CN107123161A - A kind of the whole network three-dimensional rebuilding method of contact net zero based on NARF and FPFH - Google Patents
A kind of the whole network three-dimensional rebuilding method of contact net zero based on NARF and FPFH Download PDFInfo
- Publication number
- CN107123161A CN107123161A CN201710445701.XA CN201710445701A CN107123161A CN 107123161 A CN107123161 A CN 107123161A CN 201710445701 A CN201710445701 A CN 201710445701A CN 107123161 A CN107123161 A CN 107123161A
- Authority
- CN
- China
- Prior art keywords
- mrow
- point
- msub
- cloud
- contact net
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computer Graphics (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- Other Investigation Or Analysis Of Materials By Electrical Means (AREA)
Abstract
The invention discloses a kind of the whole network three-dimensional rebuilding method of contact net zero based on NARF and FPFH, comprise the following steps：Contact net initial point cloud is gathered by the detection dolly for being provided with Kinect2.0 depth cameras；Initial point cloud is pre-processed；For contact net cloud data, correspondence key point in two frame point clouds is extracted, the NARF key points that surface-stable but neighborhood depth information change are extracted as border is detected to point cloud chart using normalization alignment radial direction feature NARF；Feature description is carried out to NARF key points using quick point feature histogram FPFH algorithms, the crucial point correspondence between two frame initial point clouds is determined；Point cloud registering is carried out using key point, Mismatching point is rejected by SAC IA；Accuracy registration is carried out using ICP algorithm, the full contact net three-dimensional point cloud model after registration is obtained.The present invention can well set up visualization contact net the whole network model, and the failure that effective detection wherein defect insulator is present.
Description
Technical field
It is particularly a kind of to be based on NARF the present invention relates to electrification railway contact net reconstructing three-dimensional model and detection field
With the FPFH the whole network three-dimensional rebuilding method of contact net zero.
Background technology
Contact net is the important component main frame of electrified high-speed railway electric power system, by contact suspension device,
The parts such as positioner, support device are constituted, and its performance directly affects the speed of service and safety of bullet train, it is ensured that contact net
It is most important for the transmission of high ferro reliable electric energy that parameter meets design specification requirement.Along Railway residing for contact net is severe, multiple
Miscellaneous working environment, causes its Frequent Troubles, just seems most important for centenary design, construction, the detection of state.
Three-dimensional values have been widely used in the various fields such as machinery, aviation, military project at present, but are directed to contact net event
Three-dimensional values research in terms of barrier does not almost have.
It is a kind of to be adapted in the text of innovatory algorithm one of online three-dimensional values provide such a method, i.e., surveyed in conventional phase
Amount technology of profiling adds Stoilov algorithms, it is applied to online three-dimensional values.Transported in traditional algorithm containing evolution and division
Calculate, for the nonlinearity erron of CCD camera, projection light is long, optical interference of surrounding environment etc. relatively it is sensitive, easily occur compared with
Big error.Therefore author utilizes Stoilov algorithms, effectively reduces the solution phase error in prediction scheme algorithmic formula, is greatly enhanced
The progresses of online three-dimensional values.In confirmatory full-scale investigation, the three-dimensional article rebuild based on the stoilov algorithms after improvement
There is body extraordinary fidelity (to refer to Zhong Lijun, Cao Yiping：A kind of innovatory algorithm for being adapted to online three-dimensional values, China swashs
Light, 2009).
In angular thread 3 D detection method and the text of experimental study one based on machine vision, author is according to spiral shell outside triangle
Oneself contour structure of line, carries out the IMAQ of surface information, then to adopting by CCD area array cameras for screw thread first
The image collected is pre-processed, such as rim detection of image enhancement processing, screw thread, extraction screw thread edge feature means；Most
Added afterwards using extracting obtained characteristic threads point and realize externally threaded Three-dimensional Gravity, it is new to be that externally threaded three-dimensional reconstruction is proposed
Feasible program (refers to ten thousand rocs：Angular thread 3 D detection method and experimental study based on machine vision, South China Science ＆ Engineering University,
2012)。
Detected in three-dimensional values technology in train wheel with the text of application study one in maintenance, three-dimensional reconstruction is answered
Car is examined for the train wheel in the railway system, its efficiency is by far above traditional artificial detection means.This method is derived from reverse
Engineering philosophy, the wheel three-dimensional values of population parameter are carried out with reference to computer picture ancillary technique.This method is in railway wheel
Under the premise of inspection criterion, optimal solution is demonstrated, the detection of wheel and maintenance are closely linked, is the railway system
Wheel maintenance (refers to Cheng Hongzhao there is provided new thinking：Three-dimensional values technology is detected in train wheel and ground with the application in maintenance
Study carefully, Southwest Jiaotong University, 2013).
Inspection Technology for Overhead Contact System based on threedimensional model is studied in a text, and each zero of contact net is obtained using optical scanner
The cloud data of part, and will be merged into certain point cloud registration algorithm without measuring the conversion of obtained multi-disc point cloud under visual angle
A complete point cloud data is formed under the same coordinate system, then through curve reestablishing and each part threedimensional model of acquisition is rendered (in detail
See Xu Jianfang：Inspection Technology for Overhead Contact System research based on threedimensional model, Southwest Jiaotong University, 2014).
Generally speaking, two dimensional touch net image, the inspection of two dimensional image are based primarily upon currently in contact net image detection
The image information that survey technology contains is relatively single, the dead angle that there is image detection, it is impossible to quick, accurate to detect that contact net is real-time
Failure.
The content of the invention
The technical problems to be solved by the invention are to provide a kind of the whole network Three-dimensional Gravity of contact net zero based on NARF and FPFH
Construction method, this method pre-processed using three-dimensional point cloud treatment technology to the contact net the whole network cloud data got, registration,
Fusion, obtains threedimensional model, and extracts wherein defect insulator, detects its failure；It can well set up visualization contact net
The whole network model, and the failure that effective detection wherein defect insulator is present.
In order to solve the above technical problems, the technical solution adopted by the present invention is：
A kind of the whole network three-dimensional rebuilding method of contact net zero based on NARF and FPFH, comprises the following steps：
Step 1：Contact net initial point cloud is gathered by the detection dolly for being provided with Kinect2.0 depth cameras, obtained
Initial cloud data be pcd formatted files, it includes some cloud positions, color and depth；
Step 2：Initial point cloud is pre-processed, including a cloud denoising, point cloud segmentation, point cloud compressing, point cloud fusion and
Point cloud derives from；
Step 3：For contact net cloud data, correspondence key point in two frame point clouds is extracted, using normalization alignment radially
Feature NARF, as border is detected, extracts the NARF keys that surface-stable but neighborhood depth information change to point cloud chart
Point, these changes include the principal direction of surface variation coefficient and change；
Step 4：Feature description is carried out to NARF key points using quick point feature histogram FPFH algorithms, at the beginning of determining two frames
Crucial point correspondence between initial point cloud；
Step 5：Point cloud registering is carried out using key point, Mismatching point is rejected by SAC-IA；
Step 6：Accuracy registration is carried out using ICP algorithm, the full contact net three-dimensional point cloud model after registration is obtained.
Further, in addition to step 7：
Step 7：Using point cloud segmentation means, the defect bracket insulator in the contact pessimistic concurrency control after point cloud registering is extracted
Out, obtained insulator surface normal is extracted in estimation, and defect insulator is detected using its normal to a surface information.
Further, the step 3 is specially：
Step 3.1：The structure separation of the big different parts of position correspondence of the point curvature of certain in cloud data, border interest value
I (p) is：
I (p)=I1(p)·I2(p)
In formula：
I1(p) factor, I are estimated for surface change intensity2(p) it is the main direction identification factor；P is boundary point, and n is point p neighbour
Element, σ is space scale, wnFor the corresponding weighted values of neighbors n, γ is the one-dimensional angle of point n correspondences；For contact net curved surface side
Edge point, wnValue is 1, and for other points, wnValue is 1- (1- ρ)3, ρ is p and its peripheral point obtained after principal component analysis
The curvature value arrived.
