CN110473239A - A kind of high-precision point cloud registration method of 3 D laser scanning - Google Patents
A kind of high-precision point cloud registration method of 3 D laser scanning Download PDFInfo
- Publication number
- CN110473239A CN110473239A CN201910731854.XA CN201910731854A CN110473239A CN 110473239 A CN110473239 A CN 110473239A CN 201910731854 A CN201910731854 A CN 201910731854A CN 110473239 A CN110473239 A CN 110473239A
- Authority
- CN
- China
- Prior art keywords
- point
- registration
- cloud
- value
- point cloud
- 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.)
- Withdrawn
Links
- 238000000034 method Methods 0.000 title claims abstract description 101
- 238000000605 extraction Methods 0.000 claims abstract description 26
- 230000009466 transformation Effects 0.000 claims abstract description 25
- 241001269238 Data Species 0.000 claims abstract description 15
- 238000013519 translation Methods 0.000 claims description 42
- 238000013459 approach Methods 0.000 claims description 35
- 239000013598 vector Substances 0.000 claims description 17
- 239000011159 matrix material Substances 0.000 claims description 13
- 230000008569 process Effects 0.000 claims description 10
- 239000000284 extract Substances 0.000 claims description 7
- 244000003240 Caesalpinia gilliesii Species 0.000 claims description 3
- 235000014161 Caesalpinia gilliesii Nutrition 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 241000208340 Araliaceae Species 0.000 claims 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims 1
- 235000003140 Panax quinquefolius Nutrition 0.000 claims 1
- 235000008434 ginseng Nutrition 0.000 claims 1
- 239000006185 dispersion Substances 0.000 abstract description 4
- 238000012545 processing Methods 0.000 description 9
- 238000011161 development Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 230000006872 improvement Effects 0.000 description 4
- 238000005452 bending Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 230000002452 interceptive effect Effects 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 208000032005 Spinocerebellar ataxia with axonal neuropathy type 2 Diseases 0.000 description 1
- 208000033361 autosomal recessive with axonal neuropathy 2 spinocerebellar ataxia Diseases 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012790 confirmation Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B11/00—Measuring arrangements characterised by the use of optical techniques
- G01B11/24—Measuring arrangements characterised by the use of optical techniques for measuring contours or curvatures
-
- 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
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
-
- 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)
- Computer Vision & Pattern Recognition (AREA)
- Theoretical Computer Science (AREA)
- Length Measuring Devices By Optical Means (AREA)
Abstract
A kind of high-precision point cloud registration method of 3 D laser scanning provided by the invention, it proposes to be registrated the strategy that two steps are walked from preposition rough registration to iteration essence, the dispersion point cloud of different perspectives is obtained by obtaining characteristic point after the feature extraction based on curvature to 3 D laser scanning, to matched characteristic point to resolving coordinate transformation parameter, obtain preposition rough registration result, the case where for two point cloud datas being not subset relation, then the same place that mistake is rejected using the condition of distance restraint is iterated essence registration and obtains final registration result.Feature point extraction is quick and precisely, Feature Points Matching accuracy is high, speed is fast, rough registration provides preferable initial position for essence registration, iteration essence is registered in iterate to certain number after, accurate result can reach very high precision, and registration accuracy can achieve grade, complicated point cloud data can be used, meet engineer application requirement.Point cloud registering speed and precision greatly improves, and method for registering is applied widely.
Description
Technical field
The present invention relates to a kind of high-precision point cloud registration method, in particular to the high-precision dot cloud of a kind of 3 D laser scanning
Method for registering belongs to three-dimensional laser point cloud technical field.
Background technique
Three-dimensional laser scanning technique does not need directly to contact object, so that it may body surface three-dimensional coordinate is quickly obtained, it should
Technology obtained extensive development in recent years and used, and became the high-new skill of one great development prospect of new era survey field
Art, three-dimensional laser scanning technique carry out threedimensional model building, and main flow is divided to field data collection and data processing two sides
Face, with the fast development of 3 D laser scanning hardware, field data acquisition obtains accurate three-dimensional coordinate and no longer there is technology
On difficulty, and data processing then become technology development key.
Due to the limitation at visual angle, the prior art can not be scanned to have obtained only by a survey station to target object
Whole data information needs to be arranged multiple survey stations.The coordinate system of each independent survey station is also independent, the different independent survey stations of conversion
The point cloud data of acquisition is into the same coordinate system, that is, the point cloud registering of 3 D laser scanning.Point cloud registering is that three-dimensional is built
The essential step of mould finds suitable method, improves the precision of cloud data registration, is the key that improve three-dimensional modeling precision,
It is also the key of industry processing technique in three-dimensional laser scanning technique, the accuracy of point cloud registering determines that can subsequent modeling work
It carries out, decides the accuracy of modeling.
The point cloud registration method of the prior art has autoregistration method, target registration method and man-machine interactive registration method etc..From
Dynamic registration is not need artificial interference, according to certain algorithms, Computer Automatic Recognition overlapping region, and it is automatic in overlapping region
Selected element realizes registration to progress registration parameter resolving, and since point cloud data is more complicated, quality is different, matches automatically
Quasi- applicable elements tend not to meet various complex environments, and the point cloud data of magnanimity not only increases the difficulty of registration, also
Increase the time of autoregistration, registration accuracy according to different data there are it is different as a result, with Quasi velosity slow, registration effect
Difference；Target, which is registered in scanning area, is arranged target, uses target to be resolved as homonymy matching point in later data processing,
Target registration method is convenient for finding homonymy matching point, but this method is in place and discomfort certain inconvenient or that cannot build target
With applicable surface is very narrow；Man-machine interactive registration method does not set up target in field data acquisition, artificial in data processing
Characteristic point is visually found, this method field data is convenient and simple when acquiring, but selects homonymy matching point inaccurate when interior industry, and
And for certain complex datas, such as terrain data etc. is not easy to visually find same place, and registration accuracy is not high, manually visualizes
It is very low to find feature point efficiency.
And in modern technical field of mapping, the numerous areas such as body surface three-dimensional modeling, deformation monitoring are all answered extensively
The data processing in later period is directly influenced with the effect of point cloud registering, registration.Therefore, seek more scientific square point cloud registering side
It is a highly important job that method, which meets the needs of Modern Surveying & Mapping,.
The prior art whether which kind of point cloud registration method, all exists clearly disadvantageous, is suitable for some environment but not
It is able to satisfy institute's some need, some method for registering do not have enough theoretical and experiment support, and registration accuracy rate is very low, serious to reduce
The quality of follow-up work；Some is selected in overlapping region automatically according to certain algorithms, Computer Automatic Recognition overlapping region
It selects a little to registration parameter resolving registration is carried out, since point cloud data is more complicated, quality is different, and applicable elements are not able to satisfy
The point cloud data of various complex environments, magnanimity not only increases the difficulty of registration, also adds the time of autoregistration, registration
Speed is slow, and registration effect is poor；Some need builds target, applicable surface it is very narrow so；Some need is manually participated in the overall process, such as
It manually visualizes and finds characteristic point, labor workload is very big, and efficiency is very low, and registration accuracy is not high, cannot make to complicated point cloud data
With not being able to satisfy engineer application requirement.
