω}\frac{1}{N}+(1-\mathrm{\ω}\underset{m=1}{\overset{M}{\Σ}}\frac{1}{M}p(x|m\left)\right)---\left(3\right)$

According to the conversion vector T formula of rigidity point cloud registration, Rigid Registration object function can be write as shown in following formula:

$Q(R,t,s)=\frac{1}{2{\mathrm{\σ}}^2}\underset{m,n=1}{\overset{M+N}{\Σ}}{P}^{old}\left(m\right|x_n){||x_n-{\mathrm{sRy}}_m-t||}^2+\frac{N_{p}D}{2}{\mathrm{log\σ}}^2---\left(4\right)$

Wherein, σ²For identical variance.

It follows that need to ask for the minimum of above-mentioned object function, the present invention intends providing following dynamic tracking certainly Dynamic registration Algorithm solves:

The first step: rough registration, chooses or by hand by being simply calculated a small amount of characteristic of correspondence point；

Second step, initializes: method for registering based on point calculates the rotation transition matrix R peace between corresponding point Move transition matrix T；Order

$s=1,0\≤\mathrm{\ω}\≤1,{\mathrm{\σ}}^2=\frac{1}{DNM}\underset{n=1}{\overset{N}{\Σ}}\underset{m=1}{\overset{M}{\Σ}}{||x_n-y_m||}^2;$

3rd step: utilize ICP optimized algorithm to solve the value determining parameters；Obtain transforming function transformation function:

T (Y)=sYR^T+1t^T；

4th step: judged whether to reach convergence, without returning to the 3rd step iteration by probability P.

From the point of view of enforcement, in rigidity point cloud registration process, it is equivalent to be to make object function minimize The process changed.During this calculates, during particularly solving object function partial derivative, it is directed to Arrive gauss hybrid models, also relate to the calculating of multiple Gaussian index function summation.Owing to M high This mixed model center of mass point and N number of data point have both participated in calculating, the calculating of the exponential term therefore directly resulted in Complexity is O (MN), when M and N data are bigger, allows for computation complexity bigger, and this will be not easy to Application.In order to reduce the computation complexity in point cloud registration process, the present invention intends at the minimization of object function During combine fast Gaussian transform to improve computational efficiency.

Specifically, accelerating for Rigid Registration, use Revised ICP algorithm, processing procedure is,

The first step: rough registration, chooses or by hand by being simply calculated a small amount of characteristic of correspondence point；

Second step: initialize: method for registering based on point calculates the rotation transition matrix R peace between corresponding point Move transition matrix T；Order

$s=1,0\≤\mathrm{\ω}\≤1,{\mathrm{\σ}}^2=\frac{1}{DNM}\underset{n=1}{\overset{N}{\Σ}}\underset{m=1}{\overset{M}{\Σ}}{||x_n-y_m||}^2;$

3rd step: first with fast Gaussian transform to calculate matrix P1 when utilizing optimized algorithm to calculate, P^T1, PX, And then calculate R, t, s and σ again²Isoparametric value, obtains transforming function transformation function:

T (Y)=sYR^T+1t^T；

4th step: judged whether to reach convergence, without returning to second step iteration by probability P.

From the point of view of flexibility registration acceleration, due to non-intellectual and the noise spot that may be present of non-rigid transformation And missing point so that registration process is likely to occur ill-conditioning problem.In order to process problem in this case, The present invention proposes a kind of thought being again based on probability, and application motion continuity is theoretical, by adjusting displacement Field adds the constraint of sharp movement continuity, makes the barycenter of gauss hybrid models move to data point consistently and complete Alignment.

The when of enforcement, can be that initial position adds a displacement function flexible transform definition, this can make Obtain registration process and become coherent, be denoted as: T (Y, v)=Y+v (Y).So, registration problems is converted to one The problem how estimating displacement function v (Y), the most how regularization v makes function T smooth.

First, define regularization term φ (v), as follows:

$\mathrm{\φ}\left(v\right)={||v||}_{H^m}^2={\∫}_{R^D}\frac{{\left|\tilde{v}\left(s\right)\right|}^2}{\tilde{G}\left(s\right)}ds---\left(5\right)$

Wherein G is the gaussian kernel function chosen,It it is the Fourier transform function of G.

Formula (5) is joined in Rigid Registration object function the object function under available flexibility registration:

$Q(v,{\mathrm{\σ}}^2)=\frac{1}{2{\mathrm{\σ}}^2}\underset{m,n=1}{\overset{M,N}{\Σ}}P\left(m\right|x_n){||x_n-(y_m+v(y_m\left)\right)||}^2+\frac{N_{p}D}{2}{\mathrm{log\σ}}^2+\frac{\mathrm{\λ}}{2}\mathrm{\φ}\left(v\right)---\left(6\right)$

Can be by v (y in above formula_m) it is defined as form:

R (7)

In order to seek the minima of object function, then can utilize the calculus of variations ask object function aboutPartial derivative, And the result zero setting that will obtain.Therefrom can obtain all parameter expressions of flexible registration Algorithm W,σ²,N_p, T, thus the correspondent transform relation between two non-rigid point cloud set is i.e. determined.