Further, the step 4 is specially：
Step 4.1：To arbitrfary point P in a cloudqInquire about all neighbor points in its k field；
Step 4.2：To any pair of point P in point P k fieldssAnd PtEstimate normal nsAnd nt, wherein s！=t, wherein one
A fixed u is defined on point, v, w local coordinate systems calculate nsAnd ntBetween deviation
Step 4.3：In the u of step 4.2, in w local coordinate systems, normal n is estimated with one group of angle by vsAnd ntBetween
Deviation：
α=vnt
In formula, | | Pt-Ps| |=d represents point PsAnd PtThe Euclidean distance of point-to-point transmission, then PqAll k fields any two points Ps
And PtRelation pass through (α, β, θ, d) represent；Four all class values of k fields are statistically put into histogram, obtain PqSpy
Levy description；
Step 4.4：For each query point Pq, a tuple α between the point and its neighborhood point is calculated,θ；Calculate
FPFH, redefines the k neighborhoods of each point, and P is calculated using neighbouring SPFHqFinal FPFH, its calculation formula isIn formula, weights omegakRepresent in given metric space, query point PqWith it
Neighbor point PkThe distance between；
Calculate and all put after the FPFH Feature Descriptors of cloud key point in point cloud frame, obtain the required key point correspondence of registration
Relation.
Further, the step 5 point cloud registration is specially：
After two frames contact site cloud corresponding relation is determined, registration is carried out to a cloud, it is divided into rough registration and essence registration；Slightly
Registration uses SAC-IA, after Mismatching point is rejected, and accuracy registration is carried out using algorithm；Obtain complete through a series of point cloud registerings
The three dimensional point cloud of contact net zero.
Further, the step 7 is specially：
Step 7.1：The characteristic vector and characteristic value of one covariance matrix of analysis, and this covariance matrix C is from each
Individual query point PiNeighbor point in construct and obtain, specific covariance matrix is as follows：
In above formula, k is query point PiThe number of neighbor point,It is the three-dimensional barycenter of closest element, λjIt is covariance matrix
J-th of characteristic value,It is j-th of characteristic vector；
Step 7.2：Establish section
For a frame cloud data, its cloud quantity is n, to calculate wherein some query point PiNormal, first set
Put plane equation：
Ax+by+cz+d=0
For carrying out constraint in above formula, when meeting a2+b2+c2When=1, plane parameter a, b, c, d are obtained；Allow point Pi
The quadratic sum numerical value that k point of proximity of surrounding reaches the plan range is minimum, that is, meets：
In formula, diIt is that cloud data concentrates either query point Pi=(xi,yi,zi) to corresponding flat apart from di=| axi+
byi+czi-d|；Extreme value is solved using method of Lagrange multipliers so that e → min, obtain function：
Local derviation is sought in equal sign both sides for d in above formula, and makes partial derivative be zero, obtains：
In formula, order
Barycenter isThen
di=| a Δs xi+bΔyi+cΔzi|
It is right againAsk inclined for a, b, c in middle equal sign both sides
Derivative, is obtained
Plane parameter a, b, c are solved, solution matrix A characteristic value and characteristic vector is converted into；Matrix A is one three section pair
Claim matrix, the solution formula of its characteristic value is：
In a2+b2+c2Under=1 constraints, try to achieveUtilize barycenter
Try to achieve d；
Step 7.3：Adjust normal direction
Judge whether all normal directions are consistent, if actual view is Vp, put all any normal n in cloudiAll point to this
Viewpoint, then it is assumed that all normal directions are consistent, that is, meet：
niVP ＞ 0
In formula, VP is vision pointpTo query point PiVector；If result of calculation is unsatisfactory for formula niVP ＞ 0 requirement, says
Bright point PiNormal vector niWith point PiTo vision pointpVector between vector angle greatly with 90 °, point PiNormal vector niShould be anti-
To；The registration of contact net three-dimensional reconstruction process point cloud is carried out according to this point cloud registration method, the three-dimensional mould of contact net the whole network is obtained
Type, and fault detect is carried out to the insulator of wherein defect.
Compared with prior art, the beneficial effects of the invention are as follows：
1st, the mode of present invention collection contact site cloud is to be mounted with Kinect2.0 depth cameras by what is set up on rail
The detection dolly of machine gathers contact net initial point cloud, and this acquisition mode meets contact net and detect requirement in real time, finally by
A series of contact net the whole network three-dimensional point cloud model that processing are obtained can highly reduce actual contact net time of day.
2nd, the present invention can intuitively, effectively detect insulation using the 3 D detection method of insulator surface normal information
Sub- defect information, meets the objective demand detected in real time today for contact net.
Brief description of the drawings
Fig. 1 is contact net point cloud acquisition scene photo.
Fig. 2 is contact net cloud data acquisition principle figure.
Fig. 3 is initial Frame1 points cloud visualization figure.
Fig. 4 is initial Frame2 points cloud visualization figure.
Fig. 5 is Frame1 and Frame2 point cloud relative position figures.
Fig. 6 is RadiusOutlierRemoval filtering principle schematic diagrames.
Fig. 7 is the schematic diagram before Frame1 down-samplings.
Fig. 8 is the schematic diagram after Frame1 down-samplings.
Fig. 9 is the Frame1 crucial point diagrams of NARF.
Figure 10 is the Frame2 crucial point diagrams of NARF.
Figure 11 is the FPFH feature histograms of Frame1 key points.
Figure 12 is the FPFH feature histograms of Frame2 key points.
Figure 13 is contact net point cloud registering result figure.
Figure 14 is extraction inclined cantilever normal insulation point cloud schematic diagram in contact site cloud.
Figure 15 is extraction inclined cantilever defect insulator point cloud schematic diagram in contact site cloud.
Figure 16 is the normal map of normal rod insulator.
Figure 17 is the normal map for but damaging rod insulator.
Figure 18 is the normal map of the sub- level cross-sectionn of normal insulation.
Figure 19 is the normal map of defect insulator level cross-sectionn.
Figure 20 is the adjacent normal angle value comparison diagram of insulator.
Embodiment
The present invention is further detailed explanation with reference to the accompanying drawings and detailed description, and the inventive method is described in detail such as
Under：
1st, contact net three dimensional point cloud is obtained and pre-processed
Due to the special existence form of contact net, it is impossible to 360 ° are carried out to it around shooting, passes through what is set up on rail
It is mounted with the detection dolly of Kinect2.0 depth cameras to gather contact net initial point cloud, it is bag to obtain initial cloud data
Contain the pcd formatted files of the information such as point cloud position, color, depth；Contact net initial point cloud data forgive complex environment, and
Many noises are pre-processed, it is necessary to carry out initial point cloud, including a cloud denoising, point cloud segmentation, point cloud compressing, point cloud merge,
A variety of processing modes such as point cloud derivation.
2nd, correspondence key point in two frame point clouds is extracted to independent, the clean contact net cloud data that above-mentioned processing is obtained,
Using normalization alignment radial direction feature (Normal Aligned Radial Feature, NARF) by point cloud chart as border is examined
Result is surveyed, surface-stable but the great NARF points of interest of neighborhood substantial variations is extracted；Then, quick point feature histogram is utilized
(Fast Point Feature Histograms, FPFH) algorithm carries out feature description for NARF key points, with this determination two
It is between frame initial point cloud and determine matching relationship；By SAC-IA (Sample Consensus Initial Alignment,
SAC-IA after) Mismatching point is rejected, finally carried out using ICP (Iterative Closest Point, ICP) algorithm accurate
Registration, obtains the full contact net three-dimensional point cloud model after registration.
3rd, extracted using point cloud segmentation means from the defect bracket insulator in the contact pessimistic concurrency control after point cloud registering,
Obtained insulator surface normal is extracted in estimation, and using its normal to a surface information, defect insulator is detected.
It is specific as follows：
1) NARF critical point detections
NARF algorithms, as border detection result, extract surface-stable but the great edge of neighborhood substantial variations by point cloud chart
Point of interest, selective analysis object structure has the advantages that typicalness and invariable rotary shape, openness stronger for design of part
Contact net cloud data is more adapted to.
Catenary Fittings are generally the big different parts of position correspondence of the point curvature of certain in body surface profile even variation, cloud data
Structure separation.Border interest value I (p) is：
I (p)=I1(p)·I2(p)
In formula：
I1(p) factor, I are estimated for surface change intensity2(p) it is the main direction identification factor.P is boundary point, and n is point p neighbour
Element, σ is space scale, wnFor the corresponding weighted values of neighbors n, γ is the one-dimensional angle of point n correspondences.For contact net curved surface side
Edge point, wnValue is 1, and for other points, wnValue is 1- (1- ρ)3, ρ is p and its peripheral point carries out principal component analysis
The curvature value obtained after (Principal Component Analysis, PCA).