Summary of the invention
In view of the deficiencies of the prior art, the high-precision point cloud registration method of a kind of 3 D laser scanning provided by the invention,
It proposes that two steps being registrated from preposition rough registration to iteration essence walk strategy, the dispersion point cloud of different perspectives is obtained to 3 D laser scanning
By obtaining characteristic point after the feature extraction based on curvature, to matched characteristic point to coordinate transformation parameter is resolved, obtain preposition
Rough registration provides more accurate initial position as a result, being registrated for successive iterations essence, is not the feelings of subset relation for two point cloud datas
Condition rejects the same place of mistake using the condition of distance restraint, is then iterated essence registration and obtains final registration result.Pass through
Measured data compares, and quick and precisely, Feature Points Matching accuracy is high, and speed is fast for feature point extraction, and rough registration is essence registration
Preferable initial position is provided, iteration essence, which is registered in, iterates to certain number, and after about 75 times, root mean square difference tends to be steady
Fixed, accurate result can reach very high precision, and registration accuracy can achieve grade, can use complicated point cloud data, meet
Engineer application requirement.Point cloud registering speed and precision greatly improves, and method for registering is applied widely, and 3 D laser scanning point cloud is matched
Quasi- processing speed is faster, precision is more quasi-, the degree of automation is higher.
To reach the above technical effect, the technical solution adopted in the present invention is as follows:
A kind of high-precision point cloud registration method of 3 D laser scanning, process includes following committed step:
The first step, point cloud characteristic point obtain, and point cloud characteristic point is obtained to be extracted using the point feature based on curvature；
Second step, Data Matching feature point search, scans for the feature point set of extraction, searches out the spy of Data Matching
Point is levied, finds out three pairs or more of homonymy matching point in principle；
Third step, constraint condition check, and are confirmed using equidistant constraint to matching double points；
4th step, preposition rough registration carry out registration calculating according to the homonymy matching point selected；
5th step deletes false same place, guarantees that iteration essence reaches enough precision with quasi-convergence and iteration；
6th step, iteration essence registration carry out closest approach again to the data after preposition rough registration and improve iteration registration resolving, obtain
To final high-precision dot cloud registration result.
A kind of high-precision point cloud registration method of 3 D laser scanning, further, including cloud characteristic point obtains, data
Matching characteristic point search, constraint condition check, preposition rough registration, delete false same place, iteration essence registration, and preposition rough registration only mentions
It takes key point to be registrated, after carrying out rotation and translation to target to be registered, obtains just being registrated achievement, preposition rough registration should not
High registration accuracy is sought, accurate initial position is only provided, the data after preposition rough registration are carried out with closest approach again and improves iteration
Algorithm essence registration obtains final registration result after second of rotation translation transformation；
It in the transformation of preposition rough registration, is resolved with characteristic point as key point, the quantity of characteristic point is much smaller than original
The quantity of data, preposition rough registration use the registration mode according to feature, minimum of three match point are only needed according to rigid body translation
Registration parameter can be calculated, but is considered based on accuracy, obtaining more points participation resolving can avoid Feature Points Matching mistake of the same name
Accidentally the case where；
In iteration essence registration, closest approach standard iterative algorithm requires there must be inclusion relation, i.e., one between two point sets
A point set must be the subset of another point set, need to judge after preposition rough registration, if registration target is specific single and surrounding
Nothing significantly interferes with environment, while meeting the subset that a point set is another point set, and the calculation of closest approach standard iteration can be used directly
Method is registrated, if inclusion relation is not present in two point sets after preposition rough registration, is needed selection overlapping region to be resolved, is overlapped
Region can be selected according to limit value condition, can also manually be chosen, if selecting according to matching characteristic point, multiselect takes some matchings
Characteristic point obtains more accurate region and is iterated smart registration.
A kind of high-precision point cloud registration method of 3 D laser scanning, further, point cloud characteristic point are obtained to use and be based on
The point feature extracting method of curvature approaches the field of the point there are a curved surface Z=a (X, Y) at each point in cloud
The curvature of point cloud, the local surface that can be fitted with this point and its field point at the certain point in cloud indicates, uses
Least square method makes the distance value of somewhere point cloud and surrounding point to parametric quadric surface minimum, indicates partial zones using quadratic surface
Domain, important curvature include principal curvatures B1、B2, average curvature C and Gaussian curvature D,
Curvature according to point cloud carries out the average curvature C that characteristic point is calculated in characteristic extraction procedure and D pairs of Gaussian curvature
Data Matching feature point search in later plays an important role, and the point feature extraction and application part sags and crests based on curvature select special
Point is levied, if certain point is some point part salient point or concave point, then it is assumed that it is a cloud characteristic point；
Using the principle of local sags and crests selection characteristic point, there is following formula:
In formula, the value of E (g) as judge whether be characteristic point standard, calculate certain point giE (g) value of field point, point
It is not denoted as E (gi1), E (gi2) ... ..., E (gin), if point giCalculated value E (gi) it is maximum value in the point calculated value of field,
That is E (gi) > max (E (gi1), E (gi2) ... ..., E (gin)), then it is assumed that the point is local salient point；If point giCalculated value E
(gi) it is minimum value in the point calculated value of field, i.e. E (gi) < min (E (gi1), E (gi2) ... ..., E (gin)), then it is assumed that this point
It is local concave point, all local salient point drawn game portion concave points is extracted, is required point cloud characteristic point.
A kind of high-precision point cloud registration method of 3 D laser scanning, further, in Data Matching feature point search, warp
The data for crossing some cloud feature point extractions obtain two feature point sets F1 and F2, calculate every bit principal curvatures B1、B2When calculate simultaneously
The average curvature C and Gaussian curvature D of point cloud, using the average curvature C of obtained characteristic point and Gaussian curvature D as an attribute
Sequence Jf=(Di, Ci), sequence of attributes J is created to the every bit in feature point set F2f=(Di, Ci), and carried out one by one in F1
Search, looks for a kind of similarity function, chooses different threshold values to similitude according to different situations and judge, find with phase
Like attribute point as match point.
A kind of high-precision point cloud registration method of 3 D laser scanning, further, the tool of Data Matching feature point search
Body step are as follows:
Step 1 calculates average curvature C and Gaussian curvature D in the feature point set F1 and F2 of extraction at each point, and creates
Build the sequence of attributes H of the pointf1iAnd Hf2i；
Step 2, to each of F2 point f2I finds the point f of the sequence of attributes having the same in F11I utilizes one
A discriminate, differentiate both whether sequence of attributes having the same, using Tuo Nimotuo coefficient, this step be converted into calculating Tuo Nimo
Hold in the palm coefficient value Li；
Step 3 scans for calculated Li value each in F2, minimum value therein is searched for, if minimum value is less than
The threshold value k of a certain setting, then the corresponding point of the value meets the condition of match point as matching double points；If search finish still without
The point for meeting threshold value then continues to scan for point next in F2, and same method is until completing all search；
Step 4 after search, completes the selection of match point；
Matched sequence of attributes is determined, the present invention uses Tuo Nimotuo coefficient, which regards one as sequence of attributes
A vector, Tuo Nimotuo coefficient consider the difference of the length and angle of two vectors simultaneously, if difference very little, theirs is similar
Property is higher；Specific Tuo Nimotuo coefficient formula is as follows:
M and N respectively represents attribute vector, and Tuo Nimotuo coefficient L meets the point of some threshold value k to as required.
A kind of high-precision point cloud registration method of 3 D laser scanning further during constraint condition checks, chooses good
It is Q1 and Q2 with the point set after point, determines that match point is correctly, using equidistant constraint to Q1 using constraint condition
Confirmed with the matching double points in Q2, for two pairs of points pair in Q1 and Q2WithP is distance,
If correctly point pair, then need to meet
If a pair of point pairWithR is setting value, meets following formula:
Then think a little pairWith point pairIt is the matching double points for meeting constraint condition.
A kind of high-precision point cloud registration method of 3 D laser scanning, further, preposition rough registration is using no characteristic point
Method for registering, calculate rotation parameter R and translation parameters T through rigid body translation, rigid body translation uses four element group methods, to process
It is equidistant the garbled point set Q1 and Q2 of constraint condition, the mass center for calculating each point set is origin, defines basis matrix H, uses
Four element groups calculate rotation parameter R and translation parameters T.
