CN103295239A - Laser-point cloud data automatic registration method based on plane base images - Google Patents
Laser-point cloud data automatic registration method based on plane base images Download PDFInfo
- Publication number
- CN103295239A CN103295239A CN2013102263806A CN201310226380A CN103295239A CN 103295239 A CN103295239 A CN 103295239A CN 2013102263806 A CN2013102263806 A CN 2013102263806A CN 201310226380 A CN201310226380 A CN 201310226380A CN 103295239 A CN103295239 A CN 103295239A
- Authority
- CN
- China
- Prior art keywords
- point
- confirmed
- match
- match point
- websites
- 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.)
- Granted
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a laser-point cloud data automatic registration method based on plane base images. According to the method, plane characteristics of two adjacent sites are respectively extracted, a datum plane is established based on the plane characteristics of each site, the base images are generated on the basis of the datum plane, homonymy point pairs of the base images of the two adjacent sites are determined by the matching of the base images of the two adjacent sites, and due to the fact the homonymy point pairs of the base images have three-dimensional coordinate information, spatial switching parameters of the two adjacent sites can be worked out through the homonymy point pairs. The laser-point cloud data automatic registration method does not need to be helped through scanning identification, is reliable and is favorable for reduction of spending on time, manpower and material resources, and non-professional users can obtain a high-quality three-dimensional data model.
Description
Technical field
The present invention relates to the autoegistration method of laser point cloud data, relate in particular to a kind of autoegistration method of the laser point cloud data based on the datum plane image.
Background technology
Although the ground laser measuring technique has become the technology of a comparative maturity and widespread use, the autoregistration of multistation laser point cloud data (being called for short " autoregistration ") still is a relatively more popular research direction, it helps the expense of reduction time, human and material resources, and makes unprofessional user can obtain high-quality three-dimensional data model.
About autoregistration, educational circles has proposed multiple diverse ways at present, and a typical method is to use artificial reflective sigh such as ball, cylinder, plane, as the link between the different scanning station, by the automatic identification to the scanning sign, set up the parameter transformational relation between the different survey stations.Yet sign must rationally distribute, and scanner scans with high-resolution, and under special circumstances, it is unpractical that sign is set, and this application for the autoregistration of identifying automatically based on sign has brought restriction.Some researchers at first carry out feature extraction to laser point cloud data, and the feature of extracting is mated, and calculate conversion parameter, and the conversion that this method calculates is not high with reference to reliability, and for example for plane of the same name, the conversion parameter between them is difficult to be determined.The difficult point of autoregistration is effective identification of three-dimensional same place between the different scanning website, therefore, needs the autoegistration method of the right more reliable laser point cloud data of a kind of same place that can effectively identify adjacent two websites of design.
Summary of the invention
The invention provides a kind of not by means of the autoegistration method of the laser point cloud data based on datum plane image scanning sign, more reliable.
Technical scheme provided by the invention is:
A kind of autoegistration method of the laser point cloud data based on the datum plane image may further comprise the steps:
Step 1, utilize laser scanner scans target buildings, and there is lap in laser scanner in the scanning of adjacent two websites, to obtain the target buildings respectively at the laser point cloud data of adjacent two websites;
Step 2, calculate the reference images of the laser point cloud data of adjacent two websites respectively, wherein the computation process of the reference images of the laser point cloud data of each website is:
(1) extract all plane characteristics of the laser point cloud data of a website, utilize all plane characteristics to calculate the point set of all reference fields and each reference field correspondence,
(2) all reference fields are calculated, wherein the computation process of each reference field is: utilize a two-dimentional regular grid that a reference field is divided into a plurality of grid units, each grid unit a plurality of points that point that this reference field correspondence should be arranged is concentrated then, in described a plurality of points, with in the point between this reference field and laser scanner to this reference field apart from the reflection strength information of the point of the maximum gray-scale value as this grid unit, with the three-dimensional coordinate information of this some three-dimensional coordinate information as the center of this grid unit, calculate the gray-scale value of possessive case net unit, each grid unit is as a pixel, thereby generates a reference images of this reference field correspondence;
Step 4, utilize the right three-dimensional coordinate information of same place of the reference images of adjacent two websites, calculate the space conversion parameter of adjacent two websites.
Preferably, in the autoegistration method of described laser point cloud data based on the datum plane image, in (1) of described step 2, the approximately equalised a plurality of plane characteristics of normal direction in all plane characteristics are merged into a reference field, with the point set of a plurality of plane characteristic correspondences point set as this reference field correspondence.