2) quick point feature histogram FPFH features description
A, the description of point feature Nogata are sub (Point Feature Histograms, PFH), are led by statistical query point and k
Between domain and in k fields a little between normal relation situation of change, to retouch the geometric properties of plain object.One query point
PqPFH to calculate influence area be scope using its own as the center of circle, round radius is r, PqThe whole phases of all k fields elements
Connect in one network, it is final to obtain PFH Nogatas by any two points in calculating field and their normal relation
Figure.Detailed process is：To arbitrfary point P in a cloudqInquire about all neighbor points in its k field.
B, to any pair of point P in point P k fieldssAnd Pt(s！=t) estimation normal nsAnd nt, it is fixed on a point wherein
One fixed u of justice, v, w local coordinate systems calculate nsAnd ntBetween deviation,
In C, the u defined in B, v, w local coordinate systems, normal n is estimated with one group of anglesAnd ntBetween deviation：
α=vnt
In formula, | | Pt-Ps| |=d represents point PsAnd PtThe Euclidean distance of point-to-point transmission, then PqAll k fields any two points
PsAnd PtRelation just can pass through (α, β, θ, d) represent.Four all class values of k fields are statistically put into histogram, just
P can be obtainedqFeature Descriptor.
D, quick point feature histogram (FPFH) are improved by PFH, and FPFH effectively reduces the computation complexity of algorithm, but
PFH most of evident characteristics are remained again, with preferable robustness and real-time.In order to simplify histogrammic feature calculation,
Its calculating process is as follows：Simplify point feature histogram (Simple Point Feature Histograms, SPFH), i.e., for every
One query point Pq, a tuple α between the point and its neighborhood point is calculated,θ；FPFH is calculated, the k for redefining each point is adjacent
Domain, P is calculated using neighbouring SPFHqFinal FPFH, its calculation formula：
In formula, weights omegakThe query point P in given metric spaceqWith its neighbor point PkThe distance between.Calculate in point cloud frame all
After the FPFH Feature Descriptors of point cloud key point, the required crucial point correspondence of registration can be obtained.
3) point cloud registering
After two frames contact site cloud corresponding relation is determined, registration is carried out to a cloud, rough registration and essence registration two can be divided into
Individual process；It is final to carry out accuracy registration using algorithm after rough registration is rejected Mismatching point using SAC-IA；Through series of points cloud
Registration obtains the complete three dimensional point cloud of contact net zero.
4) three-dimensional values based on insulator surface normal
A, the method for the surface normal estimation of object have a lot, are situated between herein for wherein simplest one kind
Continue, its core concept：The normal for calculating a body surface certain point is similar to estimate the problem of this selects phase tangent plane normal, therefore
This problem can be converted into a least square method plane fitting estimation problem.Based on this principle, it is possible to which calculating obtains thing
The characteristic vector and characteristic value of the solution of body surface normal, i.e. one covariance matrix of analysis, and this covariance matrix C
Can be from each query point PiNeighbor point in construct and obtain, specific covariance matrix is as follows：
In above formula, k is query point PiThe number of neighbor point,It is the three-dimensional barycenter of closest element, λjIt is covariance matrix
J-th of characteristic value,It is j-th of characteristic vector.The specific method of object normal estimation below.
B, establishment section
For a frame cloud data, its cloud quantity is n, to calculate wherein some query point PiNormal, then
Plane equation is first set：
Ax+by+cz+d=0
There is deviation in the point that in general cloud data is concentrated, on direction in order to eliminate these deviations as far as possible, for upper
Constraint is carried out in formula, when meeting a2+b2+c2When=1, plane parameter a, b, c, d can be obtained.It is optimal in order to obtain
Fit Plane, should now allow point PiThe quadratic sum numerical value that k point of proximity of surrounding reaches the plan range is minimum, i.e., full
Foot：
In formula, diIt is that cloud data concentrates either query point Pi=(xi,yi,zi) to corresponding flat apart from di=| axi+
byi+czi-d|.Extreme value is solved using method of Lagrange multipliers, e → min can be caused, function is obtained：
Local derviation is sought in equal sign both sides for d in above formula, and makes partial derivative be zero, obtains：
In formula, order
Barycenter isThen
di=| a Δs xi+bΔyi+cΔzi|
Now, it is then rightMiddle equal sign both sides for a, b,
C seeks partial derivative, obtains
Now, plane parameter a, b, c are solved, it is possible to be converted into solution matrix A characteristic value and characteristic vector.Matrix A
Formula one three saves symmetrical matrix, and the solution formula of its characteristic value is：
In a before2+b2+c2Under=1 constraints, it can try to achieve
Therefore, e minimum value is exactly the minimal eigenvalue of matrix A, and its corresponding characteristic vector is plane parameter a, b, c, utilizes barycenter
Can be in the hope of d.
C, adjustment normal direction：It can be calculated using the principle in a upper trifle and obtain the normal vectors of point Yun Zhongsuo a little,
But the direction of obtained normal vector often can not be consistent.In order to keep the uniformity in normal vector direction, it is necessary to make further
Processing.Judge whether all normal directions are consistent, if actual view is Vp, put all any normal n in cloudiThis is all pointed to regard
Point, then it is assumed that all normal directions are consistent, that is, meet：
niVP ＞ 0
In formula, VP is vision pointpTo query point PiVector.If result of calculation is unsatisfactory for formula (4-10) requirement, explanation
Point PiNormal vector niWith point PiTo vision pointpVector between vector angle greatly with 90 °, point PiNormal vector niShould be reverse.
The registration of contact net three-dimensional reconstruction process point cloud is carried out according to this point cloud registration method, contact net the whole network is obtained
Threedimensional model, and fault detect is carried out to the insulator of wherein defect.
Fig. 1 is contact net three-dimensional point cloud collection site
A, contact net cloud data obtain cloud data collection include optical method, computer vision method, tomographic imaging method, when
Between flight method, pumped FIR laser technology etc., the present invention carries out contact net point cloud acquisition using time flight method.Because contact net is special
Existence form, it is impossible to 360 ° are carried out to it around shooting, can only be by the inspection for being mounted with depth camera that is set up on rail
It is as shown in Figure 2 to gather contact net point cloud acquisition schematic diagram in contact net initial point cloud, the present invention to survey dolly.Detecting small garage
During entering, continuous sampling, because dolly is being moved forward, therefore different frame point cloud depth information is different, and the elevation angle is changing,
Registration is not only related to translation, also rotates, difficulty is bigger.Sample wherein two frame contact net initial point cloud such as Fig. 3, the Fig. 4 obtained
Shown, Fig. 3 is a cloud Frame2 for point cloud Frame1, Fig. 4.
B, contact net point cloud pretreatment
Original contact net cloud data amount is huge, and discrete noise is larger on feature extraction influence, it is necessary to be carried out to original cloud
Pretreatment, including denoising, background segment, down-sampling etc., by the relative position of pretreated two frames point cloud in space such as
Shown in Fig. 5.
A, point cloud denoising
By the interference such as collecting device, environment, contact net original point cloud includes the introduced outlier of measurement noise.Utilize bar
Part removes the density threshold that wave filter (Conditional Removal wave filters) defines contact site cloud, when in original point cloud
Point cloud density is noise when being less than threshold value.
B, point cloud segmentation
Point cloud segmentation：Comprising garbages such as natural background, environment debris in contact net initial point cloud, filtered using radius
Device (Radius Outlier wave filters) is that the centre of sphere sets a radius and point cloud quantity threshold than contact net geometric center, and is counted
Calculate to provide the point cloud quantity inside radius picture ball, retain when calculating point cloud quantity and being more than and provide limit value and change the time, be less than
Limit value is then rejected.Its principle such as Fig. 6, if limit value is the point deletion of the circle centre position of left circles in 1 neighbor point, Fig. 6 in figure, if limit
2 neighbor points are made as, then the point of the respective circle centre position of left circles and right circles is all deleted.
C, point cloud down-sampling
Acquired original point cloud is more intensive, and difficulty is caused to follow-up method line computation and analysis.Utilize voxel wave filter
(Voxel Grid wave filters) is that one three-dimensional voxel grid of origin cloud establishment (stand by the three-dimensional that can be interpreted as small with voxel grid
The set of cube), in each voxel grid, with voxel center of gravity a little come other points in approximate voxel, the so voxel
A little just finally represented with one focus point with regard to interior, cloud is put after being filtered after being handled for all voxels.Before and after down-sampling
The frame point cloud number of Frame1, Frame2 two is as shown in table 1, point cloud comparison diagram such as Fig. 7, Fig. 8 institute wherein before and after Frame1 down-samplings
Show, it can be seen that after down-sampling, the point quantity of same cloud data same position is significantly reduced, but its geometry
Do not change, this has very big benefit for improving the speed of follow-up points cloud processing.