From the point of view of the better embodiment of the present invention, in flexible point cloud registration process, in addition it is also necessary to use Optimized algorithm is iterated and asks optimal value to above-mentioned parameter, and step is as follows:

The first step: initialize.Order

Second step: structural matrix G: its component is as follows,

$g_{i,j}=\exp (-\frac{1}{2{\mathrm{\β}}^2}{||y_i-y_j||}^2).$

3rd step: utilize EM algorithm optimization, first calculating matrix P, wherein,

$P\left(m\right|x_n)=\frac{\exp \{-\frac{{||x_n-(y_m+GW)||}^2}{2{\mathrm{\σ}}^2}\}}{\mathrm{\ω}{\left(2{\mathrm{\π\σ}}^2\right)}^{D/2}\frac{M}{N}+(1-\mathrm{\ω})\underset{m=1}{\overset{M}{\Σ}}\exp \{-\frac{{||x_n-(y_m+GW)||}^2}{2{\mathrm{\σ}}^2}\}}$

Then solve and obtain W, σ²,N_p,T。

4th step: obtain transforming function transformation function: T=T (Y, W)=Y+GW.

5th step: judged whether to reach convergence by probability P, otherwise returns the 3rd step and calculates.

During flexible point cloud registration Algorithm optimizes, a system of linear equations is had to need to be solved out, Use during solving this equation and arrived direct matrix in verse.It is near that the present invention proposes a kind of utilization low-rank matrix The method accelerating the computational efficiency of non-rigid registration like method.

It is described as follows for flexibility registration accelerating algorithm flow process:

The first step: rough registration, chooses or by hand by being simply calculated a small amount of characteristic of correspondence point, initially Change other parameter, order

${\mathrm{\σ}}^2=\frac{1}{DNM}\underset{n=1}{\overset{N}{\Σ}}\underset{m=1}{\overset{M}{\Σ}}{||x_n-y_m||}^2,0\≤\mathrm{\ω}\≤1,\mathrm{\β}>0.$

Second step: utilize fast Gaussian transform to calculate matrix P1, P^T1, PX, and utilize low-rank matrix near Approximate matrix G is carried out like method.

3rd step: accelerate to solve according to low-rank matrix method of approximation,

W=(G+ λ σ²d^-1(P1))^-1(d^-1(P1) PX-Y), and according to N_p=1^TP1 quickly obtains N_P's Value.

4th step: obtain transforming function transformation function, T=T (Y, W)=Y+GW.

5th step: judged whether to reach convergence by probability P, otherwise returns second step and calculates.

Thus, the present invention is directed to rigidity and flexible registration situation proposes a kind of fast Gaussian transform algorithm respectively The accelerating algorithm combined with low-rank matrix method of approximation, on the one hand this algorithm reduces the computation complexity of algorithm, On the other hand the rough registration link by increasing can be that algorithm provides a more satisfactory initial value, thus carries The speed of high registration.

Finally, navigation angle is modified, in order to the navigation accuracy adverse effect causing vertebra fine motion is entered Row processes.Thus, a robust controller based on Quantitative Feedback can be set up, to improve the product of servosystem Matter, improves navigation accuracy.Specifically, following steps can be used:

The first step, establishes ambiguous model.When reality is implemented, mesh can be obtained by experience or off-line identification The uncertain region of mark vertebra fine motion.

Second step, the frequency response template of structure uncertainty plant, selection reference object, selecting one group has generation The Frequency point of table, according to targeted vertebra frequency corresponding data at selected Frequency point, generates corresponding one Group objects template.

3rd step, calculates, draws border, according to object template and navigation request, on Nicholas figure, Generating each plan boundary of reference object open-loop frequency response, the common factor taking different boundary obtains compound frequency domain Border.

4th step, designs reforming process, and on Nicholas, the open loop frequency domain response adjusting reference object is bent Line so that it is meet navigation accuracy requirement, thus obtain SERVO CONTROL rule, according to the frequency response of closed loop system Requirement, adjusts prefilter.

By above-mentioned character express and combine accompanying drawing it can be seen that use after the present invention, gather around and have the following advantages:

1, robot can be allowed to realize the navigation of X, Y, Z three-dimensional point cloud dynamic registration, solve registration at present The difficult problem that precision is undesirable, improves navigation accuracy and efficiency, makes spinal column robot registrate navigation accuracy≤0.5mm, It is avoided that ray exposes.

2, robot can be allowed to pass through probe, it is achieved zero visual area bilateral equalization data gathers, and improves further and leads Boat accuracy.

3, software programming can be coordinated to be realized the method that the present invention provides, meet automatization and implement to adjust Whole needs.

4, provide the parameters revision such as effective navigation position and angle, optimize navigation accuracy.

The above is only the preferred embodiment of the present invention, is not limited to the present invention, it is noted that For those skilled in the art, on the premise of without departing from the technology of the present invention principle, also Can make some improvement and modification, these improve and modification also should be regarded as protection scope of the present invention.