Preferably, in the autoegistration method of described laser point cloud data based on the datum plane image, with the method for average of a plurality of plane characteristics to the normal direction as this reference field, with the center of gravity of the point set of this reference field correspondence center of gravity as this reference field.
Preferably, in the autoegistration method of described laser point cloud data based on the datum plane image, in (2) of described step 2, the mesh spacing of the two-dimentional regular grid of any reference field correspondence draws by following process: calculate the mean distance between adjacent two points in the laser point cloud data of any website, and participate in the point that calculates to the distance of laser scanner less than first distance threshold.
Preferably, in the autoegistration method of described laser point cloud data based on the datum plane image, in the described step 3, adopt pyramid classification matching strategy that the reference images of adjacent two websites is mated, detailed process is: (1) generates the pyramid diagram picture of the reference images of adjacent two websites respectively, the pyramid diagram of the reference images of each website looks like to comprise one deck top layer images, n layer lower image and one deck bottom layer image, n 〉=0, (2) top layer images of adjacent two websites is carried out the SIFT coupling, it is right to obtain all match points to be confirmed, (3) successively on the n of adjacent two websites layer lower image, utilize constraint condition to all match points to be confirmed to calculating, right to reject wrong match point, it is right to obtain correct match point, (4) on the bottom layer image of adjacent two websites, utilize least square method to described correct match point to mating, right with the match point that is optimized, the match point of optimization is to being that the same place of reference images of adjacent two websites is right.
Preferably, in the autoegistration method of described laser point cloud data based on the datum plane image, in (3) of described step 3, described constraint condition comprises the distance restraint condition, described distance restraint condition is: calculate each match point of any match point centering to be confirmed to the distance of its reference field, and calculate distance poor that two match point branches are clipped to its reference field, and the difference of this distance is greater than a second distance threshold value, and it is right to reject this match point to be confirmed.
Preferably, in the autoegistration method of described laser point cloud data based on the datum plane image, in (3) of described step 3, described constraint condition comprises RANSAC constraint condition, and detailed process is:
1. initial setting up is established sample number k for infinitely great, and it is 0 that Sample Counter is counted t,
2. randomly draw 3 match points to be confirmed to setting up system of equations from all match point centerings to be confirmed, calculate 6 parameters of affined transformation function, wherein, the affined transformation function is:
x＝a
0+a
1x′+a
2y′
y＝b
0+b
1x′+b
2y′
3. calculate all the other all match points to be confirmed pair and its candidate matches point between distance, wherein the right computation process of each match point to be confirmed is, calculate one of them match point to be confirmed of this match point centering to be confirmed through the point after the affined transformation, calculate the distance between this point and another match point to be confirmed again, if this distance is less than the 3rd distance threshold, then this match point to be confirmed is to being interior point, otherwise be exterior point, wherein, interior point is right for the correct match point of RANSAC constraint checking, exterior point is that the match point of mistake of RANSAC constraint checking is right, and the right ratio of the match point of miscount:
It is exactly to be confirmed some match point logarithm that feature is counted,
4. calculate sample numerical value
η gets 0.99,
5. Sample Counter t adds 1,
6. repeat 2., 3., 4. and 5., when k＜t, stop circulation,
7. choose and have the maximum set of interior number of spots as some set in best, it is right then to reject the match point to be confirmed that does not belong to some set in this best.
Preferably, in the autoegistration method of described laser point cloud data based on the datum plane image, in (3) of described step 3, described constraint condition comprises the grayscale restraint condition, described grayscale restraint condition is: calculate the right gray scale similarity measure of any match point to be confirmed, when this this gray scale similarity measure was lower than gray scale similarity measure threshold value, then this match point to be confirmed of rejecting was right.