Table 1 puts cloud number change figure before and after pre-processing
C, contact net cloud data registration
The corresponding key point for obtaining Frame1 and the frame point clouds of Frame2 two by NARF and FPFH algorithms utilizes sampling to rear
Uniformity initial registration algorithm (SAC-IA) carries out just registration to two frame point clouds, is accurately matched somebody with somebody using ICP point cloud registration algorithms
It is accurate.
Frame1 and the frame cloud datas of Frame2 two NARF key points, as a result as shown in Figure 9, Figure 10；Frame1 and
The FPFH features of the frames of Frame2 two are described as shown in Figure 11, Figure 12, in figure, and the numerical value of abscissa is 33 from 0-32 divided
Statistics is interval, and the numerical value of ordinate is that correspondence counts the point (p included in interval1,p2,…,pk) number.As seen from the figure
Frame1 and the FPFH feature histograms of Frame2 one group of corresponding key point are quite similar, although the broken line of two figures is simultaneously endless
Complete consistent, but it can be seen that the basic trend of broken line is all consistent, the abscissa corresponding to each peak value of ordinate is also substantially
Identical, the key point of this explanation both Frame1 and Frame2 selection is characterized in very much like, and this point cloud for after is matched somebody with somebody
Accurate precision is highly beneficial.Calculate and all put after the FPFH Feature Descriptors of cloud key point in point cloud frame, needed for can obtaining registration
Crucial point correspondence.
A, SAC-IA initial registration
Sampling uniformity initial registration algorithm is broadly divided into two parts：Greedy initial registration alignment methods, sampling one
Cause property algorithm.Because point cloud has the indeformable feature of internal rotating, therefore using greedy Alignment Algorithm with very good
Good robustness.But greedy Alignment Algorithm computation complexity is higher, and it is possible to can only obtain local optimal
Solution, therefore, it is also desirable to using sampling coherence method, it is intended to identical correspondence geometrical relationship is determined, without calculating limited
All combinations of corresponding relation.
SAC-IA basic processes are as follows：
(1) sample point is chosen from a cloud Frame1, is adjusted the distance while calculating sample point and matching somebody with somebody more than predetermined threshold value most
Small distance；
(2) for individual sample point, all point clouds for meeting similarity condition are found respectively in a cloud Frame2, and at random
The some of corresponding relations to calculate sampled point of selection；
(3) corresponding relation of conjunction is converged according to an even point, rigid body translation matrix is calculated, and examine by computation measure mistake
Test the quality that conversion is put to the proof.
Error measure can be examined by Huber judgement schematics：
(4) three above step is repeated, until reaching that Optimal Error is measured.
According to above-mentioned Computing Principle, Mismatching point is rejected by SAC-IA, SAC-IA initial conversions matrix has been obtained such as
Under：
B, ICP accuracy registration
Wrong corresponding relation present in Frame1 and the frame point clouds of Frame2 two is eliminated by SAC-IA algorithms, and tried to achieve
After rough registration transition matrix, to realize high-precision point cloud registering, carried out using ICP point cloud registration algorithms most widely used at present
Accuracy registration, and obtain the transformed matrix of ICP accuracy registrations.
The process of ICP accuracy registrations can be divided as follows：
(1) contact net point set resampling：Target point cloud X={ x will be a little designated as respectively in Frame1 and Frame21,
x2,…,xm, reference point clouds Y={ y1,y2,…,yn, wherein X point clouds number is m, and Y point clouds number is n, and m≤n.
(2) point selection is matched：Accelerate searching closest approach cloud using Quaternion method initialization, and by Kd-Tree.If rotation
It is unit quaternary number q to change commutation amountR=[qx,qy,qz,qw], wherein qx>=0, and qx+qy+qz+qw=1, translation transformation to
Measure qT=[tx,ty,tz].Frame1 and Frame2 two panels point cloud center of gravity is
In this way, the optimal spin matrix R (q for obtaining two frame point cloud registerings can be calculatedR) and optimal translation vector qT,
(3) point is assigned to weights：The weight of Frame1 and Frame2 matching double points is determined by their normal vector.
If the respectively n of the normal vector to point matched1And n2, then weight be：
W=n1·n2
(4) point is to rejecting：Frame1 and Frame2 is not to be completely superposed, so general by the erroneous matching comprising edge
Relation point is to rejecting.
(5) suitable error metrics function is selected：Coordinate transform square is realized using the alternative manner based on least square method
Battle array optimization, makes error function minimum.Error function is defined as：
(6) optimize：Repeat the above steps, realize that error is minimized.Spin matrix R (the q obtained using minimal errorR)
With translation vector qTBy Frame1 and Fame2 registrations, final acquisition accuracy registration matrix：
C, contact net point cloud registering result
Extracting, describing by key point, rejected using SAC-IA Mismatching points, ICP accuracy registrations, final registration effect
As shown in figure 13, in figure relative depthwise position rearward for initial segment cloud Frame1, target point cloud Frame2 with it is registering after life
Into new point cloud almost coincide together, but can be seen that the point cloud after registration is more complete in edge.
D, the three-dimensional values based on contact net insulator surface normal
A, extraction rod insulator
The rod insulator on inclined cantilever that will be contacted in the cloud of site is extracted by way of point cloud segmentation.In Figure 14
Illustrate the process that normal insulation point cloud is extracted from contact site cloud；Illustrate to extract from contact site cloud in Figure 15 and lack
Damage the process of insulator point cloud.
B, insulator method line computation
The insulator of defect is detected using insulator surface normal information, and by the information quantization of detection.According to
Body surface normal principle is solved, their method is solved to insulator damaged in normal insulator in Figure 14 and Figure 15 respectively
Line, its normal effect is as shown in Figure 16, Figure 17.
C, rod insulator main body are to be alternately stacked the long bar type object constituted by two kinds of different cylinders of diameter, and
In complete Central Symmetry, therefore in theory, the surface normal of its cloud necessarily points to the center of circle of its place horizontal plane circle.But it is real
The insulator point cloud that border collection is obtained, its surface there is point the lacking of cloud, it is overlapping situations such as, so dry even in filtering, going
After be also impossible to complete smooth-going, it is smooth, cause its surface normal to be also impossible to point to the center of circle completely.Figure 18 and Figure 19 are opened up respectively
The normal map of the cross section of the insulator of normal insulator and defect is in the horizontal direction shown.
The right in Figure 18 and Figure 19 is the normal map of respective insulator level cross-sectionn.Two cross sections are not in figure
Complete circle, because only have taken dolly connecing in traveling process forward during the collection point cloud as shown in Figure 2
Net-fault cloud data, when dolly is not shot after contact net, so the point cloud collected is not 360 °.
Calculate the angle α between all adjacent normal line vectors in Figure 18 in the sub- cross section of normal insulationm, acquisition it is all
Angle collection { α1,a2…am}；Similarly, the folder between all adjacent normal line vectors in Figure 19 in defect insulator cross section is calculated
Angle betan, all angle collection { β of acquisition1,β2…βn}。
By angle set { α achieved above1,a2…amAnd { β1,β2…βnData induction arrange after draw line chart, can
To find out difference between the two, as shown in figure 20.
In Figure 20 line chart, what abscissa was represented is the adjacent normal angle number of two insulator cross-sectional surfaces,
{ α is have selected in this figure1,a2…amAnd { β1,β2…βnIn each 200 angle values be used as statistical sample；What ordinate was represented is
Angle.
It can be seen that the angle angle between the adjacent normal of the sub- cross-sectional surface of normal insulation that dotted line broken line is represented
Degree, because insulator puts the incomplete smooth of cloud surface in itself, this broken line rises and falls in certain amplitude, but its ordinate pair
The variance for all angles value answered is smaller, illustrates that the angle value between the sub- surface vector normal of normal insulation is more or less the same.In figure
Angle angle between the adjacent normal of defect insulator cross-sectional surface that solid line broken line is represented, this broken line is substantially and dotted line
Broken line rises and falls in the same horizontal line, except there is a unconventional peak value considerably beyond other angle values, illustrates herein two
Angle between bar normal is very big, and this position is exactly the defect location of defect insulator.The present invention is understood from Detection results
Result in the ideal defect information for detecting parts in contact net.