A kind of high-precision point cloud registration method of 3 D laser scanning, further, preposition rough registration circular
Are as follows:
Vector Q=(w, x, y, z) indicates a four element groups, if a+bi+cj+dk meets i2+j2+k2=ijk=-
1, wherein a, b, c, d are real number, and i, j, k are imaginary parts, and a+bi+cj+dk is then referred to as real quaternion, if the modulus value of the four elements group
It is 1, then is four element group of unit.
If Q=(w, x, y, z) indicates four element groups of R spin matrix, the spin matrix of rigid body translation is indicated with Q are as follows:
Wherein meet w2+x2+y2+z2=1 and w >=0；
After calculating spin matrix using the method for above-mentioned formula description, seven element representations of rigid body translation, i.e., (w, x, y,
Z, x0, y0, z0), it indicates are as follows:
After calculating rotation parameter R and translation parameters T, the result of resolving applies target point mysorethorn, realizes to point cloud data
Preposition rough registration, preposition rough registration obtained a relatively accurate initial position.
A kind of high-precision point cloud registration method of 3 D laser scanning, further, the method for deleting false same place are as follows:
Step 1, it is inevitable not belong to if that establishes is o'clock excessive between the distance two same places if two point cloud datas are U and V
It is false same place in overlapping region；
In above formula, δ is specified distance threshold, is determined using the size of sampling interval value, if CiValue be less than δ, then for
1, it is in overlapping region；It is on the contrary then be denoted as 0, it indicates false same place, is deleted；
Step 2 is closer for having there is two point cloud datas of fine initial position between same place, using being based on
The screening technique of statistics；
In above formula, the selection of ε is determined according to error in the distance between same place mean value and distance, CiValue is greater than ε threshold
The point of value, it is believed that be false same place, delete false same place.
A kind of high-precision point cloud registration method of 3 D laser scanning, further, iteration essence registration are changed using closest approach
Into iteration registration Algorithm, it is assumed that two point cloud datas U and V, two data come from unused coordinate system, find sky of the U relative to V
Between convert, be registrated U and V spatially；
It is a kind of optimal algorithm based on least square that closest approach, which improves iteration registration Algorithm, is converted each time all as rigid body
Transformation, just stops until the error of iteration meets threshold value, carries out rotation and translation to the point that U and V two o'clock is concentrated, calculates rotation
Parameter R and translation parameters T, closest approach improves iteration registration Algorithm, and specific step is as follows:
Step 1, when iteration starts, if initial position U0=U, V0=[1 0 00 00 0]T, k=0；
Step 2, most close point set U ' is calculatedk=C (Uk, V), wherein C is the operation sought a little pair, U 'kFor corresponding point set
It closes；
Step 3, registration vector is calculated；
Step 4, using registration vector, transformation point set U transformation obtains new position UK+1；
Step 5, if transformed error meets the threshold value of setting, iteration ends obtain final high-precision point cloud registering
As a result.
Compared with the prior art, the advantages of the present invention are as follows:
1. a kind of high-precision point cloud registration method of 3 D laser scanning provided by the invention, propose from preposition rough registration to
Two steps of iteration essence registration walk strategy, and the dispersion point cloud for obtaining different perspectives to 3 D laser scanning passes through the feature based on curvature
Characteristic point is obtained after extraction, to matched characteristic point to coordinate transformation parameter is resolved, obtains preposition rough registration as a result, changing to be subsequent
The case where generation essence registration provides more accurate initial position, for two point cloud datas is not subset relation, utilizes the item of distance restraint
Part rejects the same place of mistake, is then iterated essence registration and obtains final registration result.It is compared by measured data, it is special
Sign point extracts quick and precisely, and Feature Points Matching accuracy is high, and speed is fast, and rough registration provides preferable initial bit for essence registration
It sets, iteration essence, which is registered in, iterates to certain number, and after about 75 times, root mean square difference tends towards stability, and accurate result can reach
Very high precision, registration accuracy can achieve grade.
2. a kind of high-precision point cloud registration method of 3 D laser scanning provided by the invention is applied to work for the algorithm
Journey measurement and later period modeling etc. numerous areas, the limitation of existing algorithm are not able to satisfy practice needs, improvement project master
The strategy that line is walked using preposition rough registration and essence two steps of registration, preposition rough registration simultaneously do not have to all acquisition data, only extract wherein
Key point be registrated, after carrying out rotation and translation to target to be registered, obtain just being registrated achievement, rough registration is not required for
Very high registration accuracy only provides a preferable initial position, has both met the requirement of smart registration Algorithm, while can also mention
Data after rough registration are carried out smart registration, the scheme that improved two step is walked not only substantially mentions by the speed of high later period registration again
It is high to match Quasi velosity, and ensure that quality of registration, complicated point cloud data can be used, meet engineer application requirement.
3. a kind of high-precision point cloud registration method of 3 D laser scanning provided by the invention, point cloud characteristic point, which obtains, to be used
Point feature based on curvature is extracted, and the information complete and accurate of extraction, the extracting method speed of service is fast, is mentioned for matching characteristic point search
For supporting condition, it is accurate that Feature Points Matching facilitates, and the less characteristic point extracted using the method can be resolved in later period rough registration
Convenience and high-efficiency when rotating translation parameters, still needs to delete pseudo- same place.Matching is determined using constraint condition before preposition rough registration
Point is correctly, to ensure that the quality and efficiency of registration, and deleting false same place ensure that iteration essence reaches with quasi-convergence and iteration
Enough precision.
4. a kind of high-precision point cloud registration method of 3 D laser scanning provided by the invention, iteration essence registration is using nearest
The transformation that point improves iteration registration Algorithm is very accurate, as long as Registration of Measuring Data works well, obtained registration accuracy is very high
, it does not need to be split data in registration and feature extraction, does not need to set up target yet or do in data acquisition
Label, have preferable initial position by two point cloud datas of preposition rough registration, with closest approach improve iteration registration Algorithm into
Good convergence can be obtained in row registration.Point cloud registering speed and precision greatly improves, and method for registering is applied widely, and three-dimensional swashs
Optical scanning point cloud registering processing speed is faster, precision is more quasi-, the degree of automation is higher.
Detailed description of the invention
The step of Fig. 1 is a kind of high-precision point cloud registration method of 3 D laser scanning provided by the invention figure.
The step of Fig. 2 is Data Matching feature point searching method provided by the invention figure.
Fig. 3 is the step of closest approach provided by the invention improves iteration registration Algorithm figure.
Specific embodiment
With reference to the accompanying drawing, to a kind of technology of the high-precision point cloud registration method of 3 D laser scanning provided by the invention
Scheme is further described, and so that those skilled in the art is better understood the present invention and can be practiced.
Referring to Fig. 1, a kind of high-precision point cloud registration method of 3 D laser scanning provided by the invention, including point Yun Tezheng
Point acquisition, Data Matching feature point search, constraint condition check, preposition rough registration, delete false same place, iteration essence registration, preceding
It sets rough registration and only extracts key point and be registrated, after carrying out rotation and translation to target to be registered, obtain just being registrated achievement, it is preceding
It sets rough registration and does not require high registration accuracy, only provide accurate initial position, both met closest approach and improved wanting for iterative algorithm
It asks, while also improving the later period with Quasi velosity, the data after preposition rough registration are carried out with closest approach improvement iterative algorithm essence again and is matched
Standard obtains final registration result after second of rotation translation transformation.