Preferably, in the autoegistration method of described laser point cloud data based on the datum plane image, in (2) of described step 3, with an order top layer images of adjacent two websites is carried out the SIFT coupling earlier, obtain that for the first time match point to be confirmed is right, wherein, a described order is to another website by a website, in reverse order the top layer images of adjacent two websites is carried out the SIFT coupling again, obtain that for the second time match point to be confirmed is right, wherein, described opposite order is to a described website by described another website, match point to be confirmed is to right with the inconsistent match point of the match point centering to be confirmed second time for the first time in rejecting, and it is right to obtain all match points to be confirmed; In (3) of described step 3, earlier with an order successively on the n layer lower image to adjacent two websites, utilize constraint condition to all match points to be confirmed to calculating, obtain that match point to be confirmed is right for the third time, wherein, a described order is to described another website by a described website, again in reverse order on the n layer lower image to adjacent two websites, utilize constraint condition to all match points to be confirmed to calculating, it is right to obtain the 4th match point to be confirmed, wherein, described opposite order is to a described website by described another website, match point to be confirmed is to right with the 4th the inconsistent match point of match point centering to be confirmed for the third time in rejecting, and it is right to obtain correct match point.
The autoegistration method of laser point cloud data of the present invention has following beneficial effect: the present invention extracts the plane characteristic of adjacent two websites respectively, and make up reference field based on the plane characteristic of each website, regeneration reference images on the basis of reference field, the coupling of the reference images by adjacent two websites, the same place of reference images of determining adjacent two websites is right, because the same place of this reference images is to having three-dimensional coordinate information, therefore, can be by same place to calculating the space conversion parameter of adjacent two websites.Based on said process, the invention provides a kind of need be by the autoegistration method of scanning sign and more reliable laser point cloud data, help the expense of reduction time, human and material resources, and make unprofessional user can obtain high-quality three-dimensional data model.
Description of drawings
Fig. 1 is the process flow diagram of method of the present invention;
Fig. 2 is the structural representation of vault type scanning synoptic diagram in the one embodiment of the invention;
Fig. 3 is that station, left side plane characteristic extracts figure in the one embodiment of the invention;
Fig. 4 is that right station plane characteristic extracts figure in the one embodiment of the invention;
Fig. 5 is the reference images at the station, a left side of calculating in the one embodiment of the invention;
Fig. 6 is the reference images at the station, the right side of calculating in the one embodiment of the invention;
Fig. 7 is the reference images matching result under constraint condition that adopts pyramid classification matching strategy in the one embodiment of the invention;
Fig. 8 is the process flow diagram of reference images coupling in the one embodiment of the invention;
Fig. 9 is the autoregistration result of the laser point cloud data of one embodiment of the invention.
Embodiment
The present invention is described in further detail below in conjunction with accompanying drawing, can implement according to this with reference to the instructions literal to make those skilled in the art.
As shown in Figure 1, the invention provides a kind of autoegistration method of the laser point cloud data based on the datum plane image, may further comprise the steps:
Step 1, utilize laser scanner scans target buildings, and there is lap in laser scanner in the scanning of adjacent two websites, to obtain the target buildings respectively at the laser point cloud data of adjacent two websites;
Step 2, calculate the reference images of the laser point cloud data of adjacent two websites respectively, wherein the computation process of the reference images of the laser point cloud data of each website is:
(1) all plane characteristics of the laser point cloud data of a website of extraction utilize all plane characteristics to calculate the point set of all reference fields and each reference field correspondence;
(2) all reference fields are calculated, wherein the computation process of each reference field is: utilize a two-dimentional regular grid that a reference field is divided into a plurality of grid units, each grid unit a plurality of points that point that this reference field correspondence should be arranged is concentrated then, in described a plurality of points, with in the point between this reference field and laser scanner to this reference field apart from the reflection strength information of the point of the maximum gray-scale value as this grid unit, with the three-dimensional coordinate information of this some three-dimensional coordinate information as the center of this grid unit, calculate the gray-scale value of possessive case net unit, each grid unit is as a pixel, thereby generates a reference images of this reference field correspondence;
Step 4, utilize the right three-dimensional coordinate information of same place of the reference images of adjacent two websites, calculate the space conversion parameter of adjacent two websites.
In the autoegistration method of described laser point cloud data based on the datum plane image, in (1) of described step 2, the approximately equalised a plurality of plane characteristics of normal direction in all plane characteristics are merged into a reference field, with the point set of a plurality of plane characteristic correspondences point set as this reference field correspondence.
In the autoegistration method of described laser point cloud data based on the datum plane image, with the method for average of a plurality of plane characteristics to the normal direction as this reference field, with the center of gravity of the point set of this reference field correspondence center of gravity as this reference field.
In the autoegistration method of described laser point cloud data based on the datum plane image, in (2) of described step 2, the mesh spacing of the two-dimentional regular grid of any reference field correspondence draws by following process: calculate the mean distance between adjacent two points in the laser point cloud data of any website, and participate in the point that calculates to the distance of laser scanner less than first distance threshold.