Claims (6)
1. a kind of the whole network three-dimensional rebuilding method of contact net zero based on NARF and FPFH, it is characterised in that comprise the following steps：
Step 1：Contact net initial point cloud is gathered by being provided with the detection dolly of Kinect2.0 depth cameras, at the beginning of acquisition
Beginning cloud data is pcd formatted files, and it includes some cloud positions, color and depth；
Step 2：Initial point cloud is pre-processed, including a cloud denoising, point cloud segmentation, point cloud compressing, the fusion of point cloud and point cloud
Derive from；
Step 3：For contact net cloud data, correspondence key point in two frame point clouds is extracted, using normalization alignment radial direction feature
NARF, as border is detected, extracts the NARF key points that surface-stable but neighborhood depth information change to point cloud chart, this
A little changes include the principal direction of surface variation coefficient and change；
Step 4：Feature description is carried out to NARF key points using quick point feature histogram FPFH algorithms, two frame initial points are determined
Crucial point correspondence between cloud；
Step 5：Point cloud registering is carried out using key point, Mismatching point is rejected by SAC-IA；
Step 6：Accuracy registration is carried out using ICP algorithm, the full contact net three-dimensional point cloud model after registration is obtained.
2. a kind of the whole network three-dimensional rebuilding method of contact net zero based on NARF and FPFH as claimed in claim 1, its feature exists
In, in addition to step 7：
Step 7：Using point cloud segmentation means, the defect bracket insulator in the contact pessimistic concurrency control after point cloud registering is extracted
Come, obtained insulator surface normal is extracted in estimation, and defect insulator is detected using its normal to a surface information.
3. a kind of the whole network three-dimensional rebuilding method of contact net zero based on NARF and FPFH as claimed in claim 1, its feature exists
In the step 3 is specially：
Step 3.1：The structure separation of the big different parts of position correspondence of the point curvature of certain in cloud data, border interest value I (p)
For：
I (p)=I1(p)·I2(p)
In formula：
<mrow>
<msub>
<mi>I</mi>
<mn>1</mn>
</msub>
<mrow>
<mo>(</mo>
<mi>p</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munder>
<mrow>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
</mrow>
<mi>i</mi>
</munder>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<msub>
<mi>w</mi>
<msub>
<mi>n</mi>
<mi>i</mi>
</msub>
</msub>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
<mo>(</mo>
<mrow>
<mn>1</mn>
<mo>-</mo>
<mfrac>
<mrow>
<mn>10</mn>
<mo>&CenterDot;</mo>
<mo>|</mo>
<mo>|</mo>
<mi>p</mi>
<mo>-</mo>
<msub>
<mi>n</mi>
<mi>i</mi>
</msub>
<mo>|</mo>
<mo>|</mo>
</mrow>
<mi>&sigma;</mi>
</mfrac>
<mo>,</mo>
<mn>0</mn>
</mrow>
<mo>)</mo>
</mrow>
<mrow>
<msub>
<mi>I</mi>
<mn>2</mn>
</msub>
<mo>=</mo>
<munder>
<mrow>
<mi>m</mi>
<mi>a</mi>
<mi>x</mi>
</mrow>
<mrow>
<mi>i</mi>
<mo>-</mo>
<mi>j</mi>
</mrow>
</munder>
<mrow>
<mo>(</mo>
<mi>f</mi>
<mo>(</mo>
<msub>
<mi>n</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
<mi>f</mi>
<mo>(</mo>
<msub>
<mi>n</mi>
<mi>j</mi>
</msub>
<mo>)</mo>
<mo>(</mo>
<mrow>
<mn>1</mn>
<mo>-</mo>
<mo>|</mo>
<mi>c</mi>
<mi>o</mi>
<mi>s</mi>
<mrow>
<mo>(</mo>
<mrow>
<msubsup>
<mi>&gamma;</mi>
<msub>
<mi>n</mi>
<mi>i</mi>
</msub>
<mo>&prime;</mo>
</msubsup>
<mo>-</mo>
<msubsup>
<mi>&gamma;</mi>
<msub>
<mi>n</mi>
<mi>j</mi>
</msub>
<mo>&prime;</mo>
</msubsup>
</mrow>
<mo>)</mo>
</mrow>
<mo>|</mo>
</mrow>
<mo>)</mo>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mi>f</mi>
<mrow>
<mo>(</mo>
<mi>n</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msqrt>
<mrow>
<msub>
<mi>w</mi>
<mi>n</mi>
</msub>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<mfrac>
<mrow>
<mn>2</mn>
<mo>&CenterDot;</mo>
<mo>|</mo>
<mo>|</mo>
<mi>p</mi>
<mo>-</mo>
<mi>n</mi>
<mo>|</mo>
<mo>|</mo>
</mrow>
<mi>&sigma;</mi>
</mfrac>
<mo>-</mo>
<mfrac>
<mn>1</mn>
<mn>2</mn>
</mfrac>
<mo>)</mo>
</mrow>
</mrow>
</msqrt>
</mrow>
I1(p) factor, I are estimated for surface change intensity2(p) it is the main direction identification factor；P is boundary point, and n is first for point p neighbour
Element, σ is space scale, wnFor the corresponding weighted values of neighbors n, γ is the one-dimensional angle of point n correspondences；For contact net curved edges
Point, wnValue is 1, and for other points, wnValue is 1- (1- ρ)3, ρ is to be obtained after p and its peripheral point carry out principal component analysis
Curvature value.
4. a kind of the whole network three-dimensional rebuilding method of contact net zero based on NARF and FPFH as claimed in claim 1, its feature exists
In the step 4 is specially：
Step 4.1：To arbitrfary point P in a cloudqInquire about all neighbor points in its k field；
Step 4.2：To any pair of point P in point P k fieldssAnd PtEstimate normal nsAnd nt, wherein s！=t, wherein on a point
A fixed u is defined, v, w local coordinate systems calculate nsAnd ntBetween deviation
Step 4.3：In the u of step 4.2, in w local coordinate systems, normal n is estimated with one group of angle by vsAnd ntBetween deviation：
α=vnt
<mrow>
<mi>&beta;</mi>
<mo>=</mo>
<mi>u</mi>
<mo>&CenterDot;</mo>
<mfrac>
<mrow>
<mo>(</mo>
<msub>
<mi>P</mi>
<mi>t</mi>
</msub>
<mo>-</mo>
<msub>
<mi>P</mi>
<mi>s</mi>
</msub>
<mo>)</mo>
</mrow>
<mrow>
<mo>|</mo>
<mo>|</mo>
<msub>
<mi>P</mi>
<mi>t</mi>
</msub>
<mo>-</mo>
<msub>
<mi>P</mi>
<mi>s</mi>
</msub>
<mo>|</mo>
<mo>|</mo>
</mrow>
</mfrac>
</mrow>
<mrow>
<mi>&theta;</mi>
<mo>=</mo>
<mi>arctan</mi>
<mrow>
<mo>(</mo>
<mi>w</mi>
<mo>&CenterDot;</mo>
<mover>
<msub>
<mi>n</mi>
<mi>t</mi>
</msub>
<mo>&RightArrow;</mo>
</mover>
<mo>,</mo>
<mi>u</mi>
<mo>&CenterDot;</mo>
<mover>
<msub>
<mi>n</mi>
<mi>t</mi>
</msub>
<mo>&RightArrow;</mo>
</mover>
<mo>)</mo>
</mrow>
</mrow>
In formula, | | Pt-Ps| |=d represents point PsAnd PtThe Euclidean distance of point-to-point transmission, then PqAll k fields any two points PsAnd Pt
Relation pass through (α, β, θ, d) represent；Four all class values of k fields are statistically put into histogram, obtain PqFeature
Description；
Step 4.4：For each query point Pq, a tuple α between the point and its neighborhood point is calculated,θ；Calculate FPFH,
The k neighborhoods of each point are redefined, P is calculated using neighbouring SPFHqFinal FPFH, its calculation formula isIn formula, weights omegakRepresent in given metric space, query point PqWith
Its neighbor point PkThe distance between；
Calculate and all put after the FPFH Feature Descriptors of cloud key point in point cloud frame, obtain the required key point correspondence of registration and close
System.
5. a kind of the whole network three-dimensional rebuilding method of contact net zero based on NARF and FPFH as claimed in claim 1, its feature exists
In the step 5 point cloud registration is specially：
After two frames contact site cloud corresponding relation is determined, registration is carried out to a cloud, it is divided into rough registration and essence registration；Rough registration
Using SAC-IA, after Mismatching point is rejected, accuracy registration is carried out using algorithm；Complete connect is obtained through a series of point cloud registerings
Touch net zero three dimensional point cloud.