Claims (8)

1. registration air navigation aid based on skeleton three-dimensional point cloud, it is characterised in that comprise the following steps:

Step one, based on DICOM data image, constitutes X, Y, Z three-dimensional point cloud virtual coordinates System；

Step 2, sets up X, Y, Z three-dimensional point cloud world coordinate system；

Step 3, at X, Y, Z three-dimensional point cloud virtual coordinate system and the dynamic rapid registering of world coordinate system Navigation calculates, and accelerates to accelerate with flexible registration including for Rigid Registration；

Step 4, is modified navigation angle.

Registration air navigation aid based on skeleton three-dimensional point cloud the most according to claim 1, its feature exists In: in described step one, by DICOM data, to vertebra profile rapid extraction, i.e. pass through spiral CT Machine, scans osseous tissue, uses two dimensional image dividing method, extracts the vertebra profile in CT image, carries Take can bilateral measurement structure both sides vertebral surface coordinate points data, take its line midpoint set, set up X, Y, Z three-dimensional point cloud virtual coordinate system.

Registration air navigation aid based on skeleton three-dimensional point cloud the most according to claim 2, its feature exists In: the described X of foundation, the enforcement step of Y, Z three-dimensional point cloud virtual coordinate system be,

A1, to key figure layer, carries out segmentation and extracts with contour line；

A2, extracts the contour line of neighbor map layer；

A3, travels through all sequences image, obtains overall profile data；

A4, screening profile on spinous process, joint highlight bilateral can measurement structure, calculate obtain vertebra midpoint Cloud coordinate data.

Registration air navigation aid based on skeleton three-dimensional point cloud the most according to claim 2, its feature exists In: described two dimensional image dividing method includes domain method, region-growing method, edge detection method, divides based on fuzzy One or more in the method cut, drivewheel skeleton pattern extraction method combine.

Registration air navigation aid based on skeleton three-dimensional point cloud the most according to claim 1, its feature exists In: in described step 2, by probe at the vertebra rear structure bilateral such as spinous process, articular process, carry out vertebra Coordinate data gathers, and takes the midpoint of 2 lines.

Registration air navigation aid based on skeleton three-dimensional point cloud the most according to claim 1, its feature exists In: it is described that for Rigid Registration acceleration, employing Revised ICP algorithm, processing procedure is,

B1, rough registration, obtain characteristic of correspondence point；

B2, initializes, and the rotation transition matrix R peace transfer that registration based on point calculates between corresponding point is changed Matrix T, order

$s=1,0\≤\mathrm{\ω}\≤1,{\mathrm{\σ}}^2=\frac{1}{DNM}\underset{n=1}{\overset{N}{\Σ}}\underset{m=1}{\overset{M}{\Σ}}\left|\right|x_n-y_m|{|}^2;$

B3, utilizes fast Gaussian transform, calculates matrix P1, P^T1, PX, then calculate R, t, s and σ²Parameter Value, obtain transforming function transformation function,

T (Y)=sYR^T+1t^T；

B4, is judged whether to reach convergence, without returning to step 2. iteration by setting probability P.

Registration air navigation aid based on skeleton three-dimensional point cloud the most according to claim 1, its feature exists In: described for flexibility registration acceleration, use the tracking method for registering theoretical based on continuity, the mistake of acceleration Cheng Wei,

C1, rough registration, obtain characteristic of correspondence point, initialize other parameter, order

${\mathrm{\σ}}^2=\frac{1}{DNM}\underset{n=1}{\overset{N}{\Σ}}\underset{m=1}{\overset{M}{\Σ}}\left|\right|x_n-y_m|{|}^2,0\≤\mathrm{\ω}\≤1,\mathrm{\β}>0;$

C2, utilizes fast Gaussian transform to calculate matrix P1, P^T1, PX, by low-rank matrix method of approximation, closely Like matrix G；

C3, accelerates to solve according to low-rank matrix method of approximation W=(G+ λ σ²d^-1(P1))^-1(d^-1(P1) PX-Y), and according to N_p=1^TP1 obtains N_P's Value；

C4, obtains transforming function transformation function, T=T (Y, W)=Y+GW；

C5, is judged whether to reach convergence by setting probability P, otherwise returns step 2 and calculate.

Registration air navigation aid based on skeleton three-dimensional point cloud the most according to claim 1, its feature exists In: in described step 4, use following steps,

D1, establishes ambiguous model, obtains the uncertain region of targeted vertebra fine motion；

D2, the frequency response template of structure uncertainty plant, selection reference object, select a class frequency point, According to targeted vertebra frequency corresponding data at selected Frequency point, generate a corresponding group objects template；

D3, calculates, draws border, according to object template and navigation request, on Nicholas figure, generates Each plan boundary of reference object open-loop frequency response, the common factor taking different boundary obtains compound frequency domain border；

D4, designs reforming process, on Nicholas, adjusts the open loop frequency response curve of reference object, Obtain SERVO CONTROL rule, according to the frequency response requirement of closed loop system, adjust prefilter.