In the autoegistration method of described laser point cloud data based on the datum plane image, in the described step 3, adopt pyramid classification matching strategy that the reference images of adjacent two websites is mated, detailed process is: (1) generates the pyramid diagram picture of the reference images of adjacent two websites respectively, the pyramid diagram of the reference images of each website looks like to comprise one deck top layer images, n layer lower image and one deck bottom layer image, n 〉=0, (2) top layer images of adjacent two websites is carried out the SIFT coupling, it is right to obtain all match points to be confirmed, (3) successively on the n of adjacent two websites layer lower image, utilize constraint condition to all match points to be confirmed to calculating, right to reject wrong match point, it is right to obtain correct match point, (4) on the bottom layer image of adjacent two websites, utilize least square method to described correct match point to mating, right with the match point that is optimized, the match point of optimization is to being that the same place of reference images of adjacent two websites is right.
In the autoegistration method of described laser point cloud data based on the datum plane image, in (3) of described step 3, described constraint condition comprises the distance restraint condition, described distance restraint condition is: calculate each match point of any match point centering to be confirmed to the distance of its reference field, and calculate distance poor that two match point branches are clipped to its reference field, the difference of this distance is greater than a second distance threshold value, and it is right to reject this match point to be confirmed.
In the autoegistration method of described laser point cloud data based on the datum plane image, in (3) of described step 3, described constraint condition comprises RANSAC constraint condition, and detailed process is:
1. initial setting up is established sample number k for infinitely great, and it is 0 that Sample Counter is counted t,
2. randomly draw 3 match points to be confirmed to setting up system of equations from all match point centerings to be confirmed, calculate 6 parameters of affined transformation function, wherein, the affined transformation function is:
x＝a
0+a
1x′+a
2y′
y＝b
0+b
1x′+b
2y′
3. calculate all the other all match points to be confirmed pair and its candidate matches point between distance, wherein the right computation process of each match point to be confirmed is, calculate one of them match point to be confirmed of this match point centering to be confirmed through the point after the affined transformation, calculate the distance between this point and another match point to be confirmed again, if this distance is less than the 3rd distance threshold, then this match point to be confirmed is to being interior point, otherwise be exterior point, wherein, interior point is right for the correct match point of RANSAC constraint checking, exterior point is that the match point of mistake of RANSAC constraint checking is right, and the right ratio of the match point of miscount:
It is exactly to be confirmed some match point logarithm that feature is counted,
4. calculate sample numerical value
η gets 0.99,
5. Sample Counter t adds 1,
6. repeat 2., 3., 4. and 5., when k＜t, stop circulation,
7. choose and have the maximum set of interior number of spots as some set in best, it is right then to reject the match point to be confirmed that does not belong to some set in this best.
In the autoegistration method of described laser point cloud data based on the datum plane image, in (3) of described step 3, described constraint condition comprises the grayscale restraint condition, described grayscale restraint condition is: calculate the right gray scale similarity measure of any match point to be confirmed, when this this gray scale similarity measure was lower than gray scale similarity measure threshold value, then this match point to be confirmed of rejecting was right.
In the autoegistration method of described laser point cloud data based on the datum plane image, in (2) of described step 3, with an order top layer images of adjacent two websites is carried out the SIFT coupling earlier, obtain that for the first time match point to be confirmed is right, wherein, a described order is to another website by a website, in reverse order the top layer images of adjacent two websites is carried out the SIFT coupling again, obtain that for the second time match point to be confirmed is right, wherein, described opposite order is to a described website by described another website, match point to be confirmed is to right with the inconsistent match point of the match point centering to be confirmed second time for the first time in rejecting, and it is right to obtain all match points to be confirmed; In (3) of described step 3, earlier with an order successively on the n layer lower image to adjacent two websites, utilize constraint condition to all match points to be confirmed to calculating, obtain that match point to be confirmed is right for the third time, wherein, a described order is to described another website by a described website, again in reverse order on the n layer lower image to adjacent two websites, utilize constraint condition to all match points to be confirmed to calculating, it is right to obtain the 4th match point to be confirmed, wherein, described opposite order is to a described website by described another website, match point to be confirmed is to right with the 4th the inconsistent match point of match point centering to be confirmed for the third time in rejecting, and it is right to obtain correct match point.