6. a kind of the whole network three-dimensional rebuilding method of contact net zero based on NARF and FPFH as claimed in claim 2, its feature exists
In the step 7 is specially：
Step 7.1：The characteristic vector and characteristic value of one covariance matrix of analysis, and this covariance matrix C is looked into from each
Ask point PiNeighbor point in construct and obtain, specific covariance matrix is as follows：
<mrow>
<mi>C</mi>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mi>k</mi>
</mfrac>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>k</mi>
</munderover>
<mo>&CenterDot;</mo>
<mrow>
<mo>(</mo>
<msub>
<mi>P</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<mover>
<mi>P</mi>
<mo>&OverBar;</mo>
</mover>
<mo>)</mo>
</mrow>
<mo>&CenterDot;</mo>
<msup>
<mrow>
<mo>(</mo>
<msub>
<mi>P</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<mover>
<mi>P</mi>
<mo>&OverBar;</mo>
</mover>
<mo>)</mo>
</mrow>
<mi>T</mi>
</msup>
<mo>,</mo>
<mi>C</mi>
<mo>&CenterDot;</mo>
<mover>
<msub>
<mi>v</mi>
<mi>j</mi>
</msub>
<mo>&RightArrow;</mo>
</mover>
<mo>=</mo>
<msub>
<mi>&lambda;</mi>
<mi>j</mi>
</msub>
<mo>&CenterDot;</mo>
<mover>
<msub>
<mi>v</mi>
<mi>j</mi>
</msub>
<mo>&RightArrow;</mo>
</mover>
<mo>,</mo>
<mi>j</mi>
<mo>&Element;</mo>
<mo>{</mo>
<mn>0</mn>
<mo>,</mo>
<mn>1</mn>
<mo>,</mo>
<mn>2</mn>
<mo>}</mo>
</mrow>
In above formula, k is query point PiThe number of neighbor point,It is the three-dimensional barycenter of closest element, λjIt is the of covariance matrix
J characteristic value,It is j-th of characteristic vector；
Step 7.2：Establish section
For a frame cloud data, its cloud quantity is n, to calculate wherein some query point PiNormal, plane is first set
Equation：
Ax+by+cz+d=0
For carrying out constraint in above formula, when meeting a2+b2+c2When=1, plane parameter a, b, c, d are obtained；Allow point PiAround
The quadratic sum numerical value that k point of proximity reaches the plan range is minimum, that is, meets：
<mrow>
<mi>e</mi>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<msubsup>
<mi>d</mi>
<mi>i</mi>
<mn>2</mn>
</msubsup>
<mo>&RightArrow;</mo>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
</mrow>
2
In formula, diIt is that cloud data concentrates either query point Pi=(xi,yi,zi) to corresponding flat apart from di=| axi+byi+
czi-d|；Extreme value is solved using method of Lagrange multipliers so that e → min, obtain function：
<mrow>
<mi>f</mi>
<mo>=</mo>
<mi>e</mi>
<mo>-</mo>
<mi>&lambda;</mi>
<mrow>
<mo>(</mo>
<msup>
<mi>a</mi>
<mn>2</mn>
</msup>
<mo>+</mo>
<msup>
<mi>b</mi>
<mn>2</mn>
</msup>
<mo>+</mo>
<msup>
<mi>c</mi>
<mn>2</mn>
</msup>
<mo>-</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munderover>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<msup>
<mi>d</mi>
<mn>2</mn>
</msup>
<mo>-</mo>
<mi>&lambda;</mi>
<mrow>
<mo>(</mo>
<msup>
<mi>a</mi>
<mn>2</mn>
</msup>
<mo>+</mo>
<msup>
<mi>b</mi>
<mn>2</mn>
</msup>
<mo>+</mo>
<msup>
<mi>c</mi>
<mn>2</mn>
</msup>
<mo>-</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
Local derviation is sought in equal sign both sides for d in above formula, and makes partial derivative be zero, obtains：
<mrow>
<mi>d</mi>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mi>n</mi>
</mfrac>
<mrow>
<mo>(</mo>
<mi>a</mi>
<munderover>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<msub>
<mi>x</mi>
<mi>i</mi>
</msub>
<mo>+</mo>
<mi>b</mi>
<munderover>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<msub>
<mi>y</mi>
<mi>i</mi>
</msub>
<mo>+</mo>
<mi>c</mi>
<munderover>
<mi>&Sigma;</mi>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<msub>
<mi>z</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
In formula, orderMatter
The heart isThen
di=| a Δs xi+bΔyi+cΔzi|
It is right againLocal derviation is sought for a, b, c in middle equal sign both sides
Number, is obtained
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<mn>2</mn>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<mrow>
<mo>(</mo>
<msub>
<mi>a&Delta;x</mi>
<mi>i</mi>
</msub>
<mo>+</mo>
<msub>
<mi>b&Delta;y</mi>
<mi>i</mi>
</msub>
<mo>+</mo>
<msub>
<mi>c&Delta;z</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<msub>
<mi>&Delta;x</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<mn>2</mn>
<mi>&lambda;</mi>
<mi>a</mi>
<mo>=</mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mn>2</mn>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<mrow>
<mo>(</mo>
<msub>
<mi>a&Delta;x</mi>
<mi>i</mi>
</msub>
<mo>+</mo>
<msub>
<mi>b&Delta;y</mi>
<mi>i</mi>
</msub>
<mo>+</mo>
<msub>
<mi>c&Delta;z</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<msub>
<mi>&Delta;y</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<mn>2</mn>
<mi>&lambda;</mi>
<mi>b</mi>
<mo>=</mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mn>2</mn>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>n</mi>
</munderover>
<mrow>
<mo>(</mo>
<msub>
<mi>a&Delta;x</mi>
<mi>i</mi>
</msub>
<mo>+</mo>
<msub>
<mi>b&Delta;y</mi>
<mi>i</mi>
</msub>
<mo>+</mo>
<msub>
<mi>c&Delta;z</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<msub>
<mi>&Delta;z</mi>
<mi>i</mi>
</msub>
<mo>-</mo>
<mn>2</mn>
<mi>&lambda;</mi>
<mi>c</mi>
<mo>=</mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
Plane parameter a, b, c are solved, solution matrix A characteristic value and characteristic vector is converted into；Matrix A is the one three symmetrical square of section
Gust, the solution formula of its characteristic value is：
<mrow>
<mi>&lambda;</mi>
<mo>=</mo>
<mfrac>
<mrow>
<msup>
<mrow>
<mo>(</mo>
<mi>A</mi>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mi>T</mi>
</msup>
<mo>&CenterDot;</mo>
<mi>x</mi>
</mrow>
<mrow>
<msup>
<mi>x</mi>
<mi>T</mi>
</msup>
<mo>&CenterDot;</mo>
<mi>x</mi>
</mrow>
</mfrac>
<mo>,</mo>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>&NotEqual;</mo>
<mn>0</mn>
<mo>)</mo>
</mrow>
</mrow>
In a2+b2+c2Under=1 constraints, try to achieveTried to achieve using barycenter
d；
Step 7.3：Adjust normal direction
Judge whether all normal directions are consistent, if actual view is Vp, put all any normal n in cloudiThis is all pointed to regard
Point, then it is assumed that all normal directions are consistent, that is, meet：
niVP ＞ 0
In formula, VP is vision pointpTo query point PiVector；If result of calculation is unsatisfactory for formula niVP ＞ 0 requirement, illustrates a little
PiNormal vector niWith point PiTo vision pointpVector between vector angle greatly with 90 °, point PiNormal vector niShould be reverse；Root
Point cloud registration method carries out the registration of contact net three-dimensional reconstruction process point cloud accordingly, obtains contact net the whole network threedimensional model, and
Fault detect is carried out to the insulator of wherein defect.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710445701.XA CN107123161A (en) | 2017-06-14 | 2017-06-14 | A kind of the whole network three-dimensional rebuilding method of contact net zero based on NARF and FPFH |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710445701.XA CN107123161A (en) | 2017-06-14 | 2017-06-14 | A kind of the whole network three-dimensional rebuilding method of contact net zero based on NARF and FPFH |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107123161A true CN107123161A (en) | 2017-09-01 |
Family
ID=59729219
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710445701.XA Pending CN107123161A (en) | 2017-06-14 | 2017-06-14 | A kind of the whole network three-dimensional rebuilding method of contact net zero based on NARF and FPFH |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107123161A (en) |
Cited By (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108154525A (en) * | 2017-11-21 | 2018-06-12 | 四川大学 | A kind of matched bone fragments joining method of feature based |