It in the transformation of preposition rough registration, is resolved with characteristic point as key point, the quantity of characteristic point is much smaller than original
The quantity of data, it is ensured that the algorithm speed of preposition rough registration, the acquisition of point feature and matching characteristic point be really in preposition rough registration
Surely it is key, extracts accurate matching characteristic point to the core of preposition rough registration precision, preposition rough registration is used according to feature
Registration mode, only needed minimum of three match point that can calculate registration parameter according to rigid body translation, but examine based on accuracy
Consider, obtains more points and participate in resolving the situation that can avoid Feature Points Matching mistake of the same name.
A kind of high-precision point cloud registration method of 3 D laser scanning, detailed process includes following committed step:
The first step, point cloud characteristic point obtain, and point cloud characteristic point is obtained to be extracted using the point feature based on curvature；
Second step, Data Matching feature point search, scans for the feature point set of extraction, searches out the spy of Data Matching
Point is levied, finds out three pairs or more of homonymy matching point in principle, but in view of improving preposition rough registration precision, selection that can be more as far as possible
Characteristic point is resolved；
Third step, constraint condition check, and are confirmed using equidistant constraint to matching double points；
4th step, preposition rough registration carry out registration calculating according to the homonymy matching point selected；
5th step deletes false same place, guarantees that iteration essence reaches enough precision with quasi-convergence and iteration；
6th step, iteration essence registration carry out closest approach again to the data after preposition rough registration and improve iteration registration resolving, obtain
To final high-precision dot cloud registration result.
Accurate initial position is obtained after preposition rough registration, might not be able to satisfy wanting for iteration essence registration Algorithm
It asks, requires there must be inclusion relation between two point sets in closest approach standard iterative algorithm, i.e. in addition a point set must be
The subset of one point set needs to judge after preposition rough registration, if registration target is specific single and surrounding is without environment is significantly interfered with, together
When meet a point set be another point set subset, closest approach standard iterative algorithm can be used directly and be registrated, registration
Precision is higher, if inclusion relation is not present in two point sets after preposition rough registration, cannot directly be calculated using closest approach standard iteration
Method is registrated, and need to be chosen overlapping region and be resolved, and overlapping region can be selected according to limit value condition, can also manually be selected
It takes, if selecting according to matching characteristic point, multiselect takes some matching characteristic points, obtains more accurate region and be iterated essence matching
It is quasi-.
One, point cloud characteristic point obtains
The characteristic point of three-dimensional laser point cloud is the point set that can indicate the geometric error modeling characteristic of object, puts the correct of cloud characteristic point
The description three-dimensional model information for being that the few point of the amount of using up is more as far as possible is chosen, the accuracy of subsequent registration work can either be promoted,
And less point participates in resolving and capable of improving computational efficiency.Therefore point cloud characteristic point acquisition is not only the important of Point Cloud Processing
Step, while being also one step of key of the high-precision point cloud registration method of 3 D laser scanning, point cloud characteristic point, which obtains, uses base
In the point feature extracting method of curvature.
The bending degree of a certain curve of curvature representation, there are a curved surface Z=a (X, Y) at each point in cloud,
The field point cloud for approaching the point can use the local surface that this point and its field point fit at the certain point in cloud
Curvature indicates.Make the distance value of somewhere point cloud and surrounding point to parametric quadric surface minimum using least square method, use is secondary
Representation of a surface regional area.Important curvature includes principal curvatures B1、B2, average curvature C and Gaussian curvature D, average curvature C are locally retouched
State the curvature that a certain curved surface falls into surrounding space；Gaussian curvature D reflect curved surface bending situation, Gaussian curvature D variation comparatively fast compared with
Illustrate that the smoothness of the curved surface is lower when big.
Since the characteristic point of extraction only carries out preposition rough registration, the resolving of preposition rough registration only needs minimum of three point just
It can carry out, so there is no need to largely put to participate in preposition rough registration, only need to extract general profile meets preposition rough registration requirement,
And average curvature C and Gaussian curvature D that characteristic point is calculated in characteristic extraction procedure, average curvature are carried out according to the curvature of cloud
C and Gaussian curvature D plays an important role for Data Matching feature point search later, and follow-up work needs to use average curvature C
With Gaussian curvature D, the point feature extraction and application part sags and crests based on curvature select characteristic point, if certain point is some point part
Salient point or concave point, then it is assumed that it is a cloud characteristic point.
Using the principle of local sags and crests selection characteristic point, there is following formula:
In formula, the value of E (g) as judge whether be characteristic point standard, calculate certain point giE (g) value of field point, point
It is not denoted as E (gi1), E (gi2) ... ..., E (gin), if point giCalculated value E (gi) it is maximum value in the point calculated value of field,
That is E (gi) > max (E (gi1), E (gi2) ... ..., E (gin)), then it is assumed that the point is local salient point；If point giCalculated value E
(gi) it is minimum value in the point calculated value of field, i.e. E (gi) < min (E (gi1), E (gi2) ... ..., E (gin)), then it is assumed that this point
It is local concave point, all local salient point drawn game portion concave points is extracted, is required point cloud characteristic point.
Two, Data Matching feature point search
The data of passing point cloud feature point extraction obtain two feature point sets F1 and F2, the purpose of match point cloud characteristic point
It is the common point in order to find two o'clock cloud, then carries out rotation by the way of foundation feature registration and translate preposition rough registration, with
Obtain accurate initial position.
When cloud characteristic point obtains, the present invention uses the point feature extracting method based on curvature, calculates every bit principal curvatures
B1、B2When calculate the average curvature C and Gaussian curvature D of a cloud simultaneously, utilize the average curvature C and Gauss of obtained characteristic point bent
Rate D is as a sequence of attributes Jf=(Di, Ci), even if the point coordinate of same position is different in two o'clock cloud, but the variation of coordinate is not
It can cause the difference of average curvature C and Gaussian curvature D, therefore the same point under different coordinates in two o'clock cloud must have identical attribute sequence
Arrange Jf=(Di, Ci), so creating sequence of attributes J to the every bit in feature point set F2f=(Di, Ci), and in F1 one by one into
Row search, finds the point with same alike result as match point.
In practice, it is impossible to which there are corresponding average curvature C and Gaussian curvature D identical under theoretical case, also not
There may be identical sequence of attributes pair, thus cannot according to the identical carry out matching double points selection of sequence of attributes,
A kind of similarity function is looked for, different threshold values is chosen to similitude according to different situations and is judged.
Referring to fig. 2, the specific steps of Data Matching feature point search are as follows:
Step 1 calculates average curvature C and Gaussian curvature D in the feature point set F1 and F2 of extraction at each point, and creates
Build the sequence of attributes H of the pointf1iAnd Hf2i；
Step 2, to each of F2 point f2I finds the point f of the sequence of attributes having the same in F11I utilizes one
A discriminate, differentiate both whether sequence of attributes having the same, using Tuo Nimotuo coefficient, this step is converted into design factor value
Li；
Step 3 scans for calculated Li value each in F2, minimum value therein is searched for, if minimum value is less than
The threshold value k of a certain setting, then the corresponding point of the value meets the condition of match point as matching double points；If search finish still without
The point for meeting threshold value then continues to scan for point next in F2, and same method is until completing all search；
Step 4 after search, completes the selection of match point；
Matched sequence of attributes is determined, the present invention uses Tuo Nimotuo coefficient, which regards one as sequence of attributes
A vector, Tuo Nimotuo coefficient consider the difference of the length and angle of two vectors simultaneously, if difference very little, theirs is similar
Property is higher；Specific Tuo Nimotuo coefficient formula is as follows:
M and N respectively represent attribute vector.Meet the point of some threshold value k to as required.