Modern website formula territorial laser scanning instrument, generally scan (as Fig. 2) in fixing vault type mode through the difference meridional difference, the laser point cloud data of gathering has lattice characteristic, though the initial cloud data of gathering can directly be exported with image mode, but different websites lack public projecting plane, make that the image registration work between different websites is relatively more difficult, determine that therefore the reference field of reference images becomes the key of dealing with problems.
The present invention is converted to the two-dimensional grid structure with the laser point cloud data of three-dimensional, utilizes the image registration algorithm extraction same place of two dimension right, again same place is back to three dimensions to data from two-dimensional space, thereby realizes the extraction that three-dimensional same place is right.
Substantially based on plane, cylinder, sphere, the conical surface etc., wherein the planar element feature is much abundanter than other elemental characteristics in the design of modern architecture or structures.Therefore can between overlapping website, select the common plane feature as the projecting plane, to realize the laser point cloud data registration based on image feature.Just be based on this, by the detection to scanning scene midplane feature, setting up the reference field of laser point cloud data, generating reference images by projection again, realizing the registration of laser point cloud data with Image registration.
The present invention mainly comprises four parts: the plane characteristic of each website extracts and the reference images generation in adjacent two websites, and the coupling of the reference images between adjacent two websites, and the registration of laser point cloud data.Among the present invention adjacent two websites being called station, a left side and the right side stands.
1, plane characteristic extracts
At present, the method for plane characteristic extraction mainly contains: based on the method for curvature, based on the method for Gaussian sphere cluster, based on classifications such as fitting method.Method based on curvature mainly comprises: Gaussian curvature, mean curvature, approximate curvature, curvature variance etc., and utilize planar curvature to be approximately zero this feature extraction plane characteristic.This class methods concept is simple, and algorithm is easy to implement, but curvature calculate and generally to finish with Local Polynomial match, local principal component analysis, be subjected to neighborhood and noise effect bigger.
Method based on the Gaussian sphere cluster: at first will be to cloud data point-by-point method vector, more approaching to direction ratio according to same planar process, the normal direction angle is less between points, on Gaussian sphere, the normal direction autopolymerization together, each classification is cut apart according to spatial relation again, thereby extracted each dough sheet.Obtaining at present of normal direction information in this method mainly undertaken by methods such as Local Polynomial match, least square plane match, principal component analysis (PCA)s, calculate similar to curvature, be subjected to neighborhood size, The noise, in edge and data sudden change place, the influence that is subjected to is bigger.
Mainly contain based on fitting method: HOUGH conversion and RANSAC method.The HOUGH conversion is the method for detection of a target image parameter, often is used to survey features such as straight line, circle.In the three-dimensional point cloud target detection, be mainly used to the detection plane feature, the detection of complex object is because of features such as its parametric equation complexity, and in computation process, counting yield is lower, and the application of algorithm is restricted.RANSAC (RANdom SAMPLE CONSENSUS, the consistance of sampling immediately) is a kind of method from data centralization iteration estimation model parameter, and it makes up a basic subclass of only being made up of interior point data by grab sample.When carrying out parameter estimation, at first set a judgment criterion, utilize this criterion to reject those and the inconsistent input data of estimated model (exterior point) iteratively, use the input data that meet this model to come accurate estimation model parameter.The information of RANSAC algorithm input has: data set, model classification, threshold information.This method is approved in the industry with the simple easy to understand of its concept and distinctive robustness, is widely used in feature extraction, fields such as image coupling.Therefore adopt the RANSAC method to extract the plane characteristic of laser point cloud data.Fig. 3 has provided plane characteristic extraction figure in station, a left side among the embodiment, and Fig. 4 has provided right station plane characteristic extraction figure.
In fact, for any website, the plane characteristic leaching process through above-mentioned may obtain a more than plane characteristic, and these plane characteristics have constituted a plane set, and each plane characteristic is to there being a plane characteristic point set.
2, reference images generates
In the plane set that the RANSAC algorithm extracts, because the relation of locus, there is the normal direction approximately equal but non-conterminous plane, position, so a plurality of plane characteristics are merged into a reference field, and the point set of a plurality of plane characteristics is classified as the point set of a reference field correspondence.The method of average of calculating above-mentioned a plurality of plane characteristics to and centre of gravity place, represent reference field with these two parameters.A plane characteristic may be arranged just as the situation of a reference field herein, therefore, for a website, may calculate a more than reference field, can be to the ordering of all reference fields according to the number of the point of the point set of reference field correspondence, and according to pre-set threshold, only keep the reference field of the threshold value that the number with the point of the point set of correspondence meets.