| CN108171796A (en) * | 2017-12-25 | 2018-06-15 | 燕山大学 | A kind of inspection machine human visual system and control method based on three-dimensional point cloud |
| CN108230432A (en) * | 2017-12-12 | 2018-06-29 | 中国南方电网有限责任公司超高压输电公司广州局 | A kind of insulator laser point cloud three-dimensional rebuilding method based on CS-RBF |
| CN108564575A (en) * | 2018-04-11 | 2018-09-21 | 西南交通大学 | A kind of contactless catenary's parameters detection method based on three dimensional point cloud |
| CN108764157A (en) * | 2018-05-30 | 2018-11-06 | 湖北大学 | Building laser footpoint extracting method and system based on normal vector Gaussian Profile |
| CN108776991A (en) * | 2018-04-17 | 2018-11-09 | 深圳清创新科技有限公司 | Three-dimensional modeling method, device, storage medium and computer equipment |
| CN108921164A (en) * | 2018-06-15 | 2018-11-30 | 西南交通大学 | A kind of contact net positioner slope detection method based on three-dimensional point cloud segmentation |
| CN110473223A (en) * | 2019-08-15 | 2019-11-19 | 西南交通大学 | Two dimensional image auxiliary partition method based on contact network cantilever system three-dimensional point cloud |
| CN110503040A (en) * | 2019-08-23 | 2019-11-26 | 斯坦德机器人（深圳）有限公司 | Obstacle detection method and device |
| CN110866969A (en) * | 2019-10-18 | 2020-03-06 | 西北工业大学 | Engine blade reconstruction method based on neural network and point cloud registration |
| CN110910463A (en) * | 2019-10-14 | 2020-03-24 | 中国科学院自动化研究所 | Full-view-point cloud data fixed-length ordered encoding method and equipment and storage medium |
| CN110956664A (en) * | 2019-12-17 | 2020-04-03 | 武汉易维晟医疗科技有限公司 | Real-time camera position repositioning method for handheld three-dimensional scanning system |
| CN111316330A (en) * | 2017-10-06 | 2020-06-19 | 交互数字Vc控股公司 | Method and apparatus for upsampling a point cloud |
| CN111429492A (en) * | 2020-03-20 | 2020-07-17 | 南京航空航天大学 | Airplane C-shaped beam registration method based on local invariance |
| CN111611996A (en) * | 2020-04-22 | 2020-09-01 | 青岛联合创智科技有限公司 | Computing method of point cloud characteristic point descriptor |
| CN111612909A (en) * | 2020-05-14 | 2020-09-01 | 北京林业大学 | Three-dimensional branch breaking effect simulation algorithm based on point cloud data |
| CN111986296A (en) * | 2020-08-20 | 2020-11-24 | 叠境数字科技（上海）有限公司 | CG animation synthesis method for bullet time |
| CN111986219A (en) * | 2020-08-10 | 2020-11-24 | 中国科学院光电技术研究所 | Matching method of three-dimensional point cloud and free-form surface model |
| CN112365528A (en) * | 2020-07-23 | 2021-02-12 | 哈尔滨岛田大鹏工业股份有限公司 | Three-dimensional point cloud model gradual refinement and rapid registration method based on principal component analysis |
| CN112785711A (en) * | 2021-01-21 | 2021-05-11 | 浙江科技学院 | Insulator creepage distance detection method and detection system based on three-dimensional reconstruction |
| CN113076949A (en) * | 2021-03-31 | 2021-07-06 | 成都唐源电气股份有限公司 | Method and system for quickly positioning parts of contact net |
| CN113177477A (en) * | 2021-04-29 | 2021-07-27 | 湖南大学 | Target detection and identification method based on three-dimensional point cloud analysis |
| CN113408074A (en) * | 2021-06-28 | 2021-09-17 | 吉林大学 | Wheel set tread parameter measuring method and device |
| CN113469195A (en) * | 2021-06-25 | 2021-10-01 | 浙江工业大学 | Target identification method based on self-adaptive color fast point feature histogram |
| CN113496199A (en) * | 2020-03-20 | 2021-10-12 | 安波福技术有限公司 | Histogram-based L-shaped detection of target objects |
| CN113516694A (en) * | 2021-04-13 | 2021-10-19 | 同济大学 | Multi-Kinect-based dynamic visual control method for remote medical scene |
| CN113808133A (en) * | 2021-11-19 | 2021-12-17 | 西南交通大学 | Subway brake shoe fault detection method based on three-dimensional point cloud |
| CN113947729A (en) * | 2021-11-12 | 2022-01-18 | 华南农业大学 | Plant point cloud data depth image processing method and system based on competitive network |
| CN113971768A (en) * | 2021-10-27 | 2022-01-25 | 国网安徽省电力有限公司马鞍山供电公司 | Unmanned aerial vehicle-based three-dimensional dynamic detection method for power transmission line illegal building |
| CN115063458A (en) * | 2022-07-27 | 2022-09-16 | 武汉工程大学 | Material pile volume calculation method based on three-dimensional laser point cloud |
| CN115205354A (en) * | 2022-06-23 | 2022-10-18 | 中国人民解放军国防科技大学 | Phased array laser radar imaging method based on RANSAC and ICP point cloud registration |
| CN115953604A (en) * | 2023-03-13 | 2023-04-11 | 泰安市金土地测绘整理有限公司 | Real estate geographic information mapping data acquisition method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104299260A (en) * | 2014-09-10 | 2015-01-21 | 西南交通大学 | Contact network three-dimensional reconstruction method based on SIFT and LBP point cloud registration |
| CN104794748A (en) * | 2015-03-17 | 2015-07-22 | 上海海洋大学 | Three-dimensional space map construction method based on Kinect vision technology |
| CN106803267A (en) * | 2017-01-10 | 2017-06-06 | 西安电子科技大学 | Indoor scene three-dimensional rebuilding method based on Kinect |
-
2017
- 2017-06-14 CN CN201710445701.XA patent/CN107123161A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104299260A (en) * | 2014-09-10 | 2015-01-21 | 西南交通大学 | Contact network three-dimensional reconstruction method based on SIFT and LBP point cloud registration |
| CN104794748A (en) * | 2015-03-17 | 2015-07-22 | 上海海洋大学 | Three-dimensional space map construction method based on Kinect vision technology |
| CN106803267A (en) * | 2017-01-10 | 2017-06-06 | 西安电子科技大学 | Indoor scene three-dimensional rebuilding method based on Kinect |
Non-Patent Citations (5)
| Title |
|---|
| 曹诗卉等: "基于最小二乘法平面拟合的点云法矢算法", 《空军预警学院学报》 * |
| 朱德海: "《点云库PCL学习教程》", 31 October 2012, 北京航空航天大学出版社 * |
| 王颖等: "圆结构光视觉三维点管道缺陷检测及重构", 《应用光学》 * |
| 胡修祥等: "结合NARF特征的改进型3D-NDT多视点云匹配", 《信号处理》 * |
| 艾达等: "基于FPFH特征的三维场景重建", 《计算机测量与控制》 * |
Cited By (45)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111316330A (en) * | 2017-10-06 | 2020-06-19 | 交互数字Vc控股公司 | Method and apparatus for upsampling a point cloud |
| CN108154525A (en) * | 2017-11-21 | 2018-06-12 | 四川大学 | A kind of matched bone fragments joining method of feature based |
| CN108154525B (en) * | 2017-11-21 | 2022-06-07 | 四川大学 | Bone fragment splicing method based on feature matching |
| CN108230432A (en) * | 2017-12-12 | 2018-06-29 | 中国南方电网有限责任公司超高压输电公司广州局 | A kind of insulator laser point cloud three-dimensional rebuilding method based on CS-RBF |
| CN108171796A (en) * | 2017-12-25 | 2018-06-15 | 燕山大学 | A kind of inspection machine human visual system and control method based on three-dimensional point cloud |
| CN108564575A (en) * | 2018-04-11 | 2018-09-21 | 西南交通大学 | A kind of contactless catenary's parameters detection method based on three dimensional point cloud |
| CN108564575B (en) * | 2018-04-11 | 2022-03-15 | 西南交通大学 | Non-contact type contact network parameter detection method based on three-dimensional point cloud data |
| CN108776991A (en) * | 2018-04-17 | 2018-11-09 | 深圳清创新科技有限公司 | Three-dimensional modeling method, device, storage medium and computer equipment |
| CN108776991B (en) * | 2018-04-17 | 2023-02-28 | 深圳一清创新科技有限公司 | Three-dimensional modeling method, three-dimensional modeling device, storage medium and computer equipment |
| CN108764157A (en) * | 2018-05-30 | 2018-11-06 | 湖北大学 | Building laser footpoint extracting method and system based on normal vector Gaussian Profile |
| CN108921164A (en) * | 2018-06-15 | 2018-11-30 | 西南交通大学 | A kind of contact net positioner slope detection method based on three-dimensional point cloud segmentation |
| CN108921164B (en) * | 2018-06-15 | 2021-10-08 | 西南交通大学 | Contact net locator gradient detection method based on three-dimensional point cloud segmentation |