Three, constraint condition checks
For feature point set F1 and F2, the point set chosen after match point is Q1 and Q2, carries out preposition rough registration to it
Before resolving, it is necessary to determine that match point is correctly, using equidistant constraint in Q1 and Q2 first with constraint condition
Matching double points are confirmed, for two pairs of points pair in Q1 and Q2WithP is distance, if correctly
Point pair, then need to meet
If a pair of point pairWithR is setting value, meets following formula:
Then it is believed that point pairWith point pairIt is the matching double points for meeting constraint condition.
Four, preposition rough registration
It obtains after the match point of confirmation, starts to carry out preposition rough registration, preposition rough registration uses matching without characteristic point
Quasi- method calculates rotation parameter R and translation parameters T through rigid body translation, and rigid body translation uses four element group methods.For front
By being equidistant the garbled point set Q1 and Q2 of constraint condition, the mass center for calculating each point set is origin, defines basis matrix
H calculates rotation parameter R and translation parameters T with four element groups.
Circular are as follows:
Registration is the coordinate transform that required coordinate system is moved to a certain rotation of coordinates, if P point in S1 (scan1) and
The coordinate of S2 (scan2) is respectively P1(x1, y1, z1)TAnd P2(x2, y2, z2)T, coordinate of the P point in S1 is P1, carry out coordinate rotation
Turn the coordinate P that peace transfer is changed in S22。
Wherein x0, y0, z0It is the translation parameters in reference axis respectively, R is rotational transformation matrix, and the process of point cloud registering is
The process of one rigid body translation is to find the R and T for meeting condition.
Vector Q=(w, x, y, z) indicates a four element groups, if a+bi+cj+dk meets i2+j2+k2=ijk=-
1, wherein a, b, c, d are real number, and i, j, k are imaginary parts, and a+bi+cj+dk is then referred to as real quaternion, if the modulus value of the four elements group
It is 1, then is four element group of unit.
If Q=(w, x, y, z) indicates four element groups of R spin matrix, the spin matrix of rigid body translation is indicated with Q are as follows:
Wherein meet w2+x2+y2+z2=1 and w >=0.
After calculating spin matrix using the method for above-mentioned formula description, rigid body translation can use seven element representations, i.e., (w,
X, y, z, x0, y0, z0), it indicates are as follows:
Four element group methods not only do not need complicated calculations, but also can freely convert, and are a kind of more handy representations.
After calculating rotation parameter R and translation parameters T, the result of resolving applies target point mysorethorn, realizes to point cloud data
Preposition rough registration, preposition rough registration obtained a relatively accurate initial position, creates for successive iterations essence registration good
Good condition.
Five, false same place is deleted
Iteration essence registration improves iteration registration Algorithm using closest approach, improves iteration registration Algorithm in closest approach and is iterated
Before essence registration, judge the two o'clock cloud for participating in iteration essence registration with the presence or absence of inclusion relation, i.e., one of point cloud data whether be
The subset of another point cloud data.If subset, then the preposition rough registration data completed to it directly carry out closest approach and change
It is registrated into iteration registration Algorithm iteration essence, obtains final transformation results, and carry out transformation registration.
If not subset, iteration registration Algorithm directly cannot be improved with closest approach and be iterated smart registration, although two o'clock cloud
Inclusion relation is not present between data, but there will necessarily be overlapping region therebetween, extracts lap between the two, it is right
Overlapping region improves iteration registration Algorithm using closest approach and is iterated smart registration, and obtained registration parameter is suitable for global point
Cloud data then do whole rotation translation transformation and obtain registration result to the end.
Whether the point cloud data that there is inclusion relation is iterated smart registration, or the point cloud containing the region that overlaps
Data are iterated smart registration, all without establishing point to relationship to all the points, establish one in two o'clock cloud using KNN searching algorithm
Quantitative corresponding dot pair, then foundation puts quadratic sum minimum of adjusting the distance and establishes error equation, solves conversion parameter.
For partly overlapping point cloud data, the selection of overlapping region is the process of deletion error same place, is called vacation
Same place.Since there are noise spots and outlier, even if still needing to delete false same in the point cloud data there are inclusion relation
Famous cake, deleting false same place just can guarantee that iteration convergence and iteration reach enough precision.
The method for deleting false same place is as follows:
Step 1, if two point cloud datas are U and V, by the relative position the U and V very little of preposition rough registration, if establishing
It is o'clock excessive between the distance two same places, then be necessarily not belonging to overlapping region, be false same place；
In above formula, δ is specified distance threshold, is determined using the size of sampling interval value, if CiValue be less than δ, then for
1, it is in overlapping region；It is on the contrary then be denoted as 0, it indicates false same place, is deleted；
Step 2 is closer for having there is two point cloud datas of fine initial position between same place, using being based on
The screening technique of statistics；
In above formula, the selection of ε is determined according to error in the distance between same place mean value and distance, CiValue is greater than ε threshold
The point of value, it is believed that be false same place, delete false same place.
The method for deleting false same place both can be used alone, and can also use multiple rejecting processes to data, it is ensured that leave just
True corresponding dot pair participates in calculating.
Six, iteration essence is registrated
In laser point cloud data iteration essence with punctual, closest approach improve iteration registration Algorithm assume two point cloud data U and
V, two data come from unused coordinate system, find spatial alternation of the U relative to V, U and V can be made spatially to be registrated.
As shown in above formula, it is assumed that use UiAnd ViIndicate that the point in two space point sets, the registration transformation of two point sets alignment require to make
The objective function for obtaining above formula is minimum.It is a kind of optimal algorithm based on least square, iteration that closest approach, which improves iteration registration Algorithm,
The process for carrying out optimal transformation converts all as rigid body translation each time, just stops until the error of iteration meets threshold value.According to upper
The variation for stating formula carries out rotation and translation to the point that U and V two o'clock is concentrated, calculates rotation parameter R and translation parameters T, so that
Optimal registration of two data under certain criterion.
Referring to Fig. 3, closest approach improves iteration registration Algorithm, and specific step is as follows:
Step 1, when iteration starts, if initial position U0=U, V0=[1 00000 0]T, k=0；
Step 2, most close point set U ' is calculatedk=C (Uk, V), wherein C is the operation sought a little pair, U 'kFor corresponding point set
It closes；
Step 3, registration vector is calculated；
Step 4, using registration vector, transformation point set U transformation obtains new position UK+1；
Step 5, if transformed error meets the threshold value of setting, iteration ends obtain final high-precision point cloud registering
As a result.
The transformation that iteration essence registration improves iteration registration Algorithm using closest approach is very accurate, it is a kind of based on minimum two
Multiply the iterative algorithm of optimal solution, as long as Registration of Measuring Data works well, obtained registration accuracy is very high.Closest approach improvement changes
It is a kind of no feature registration algorithm for registration Algorithm, does not need to be split data in registration and feature extraction, in data
It does not need to set up target in acquisition yet or make marks.There is preferable initial position by two point cloud datas of preposition rough registration,
Carrying out registration with closest approach improvement iteration registration Algorithm can be obtained good convergence.
A kind of high-precision point cloud registration method of 3 D laser scanning provided by the invention is proposed from preposition rough registration to repeatedly
The strategy of generation essence registration obtains the dispersion point cloud of different perspectives by obtaining after the feature extraction based on curvature to 3 D laser scanning
To characteristic point, to matched characteristic point to coordinate transformation parameter is resolved, preposition rough registration is obtained as a result, being registrated for successive iterations essence
The case where providing more accurate initial position, for two point cloud datas being not subset relation, utilizes the condition of distance restraint to reject wrong
Then same place accidentally is iterated essence registration and obtains final registration result.It is compared by measured data, feature point extraction
Quick and precisely, Feature Points Matching accuracy is high, and speed is fast, and rough registration provides preferable initial position, iteration essence for essence registration
It is registered in and iterates to certain number, after about 70 times, root mean square difference tends towards stability, and accurate result can reach very high essence
Degree, registration accuracy can achieve grade.