The ground laser point cloud data that single station obtains, be laser scanner according to fixing obtaining through difference and meridional difference scanning, so its data are orderly, have lattice characteristic.Therefore adjacent interval between points, can ask for by lattice structure, in the application of reality, can be as required, rejection of data that will be far away apart from laser scanner, only calculate the equispaced between points apart from laser scanner certain limit (i.e. first distance threshold), with this mesh spacing as reference images.
In laser point cloud data, will arrive the spot projection of distance in certain given threshold range of a reference field to this reference field.
Two-dimentional regular grid delimited as mesh spacing in the interval between points that obtains above to fall into a trap.Direction with the sensing buildings outside is positive dirction, the interior point of point set that calculates a reference field correspondence arrives the symbolic distance of this reference field, the point (point between reference field and the laser scanner just) that then is positioned at the reference field outside is positive to the symbol of the distance of this reference field, and the point that is positioned at the reference field inboard is born to the symbol of the distance of this reference field.Each grid unit can be to the concentrated a plurality of points of point that a reference field correspondence should be arranged.In same grid unit inside, according to the symbolic distance ordering of point to reference field, the point that keeps the distance value maximum put corresponding reflection strength information as the half-tone information of image with this, and the three-dimensional coordinate information that will put is as the three-dimensional coordinate information at the center of this grid unit.The purpose of this design is that when actual scanning, the nearer point of range sweep instrument comes from the nearer object of range sweep instrument, and this object is to wish the structure that embodies when making up three-dimensional data model, be with such point reflection on reference images.Each grid unit is as a pixel, thereby will set up a reference images corresponding fully with reference field.This reference images correspondence a three-dimensional point array.
In the process of above-mentioned calculating gray-scale value, be the reflection strength information put as the gray-scale value of grid unit, this be because: if gray-scale value is range information, then reference images is apart from reference images, if reflection strength information, it then is the reflection strength reference images, if RGB information is the colour reference image, however not abundant apart from the texture information of reference images, the information of colour reference image is unreliable, so select reflection strength information as the half-tone information of reference images.
Each reference field has calculated a reference images, and a website just may calculate a more than reference images so.Fig. 5 is the reference images at the station, a left side of one embodiment of the invention, and Fig. 6 is the reference images at the station, the right side of one embodiment of the invention.
3, the coupling of reference images
Because the uncertainty of sweep object, sweep limit and scanner initial orientation, phenomenons such as rotation, convergent-divergent may appear in two overlapping reflection strength benchmark images, so conventional image matching method is not suitable for the coupling of this kind image, the present invention adopts for the SIFT operator that rotates, convergent-divergent has unchangeability and carries out images match (SIFT feature calculation method is not described in detail in this).
The present invention relates to following several constraint condition:
Change and all maintain the invariance though the SIFT feature has rotation, yardstick convergent-divergent, brightness, but be the local feature of image, reference images centering at adjacent two websites is not pure man-to-man relation, if do not do constraint, matching result still has the existence of erroneous matching, for fear of this situation, the present invention has carried out following constraint to matching process.The match point that draws of SIFT operator coupling is right to being called match point to be confirmed with at first utilizing, and uses following constraint condition to do further judgement.
(1) range information constraint condition
When generating the reflection strength benchmark image, also generated consistent with it three-dimensional information and range information index, between adjacent two websites at the reference field of same position on the locus or equate perhaps in the degree of depth certain displacement to be arranged.In theory, between the same place of two benchmark images should be near certain definite value apart from difference, so, same place is judged to the difference of reference field, if very big jump has appearred in difference, just reject as the mistake match point.
Each match point that calculates any match point centering to be confirmed is to the distance of its reference field, and calculates distance poor that two match point branches are clipped to its reference field, and the difference of this distance is greater than a second distance threshold value, and it is right to reject this match point to be confirmed.
(2) RANSAC constraint condition
The RANSAC method is a kind of sane method for parameter estimation, aspect the rejecting of image coupling mistake coupling a lot of application is being arranged.Its basic thought is, at first design certain objective function according to particular problem, then by extracting minimum point set repeatedly, estimate the initial value of parameter in this function, utilize these initial parameters that all data are divided into so-called " interior point " (namely satisfying the point of estimated parameter) and " exterior point " (namely not satisfying the point of estimated parameter), use at last all " interior point " to recomputate parameter with estimation function conversely.