| CN110473223B (en) * | 2019-08-15 | 2023-05-05 | 西南交通大学 | Two-dimensional image auxiliary segmentation method based on three-dimensional point cloud of catenary cantilever system |
| CN110473223A (en) * | 2019-08-15 | 2019-11-19 | 西南交通大学 | Two dimensional image auxiliary partition method based on contact network cantilever system three-dimensional point cloud |
| CN110503040A (en) * | 2019-08-23 | 2019-11-26 | 斯坦德机器人（深圳）有限公司 | Obstacle detection method and device |
| CN110910463A (en) * | 2019-10-14 | 2020-03-24 | 中国科学院自动化研究所 | Full-view-point cloud data fixed-length ordered encoding method and equipment and storage medium |
| CN110866969A (en) * | 2019-10-18 | 2020-03-06 | 西北工业大学 | Engine blade reconstruction method based on neural network and point cloud registration |
| CN110866969B (en) * | 2019-10-18 | 2022-06-14 | 西北工业大学 | Engine blade reconstruction method based on neural network and point cloud registration |
| CN110956664A (en) * | 2019-12-17 | 2020-04-03 | 武汉易维晟医疗科技有限公司 | Real-time camera position repositioning method for handheld three-dimensional scanning system |
| CN110956664B (en) * | 2019-12-17 | 2023-06-16 | 武汉易维晟医疗科技有限公司 | Real-time repositioning method for camera position of handheld three-dimensional scanning system |
| CN113496199A (en) * | 2020-03-20 | 2021-10-12 | 安波福技术有限公司 | Histogram-based L-shaped detection of target objects |
| CN113496199B (en) * | 2020-03-20 | 2024-03-29 | 安波福技术有限公司 | Histogram-based L-shaped detection of target objects |
| CN111429492A (en) * | 2020-03-20 | 2020-07-17 | 南京航空航天大学 | Airplane C-shaped beam registration method based on local invariance |
| CN111611996A (en) * | 2020-04-22 | 2020-09-01 | 青岛联合创智科技有限公司 | Computing method of point cloud characteristic point descriptor |
| CN111612909A (en) * | 2020-05-14 | 2020-09-01 | 北京林业大学 | Three-dimensional branch breaking effect simulation algorithm based on point cloud data |
| CN112365528A (en) * | 2020-07-23 | 2021-02-12 | 哈尔滨岛田大鹏工业股份有限公司 | Three-dimensional point cloud model gradual refinement and rapid registration method based on principal component analysis |
| CN111986219B (en) * | 2020-08-10 | 2023-09-19 | 中国科学院光电技术研究所 | Matching method of three-dimensional point cloud and free-form surface model |
| CN111986219A (en) * | 2020-08-10 | 2020-11-24 | 中国科学院光电技术研究所 | Matching method of three-dimensional point cloud and free-form surface model |
| CN111986296B (en) * | 2020-08-20 | 2024-05-03 | 叠境数字科技（上海）有限公司 | CG animation synthesis method for bullet time |
| CN111986296A (en) * | 2020-08-20 | 2020-11-24 | 叠境数字科技（上海）有限公司 | CG animation synthesis method for bullet time |
| CN112785711B (en) * | 2021-01-21 | 2024-05-17 | 浙江科技学院 | Insulator creepage distance detection method and detection system based on three-dimensional reconstruction |
| CN112785711A (en) * | 2021-01-21 | 2021-05-11 | 浙江科技学院 | Insulator creepage distance detection method and detection system based on three-dimensional reconstruction |
| CN113076949A (en) * | 2021-03-31 | 2021-07-06 | 成都唐源电气股份有限公司 | Method and system for quickly positioning parts of contact net |
| CN113516694A (en) * | 2021-04-13 | 2021-10-19 | 同济大学 | Multi-Kinect-based dynamic visual control method for remote medical scene |
| CN113177477A (en) * | 2021-04-29 | 2021-07-27 | 湖南大学 | Target detection and identification method based on three-dimensional point cloud analysis |
| CN113469195A (en) * | 2021-06-25 | 2021-10-01 | 浙江工业大学 | Target identification method based on self-adaptive color fast point feature histogram |
| CN113469195B (en) * | 2021-06-25 | 2024-02-06 | 浙江工业大学 | Target identification method based on self-adaptive color quick point feature histogram |
| CN113408074A (en) * | 2021-06-28 | 2021-09-17 | 吉林大学 | Wheel set tread parameter measuring method and device |
| CN113971768A (en) * | 2021-10-27 | 2022-01-25 | 国网安徽省电力有限公司马鞍山供电公司 | Unmanned aerial vehicle-based three-dimensional dynamic detection method for power transmission line illegal building |
| CN113947729A (en) * | 2021-11-12 | 2022-01-18 | 华南农业大学 | Plant point cloud data depth image processing method and system based on competitive network |
| CN113808133A (en) * | 2021-11-19 | 2021-12-17 | 西南交通大学 | Subway brake shoe fault detection method based on three-dimensional point cloud |
| CN115205354B (en) * | 2022-06-23 | 2023-04-07 | 中国人民解放军国防科技大学 | Phased array laser radar imaging method based on RANSAC and ICP point cloud registration |
| CN115205354A (en) * | 2022-06-23 | 2022-10-18 | 中国人民解放军国防科技大学 | Phased array laser radar imaging method based on RANSAC and ICP point cloud registration |
| CN115063458A (en) * | 2022-07-27 | 2022-09-16 | 武汉工程大学 | Material pile volume calculation method based on three-dimensional laser point cloud |
| CN115953604A (en) * | 2023-03-13 | 2023-04-11 | 泰安市金土地测绘整理有限公司 | Real estate geographic information mapping data acquisition method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107123161A (en) | A kind of the whole network three-dimensional rebuilding method of contact net zero based on NARF and FPFH | |
| CN112581457B (en) | Pipeline inner surface detection method and device based on three-dimensional point cloud | |
| CN107093205B (en) | A kind of three-dimensional space building window detection method for reconstructing based on unmanned plane image | |
| WO2020083407A1 (en) | Three-dimensional finger vein feature extraction method and matching method therefor | |
| Zhou et al. | Building detection in Digital surface model | |
| CN103234969B (en) | Method for measuring fabric weft density based on machine vision | |
| CN108564616A (en) | Method for reconstructing three-dimensional scene in the rooms RGB-D of fast robust | |
| CN109816664B (en) | Three-dimensional point cloud segmentation method and device | |
| CN103914837B (en) | A kind of cylindrical neighborhood being applicable to various visual angles points cloud processing and searching method thereof | |
| CN114972459B (en) | Point cloud registration method based on low-dimensional point cloud local feature descriptor | |
| CN106204557A (en) | A kind of extracting method of the non-complete data symmetrical feature estimated with M based on extension Gaussian sphere | |
| CN113628263A (en) | Point cloud registration method based on local curvature and neighbor characteristics thereof | |
| CN105387814A (en) | Automatic measurement system for surface parameters of prefabricated component of three-dimension braiding composite material | |
| CN114648445B (en) | Multi-view high-resolution point cloud splicing method based on feature point extraction and fine registration optimization | |
| CN109993747A (en) | Merge the rapid image matching method of dotted line feature | |
| CN111862315A (en) | Human body multi-size measuring method and system based on depth camera | |
| CN116358449A (en) | Aircraft rivet concave-convex amount measuring method based on binocular surface structured light | |
| CN112884886A (en) | Three-dimensional point cloud pipeline extraction and modeling method capable of self-adapting to search radius | |
| CN103968781B (en) | High accuracy fast phase unwrapping method based on structure limit | |
| Yihong et al. | Fabric defect detection with optimal Gabor wavelet based on radon | |
| CN104050639A (en) | Multi-view dense point cloud data fusion method based on two-sided filter | |
| Li et al. | A center-line extraction algorithm of laser stripes based on multi-Gaussian signals fitting | |
| CN105783785A (en) | Wavelet-ridge phase extraction method | |
| Wang et al. | Mapping road based on multiple features and B-GVF snake | |
| Klose et al. | Reconstructing Shape and Motion from Asynchronous Cameras. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170901 |
|
| WD01 | Invention patent application deemed withdrawn after publication |