Claims (10)
1. a kind of high-precision point cloud registration method of 3 D laser scanning, it is characterised in that: process includes following committed step:
The first step, point cloud characteristic point obtain, and point cloud characteristic point is obtained to be extracted using the point feature based on curvature；
Second step, Data Matching feature point search, scans for the feature point set of extraction, searches out the feature of Data Matching
Point finds out three pairs or more of homonymy matching point in principle；
Third step, constraint condition check, and are confirmed using equidistant constraint to matching double points；
4th step, preposition rough registration carry out registration calculating according to the homonymy matching point selected；
5th step deletes false same place, guarantees that iteration essence reaches enough precision with quasi-convergence and iteration；
6th step, iteration essence registration carry out closest approach again to the data after preposition rough registration and improve iteration registration resolving, obtain most
Whole high-precision dot cloud registration result.
2. a kind of high-precision point cloud registration method of 3 D laser scanning according to claim 1, it is characterised in that: including
Point cloud characteristic point acquisition, Data Matching feature point search, constraint condition check, preposition rough registration, delete false same place, iteration essence
Registration, preposition rough registration only extract key point and are registrated, and after carrying out rotation and translation to target to be registered, obtain just being registrated
Achievement, preposition rough registration do not require high registration accuracy, only provide accurate initial position, again to the data after preposition rough registration
It carries out closest approach and improves iterative algorithm essence registration, obtain final registration result after second of rotation translation transformation；
It in the transformation of preposition rough registration, is resolved with characteristic point as key point, the quantity of characteristic point is much smaller than initial data
Quantity, preposition rough registration use according to feature registration mode, minimum of three match point is only needed according to rigid body translation
Registration parameter is calculated, but is considered based on accuracy, obtaining more points participation resolving can avoid Feature Points Matching mistake of the same name
Situation；
In iteration essence registration, closest approach standard iterative algorithm requires there must be inclusion relation, i.e. a point between two point sets
Collection must be the subset of another point set, need to judge after preposition rough registration, if registration target is specific single and surrounding is without bright
Aobvious interference environment, while meeting the subset that a point set is another point set, can be used directly closest approach standard iterative algorithm into
Row registration needs selection overlapping region to be resolved, overlapping region if inclusion relation is not present in two point sets after preposition rough registration
It can select, can also manually be chosen according to limit value condition, if selecting according to matching characteristic point, multiselect takes some matching characteristics
Point obtains more accurate region and is iterated smart registration.
3. a kind of high-precision point cloud registration method of 3 D laser scanning according to claim 1, it is characterised in that: point cloud
Characteristic point, which obtains, uses the point feature extracting method based on curvature, and there are a curved surface Z=a at each point in cloud
(X, Y) approaches the field point cloud of the point, the part that can be fitted with this point and its field point at the certain point in cloud
The curvature of curved surface indicates, makes the distance value of somewhere point cloud and surrounding point to parametric quadric surface minimum using least square method, adopts
Regional area is indicated with quadratic surface, and important curvature includes principal curvatures B1、B2, average curvature C and Gaussian curvature D,
Curvature according to point cloud carries out calculating the average curvature C and Gaussian curvature D of characteristic point for it in characteristic extraction procedure
Data Matching feature point search afterwards plays an important role, and the point feature extraction and application part sags and crests based on curvature select feature
Point, if certain point is some point part salient point or concave point, then it is assumed that it is a cloud characteristic point；
Using the principle of local sags and crests selection characteristic point, there is following formula:
In formula, the value of E (g) as judge whether be characteristic point standard, calculate certain point giE (g) value of field point, remembers respectively
For E (gi1), E (gi2) ... ..., E (gin), if point giCalculated value E (gi) it is maximum value in the point calculated value of field, i.e. E
(gi) > max (E (gi1), E (gi2) ... ..., E (gin)), then it is assumed that the point is local salient point；If point giCalculated value E (gi)
It is the minimum value in the point calculated value of field, i.e. E (gi) < min (E (gi1), E (gi2) ... ..., E (gin)), then it is assumed that the point is office
Portion's concave point extracts all local salient point drawn game portion concave points, is required point cloud characteristic point.
4. a kind of high-precision point cloud registration method of 3 D laser scanning according to claim 1, it is characterised in that: data
In matching characteristic point search, the data of passing point cloud feature point extraction obtain two feature point sets F1 and F2, calculate every bit master
Curvature B1、B2When calculate the average curvature C and Gaussian curvature D of a cloud simultaneously, utilize the average curvature C and height of obtained characteristic point
This curvature D is as a sequence of attributes Jf=(Di, Ci), sequence of attributes J is created to the every bit in feature point set F2f=(Di,
Ci), and scanned for one by one in F1, a kind of similarity function is looked for, chooses different thresholds to similitude according to different situations
Value is judged, finds the point with like attribute as match point.
5. a kind of high-precision point cloud registration method of 3 D laser scanning according to claim 1, Data Matching characteristic point
The specific steps of search are as follows:
Step 1 calculates average curvature C and Gaussian curvature D in the feature point set F1 and F2 of extraction at each point, and creating should
The sequence of attributes H of pointf1iAnd Hf2i；
Step 2 finds the point f1i of the sequence of attributes having the same in F1, that is, utilizes one to each of F2 point f2i
Discriminate, differentiate both whether sequence of attributes having the same, using Tuo Nimotuo coefficient, this step be converted into calculating Tuo Nimotuo
Coefficient value Li；
Step 3 scans for calculated Li value each in F2, searches for minimum value therein, if minimum value is less than a certain
The threshold value k of setting, then the corresponding point of the value meets the condition of match point as matching double points；If search is finished still without satisfaction
The point of threshold value then continues to scan for point next in F2, and same method is until completing all search；
Step 4 after search, completes the selection of match point；
For matched sequence of attributes determine, the present invention use Tuo Nimotuo coefficient, the coefficient sequence of attributes regard as one to
Amount, Tuo Nimotuo coefficient consider the difference of the length and angle of two vectors simultaneously, if difference very little, their similitude is more
It is high；Specific Tuo Nimotuo coefficient formula is as follows:
M and N respectively represents attribute vector, and Tuo Nimotuo coefficient L meets the point of some threshold value k to as required.
6. a kind of high-precision point cloud registration method of 3 D laser scanning according to claim 1, it is characterised in that: constraint
During condition checks, having chosen the point set after match point is Q1 and Q2, determines that match point is correctly, to use using constraint condition
Equidistant constraint confirms the matching double points in Q1 and Q2, for two pairs of points pair in Q1 and Q2WithP is distance, if correctly point pair, then need to meet
If a pair of point pairWithR is setting value, meets following formula:
Then think a little pairWith point pairIt is the matching double points for meeting constraint condition.
7. a kind of high-precision point cloud registration method of 3 D laser scanning according to claim 1, it is characterised in that: preposition
Rough registration uses the method for registering without characteristic point, calculates rotation parameter R and translation parameters T through rigid body translation, rigid body translation is adopted
The mass center of each point set is calculated as original to by being equidistant the garbled point set Q1 and Q2 of constraint condition with four element group methods
Point defines basis matrix H, calculates rotation parameter R and translation parameters T with four element groups.
8. a kind of high-precision point cloud registration method of 3 D laser scanning according to claim 1, it is characterised in that: preposition
Rough registration circular are as follows:
Vector Q=(w, x, y, z) indicates a four element groups, if a+bi+cj+dk meets i2+j2+k2=ijk=-1,
Middle a, b, c, d are real number, and i, j, k are imaginary parts, and a+bi+cj+dk is then referred to as real quaternion, if the modulus value of the four elements group is 1,
It is then four element group of unit；
If Q=(w, x, y, z) indicates four element groups of R spin matrix, the spin matrix of rigid body translation is indicated with Q are as follows:
Wherein meet w2+x2+y2+z2=1 and w >=0；
After calculating spin matrix using the method for above-mentioned formula description, rigid body translation seven element representations, i.e. (w, x, y, z, x0,
y0, z0), it indicates are as follows:
After calculating rotation parameter R and translation parameters T, the result of resolving applies target point mysorethorn, before realizing to point cloud data
Rough registration is set, preposition rough registration has obtained a relatively accurate initial position.
9. a kind of high-precision point cloud registration method of 3 D laser scanning according to claim 1, it is characterised in that: delete
The method of false same place is as follows:
Step 1, if that establishes is o'clock excessive between the distance two same places, is necessarily not belonging to weight if two point cloud datas are U and V
Folded region, is false same place；
In above formula, δ is specified distance threshold, is determined using the size of sampling interval value, if CiValue is less than δ, then is 1, is in
Overlapping region；It is on the contrary then be denoted as 0, it indicates false same place, is deleted；
Step 2 is closer, using based on statistics for having there is two point cloud datas of fine initial position between same place
Screening technique；
In above formula, the selection of ε is determined according to error in the distance between same place mean value and distance, CiValue is greater than ε threshold value
Point, it is believed that be false same place, delete false same place.
10. a kind of high-precision point cloud registration method of 3 D laser scanning according to claim 1, it is characterised in that: repeatedly
Generation essence registration improves iteration registration Algorithm using closest approach, it is assumed that two point cloud datas U and V, two data come from unused seat
Mark system, finds spatial alternation of the U relative to V, is registrated U and V spatially；
It is a kind of optimal algorithm based on least square that closest approach, which improves iteration registration Algorithm, converts all become for rigid body each time
It changes, just stops until the error of iteration meets threshold value, rotation and translation is carried out to the point that U and V two o'clock is concentrated, calculates rotation ginseng
Number R and translation parameters T, closest approach improves iteration registration Algorithm, and specific step is as follows:
Step 1, when iteration starts, if initial position U0=U, V0=[1 00000 0]T, k=0；
Step 2, most close point set U ' is calculatedk=C (Uk, V), wherein C is the operation sought a little pair, U 'kFor corresponding point set；
Step 3, registration vector is calculated；
Step 4, using registration vector, transformation point set U transformation obtains new position UK+1；
Step 5, if transformed error meets the threshold value of setting, iteration ends obtain final high-precision dot cloud registration result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910731854.XA CN110473239A (en) | 2019-08-08 | 2019-08-08 | A kind of high-precision point cloud registration method of 3 D laser scanning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910731854.XA CN110473239A (en) | 2019-08-08 | 2019-08-08 | A kind of high-precision point cloud registration method of 3 D laser scanning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110473239A true CN110473239A (en) | 2019-11-19 |
Family
ID=68510527
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910731854.XA Withdrawn CN110473239A (en) | 2019-08-08 | 2019-08-08 | A kind of high-precision point cloud registration method of 3 D laser scanning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110473239A (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111080684A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨工程大学 | Point cloud registration method for point neighborhood scale difference description |
| CN111710040A (en) * | 2020-06-03 | 2020-09-25 | 纵目科技（上海）股份有限公司 | Method, system, terminal and storage medium for constructing high-precision map |
| CN113223062A (en) * | 2021-06-04 | 2021-08-06 | 武汉工控仪器仪表有限公司 | Point cloud registration method based on angular point feature point selection and quick descriptor |
| CN113393507A (en) * | 2021-06-30 | 2021-09-14 | 南京工业大学 | Unmanned aerial vehicle point cloud and ground three-dimensional laser scanner point cloud registration method |
| CN113393504A (en) * | 2021-05-28 | 2021-09-14 | 广东电网有限责任公司广州供电局 | Switch cabinet point cloud registration method and system, computer equipment and storage medium |
| CN113592926A (en) * | 2021-07-23 | 2021-11-02 | 浙江工业大学 | Rapid selection and registration method for 3D laser radar point cloud data prediction points |
| CN113658166A (en) * | 2021-08-24 | 2021-11-16 | 凌云光技术股份有限公司 | Point cloud defect detection method and device based on grid model |
| CN113701655A (en) * | 2021-06-30 | 2021-11-26 | 中国地质调查局水文地质环境地质调查中心 | Method for obtaining landslide high-precision three-dimensional deformation information |
| CN113963138A (en) * | 2021-10-26 | 2022-01-21 | 王彬 | Complete and accurate extraction method of three-dimensional laser point cloud characteristic point line |
| CN114037745A (en) * | 2021-11-17 | 2022-02-11 | 北京容积视觉科技有限公司 | Multi-view three-dimensional point cloud data rough registration method based on branch and bound |
| CN114353780A (en) * | 2021-12-31 | 2022-04-15 | 高德软件有限公司 | Attitude optimization method and equipment |
| CN114612534A (en) * | 2022-03-17 | 2022-06-10 | 南京航空航天大学 | Whole-machine point cloud registration method of multi-view aircraft based on spherical harmonic characteristics |
| CN114781056A (en) * | 2022-04-13 | 2022-07-22 | 南京航空航天大学 | Aircraft complete machine shape measuring method based on feature matching |
| CN115100258A (en) * | 2022-08-29 | 2022-09-23 | 杭州三坛医疗科技有限公司 | Hip joint image registration method, device, equipment and storage medium |
| CN115546266A (en) * | 2022-11-24 | 2022-12-30 | 中国铁路设计集团有限公司 | Multi-strip airborne laser point cloud registration method based on local normal correlation |
| CN115661215A (en) * | 2022-10-17 | 2023-01-31 | 北京四维远见信息技术有限公司 | Vehicle-mounted laser point cloud data registration method and device, electronic equipment and medium |
| CN117456146A (en) * | 2023-12-21 | 2024-01-26 | 绘见科技(深圳)有限公司 | Laser point cloud splicing method, device, medium and equipment |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103955939A (en) * | 2014-05-16 | 2014-07-30 | 重庆理工大学 | Boundary feature point registering method for point cloud splicing in three-dimensional scanning system |
| US20160125226A1 (en) * | 2013-09-17 | 2016-05-05 | Shenzhen Institutes Of Advanced Technology Chinese Academy Of Sciences | Method and system for automatically optimizing quality of point cloud data |
| CN106780459A (en) * | 2016-12-12 | 2017-05-31 | 华中科技大学 | A kind of three dimensional point cloud autoegistration method |
| CN108376408A (en) * | 2018-01-30 | 2018-08-07 | 清华大学深圳研究生院 | A kind of three dimensional point cloud based on curvature feature quickly weights method for registering |
| CN109816705A (en) * | 2019-01-03 | 2019-05-28 | 咸阳师范学院 | A kind of Characteristic points match method lacking skull |
| CN109949350A (en) * | 2019-03-11 | 2019-06-28 | 中国矿业大学(北京) | A kind of multidate point cloud autoegistration method based on form invariant features |
-
2019
- 2019-08-08 CN CN201910731854.XA patent/CN110473239A/en not_active Withdrawn
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160125226A1 (en) * | 2013-09-17 | 2016-05-05 | Shenzhen Institutes Of Advanced Technology Chinese Academy Of Sciences | Method and system for automatically optimizing quality of point cloud data |
| CN103955939A (en) * | 2014-05-16 | 2014-07-30 | 重庆理工大学 | Boundary feature point registering method for point cloud splicing in three-dimensional scanning system |
| CN106780459A (en) * | 2016-12-12 | 2017-05-31 | 华中科技大学 | A kind of three dimensional point cloud autoegistration method |
| CN108376408A (en) * | 2018-01-30 | 2018-08-07 | 清华大学深圳研究生院 | A kind of three dimensional point cloud based on curvature feature quickly weights method for registering |
| CN109816705A (en) * | 2019-01-03 | 2019-05-28 | 咸阳师范学院 | A kind of Characteristic points match method lacking skull |
| CN109949350A (en) * | 2019-03-11 | 2019-06-28 | 中国矿业大学(北京) | A kind of multidate point cloud autoegistration method based on form invariant features |
Cited By (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111080684A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨工程大学 | Point cloud registration method for point neighborhood scale difference description |
| CN111080684B (en) * | 2019-12-12 | 2023-11-17 | 哈尔滨工程大学 | Point cloud registration method for point neighborhood scale difference description |
| CN111710040A (en) * | 2020-06-03 | 2020-09-25 | 纵目科技（上海）股份有限公司 | Method, system, terminal and storage medium for constructing high-precision map |
| CN111710040B (en) * | 2020-06-03 | 2024-04-09 | 纵目科技（上海）股份有限公司 | High-precision map construction method, system, terminal and storage medium |
| CN113393504A (en) * | 2021-05-28 | 2021-09-14 | 广东电网有限责任公司广州供电局 | Switch cabinet point cloud registration method and system, computer equipment and storage medium |
| CN113223062A (en) * | 2021-06-04 | 2021-08-06 | 武汉工控仪器仪表有限公司 | Point cloud registration method based on angular point feature point selection and quick descriptor |
| CN113223062B (en) * | 2021-06-04 | 2024-05-07 | 武汉工控仪器仪表有限公司 | Point cloud registration method based on corner feature point selection and quick description |
| CN113393507B (en) * | 2021-06-30 | 2023-07-28 | 南京工业大学 | Unmanned aerial vehicle point cloud and ground three-dimensional laser scanner point cloud registration method |
| CN113393507A (en) * | 2021-06-30 | 2021-09-14 | 南京工业大学 | Unmanned aerial vehicle point cloud and ground three-dimensional laser scanner point cloud registration method |
| CN113701655A (en) * | 2021-06-30 | 2021-11-26 | 中国地质调查局水文地质环境地质调查中心 | Method for obtaining landslide high-precision three-dimensional deformation information |
| CN113701655B (en) * | 2021-06-30 | 2024-03-29 | 中国地质调查局水文地质环境地质调查中心 | Method for acquiring landslide high-precision three-dimensional deformation information |
| CN113592926B (en) * | 2021-07-23 | 2024-02-02 | 浙江工业大学 | Quick selection and registration method for 3D laser radar point cloud data prediction points |
| CN113592926A (en) * | 2021-07-23 | 2021-11-02 | 浙江工业大学 | Rapid selection and registration method for 3D laser radar point cloud data prediction points |
| CN113658166B (en) * | 2021-08-24 | 2024-04-12 | 凌云光技术股份有限公司 | Point cloud defect detection method and device based on grid model |
| CN113658166A (en) * | 2021-08-24 | 2021-11-16 | 凌云光技术股份有限公司 | Point cloud defect detection method and device based on grid model |
| CN113963138A (en) * | 2021-10-26 | 2022-01-21 | 王彬 | Complete and accurate extraction method of three-dimensional laser point cloud characteristic point line |
| CN114037745A (en) * | 2021-11-17 | 2022-02-11 | 北京容积视觉科技有限公司 | Multi-view three-dimensional point cloud data rough registration method based on branch and bound |
| CN114353780B (en) * | 2021-12-31 | 2024-04-02 | 高德软件有限公司 | Gesture optimization method and device |
| CN114353780A (en) * | 2021-12-31 | 2022-04-15 | 高德软件有限公司 | Attitude optimization method and equipment |
| CN114612534A (en) * | 2022-03-17 | 2022-06-10 | 南京航空航天大学 | Whole-machine point cloud registration method of multi-view aircraft based on spherical harmonic characteristics |
| CN114781056B (en) * | 2022-04-13 | 2023-02-03 | 南京航空航天大学 | Aircraft complete machine shape measuring method based on feature matching |
| CN114781056A (en) * | 2022-04-13 | 2022-07-22 | 南京航空航天大学 | Aircraft complete machine shape measuring method based on feature matching |
| CN115100258B (en) * | 2022-08-29 | 2023-02-07 | 杭州三坛医疗科技有限公司 | Hip joint image registration method, device, equipment and storage medium |
| CN115100258A (en) * | 2022-08-29 | 2022-09-23 | 杭州三坛医疗科技有限公司 | Hip joint image registration method, device, equipment and storage medium |
| CN115661215A (en) * | 2022-10-17 | 2023-01-31 | 北京四维远见信息技术有限公司 | Vehicle-mounted laser point cloud data registration method and device, electronic equipment and medium |
| CN115546266A (en) * | 2022-11-24 | 2022-12-30 | 中国铁路设计集团有限公司 | Multi-strip airborne laser point cloud registration method based on local normal correlation |
| CN117456146A (en) * | 2023-12-21 | 2024-01-26 | 绘见科技(深圳)有限公司 | Laser point cloud splicing method, device, medium and equipment |
| CN117456146B (en) * | 2023-12-21 | 2024-04-12 | 绘见科技(深圳)有限公司 | Laser point cloud splicing method, device, medium and equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110473239A (en) | A kind of high-precision point cloud registration method of 3 D laser scanning | |
| CN106023298B (en) | Point cloud Rigid Registration method based on local Poisson curve reestablishing | |
| CN106780386B (en) | A kind of 3 D laser scanning deformation extraction method for evaluating reliability | |
| CN103295239B (en) | A kind of autoegistration method of the laser point cloud data based on datum plane image | |
| CN108171780A (en) | A kind of method that indoor true three-dimension map is built based on laser radar | |
| CN107886529A (en) | A kind of point cloud registration method for three-dimensional reconstruction | |
| CN109887015A (en) | A kind of point cloud autoegistration method based on local surface feature histogram | |
| CN105654483B (en) | The full-automatic method for registering of three-dimensional point cloud | |
| CN103955939A (en) | Boundary feature point registering method for point cloud splicing in three-dimensional scanning system | |
| CN109767463A (en) | A kind of three-dimensional point cloud autoegistration method | |
| CN113327275B (en) | Point cloud double-view-angle fine registration method based on multi-constraint point to local curved surface projection | |
| CN112396641B (en) | Point cloud global registration method based on congruent two-baseline matching | |
| CN110222642A (en) | A kind of planar architectural component point cloud contour extraction method based on global figure cluster | |
| CN104484508A (en) | Optimizing method for noncontact three-dimensional matching detection of complex curved-surface part | |
| CN113516695B (en) | Point cloud registration strategy in laser profiler flatness measurement | |
| CN110136178B (en) | Three-dimensional laser point cloud registration method and device based on endpoint fitting | |
| CN109871818A (en) | Face identification method based on normal vector distribution histogram and covariance description | |
| CN109523582A (en) | A kind of point cloud rough registration method for taking normal vector and multiple dimensioned sparse features into account | |
| CN106338277A (en) | Baseline-based building change detection method | |
| CN114648445B (en) | Multi-view high-resolution point cloud splicing method based on feature point extraction and fine registration optimization | |
| CN112884886A (en) | Three-dimensional point cloud pipeline extraction and modeling method capable of self-adapting to search radius | |
| CN110942077B (en) | Feature line extraction method based on weight local change degree and L1 median optimization | |
| CN110310322A (en) | Method for detecting assembly surface of 10-micron-level high-precision device | |
| CN111899291B (en) | Automatic registration method of urban point cloud from coarse to fine based on multisource dimension decomposition | |
| CN113267122A (en) | Industrial part size measurement method based on 3D vision sensor |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20191119 |