The difference of RANSAC method and traditional optimization method is: traditional method calculates initial parameter value to whole data points as interior point earlier, puts and exterior point in recomputating then and adding up; And RANSAC to begin most be to utilize partial data to obtain initial value as interior point, seek points in the data centralization all other then.
The present invention adopts affined transformation as objective function:
x＝a
0+a
1x′+a
2y′
y＝b
0+b
1x′+b
2y′
If η is confidence level, p be match point to be confirmed to being the probability of interior point, then w=1-p represents that match point to be confirmed is to being the probability of exterior point.N represents that the needed minimum point of computational transformation parameter is to number.N is 3 in the affined transformation.1-w so
nThe probability that match point to be confirmed is centering to rare a pair of exterior point is chosen in expression, and k is maximum sampling numbers, (1-w
n)
kThe probability that it all is interior point that n point never chosen in expression, then:
(1-w
n)
k＝1-η
By following formula, can calculate:
That is to say, carry out k sampling, can obtain correct solution at confidence level η.But, in most cases, and do not know error matching points to shared ratio, proportion is automatically adjusted with error matching points for making cycle index, use RANSAC to estimate that the concrete steps of affine transformation parameter are as follows:
1. initial setting up is established sample number k for infinitely great, and it is 0 that Sample Counter is counted t.
2. from all SIFT matching characteristic points, randomly draw the match point to be confirmed of 3 SIFT matching characteristic points to setting up system of equations, calculate 6 parameters of affined transformation function.
3. calculate all the other all match points to be confirmed pair and its candidate matches point between distance, wherein the right computation process of each match point to be confirmed is, calculate one of them match point to be confirmed of this match point centering to be confirmed through the point after the affined transformation, calculate the distance between this point and another match point to be confirmed again, if distance is less than given threshold value (being the second distance threshold value), then this match point to be confirmed is to being interior point (correct match point to), otherwise is exterior point (match point of mistake to).And miscount match point Comparative Examples
It is exactly to be confirmed some match point logarithm that feature is counted,
4. calculate sample numerical value
η gets 0.99.
5. Sample Counter t adds 1.
6. repeat 2., 3., 4. and 5., when k＜t, stop circulation.
7. after iteration stops, choose and have the maximum set of interior number of spots and gather as point in best.Since last conversion coefficient only by 3 points of sample to calculating, in order to make the transformation matrix that obtains more stable, at last must with obtain all in point recomputate transformation parameter.
(3) grayscale restraint condition
Calculate the right gray scale similarity measure of any match point to be confirmed, when this this gray scale similarity measure was lower than gray scale similarity measure threshold value, then this match point to be confirmed of rejecting was right.
(4) reverse matching constraint
Even the method for using is as above missed the rejecting of mating, still the phenomenon that has the mistake coupling sometimes, if last coupling is carried out to right image from left image, the mode from right image to left image more then, oppositely mate, matching constraints is consistent with above-mentioned (1) (2) (3) the way of restraint, and match point to be confirmed to carrying out primary purification again, is reached not the purpose of coupling by mistake.
(5) least square accurately mates
Least square method was applied in the middle of the coupling since the eighties in 20th century, and professor Ackernann of Germany has proposed least square image coupling LSM.The information that this method has taken full advantage of in the imaging window is carried out compensating computation, makes the image coupling can reach the high precision coupling of 1/10 even 1/100 pixel, is called as the high precision correction of image.It not only can solve the matching problem of single-point, can also introduce rough error simultaneously easily and survey, and improves the reliability of image coupling greatly.
The matching constraint condition is in order to reject the mistake match point, in order to avoid rough error exerts an influence to registration results.But be block unique point (blob-like) because the SIFT algorithm extracts, be with each block central point as match point, make match point like this and have certain error between the match point really.The least square coupling is called as high precision image coupling, uses the least square coupling at this SIFT match point is optimized.
Based on above-mentioned constraint condition, the present invention adopts following matching strategy to carry out the coupling of benchmark image.
(1) benchmark image to two websites carries out SIFT feature extraction and coupling.
(2) the information constrained condition of applications distances is to the rejecting of match point to be confirmed to carry out once mating by mistake.
(3) use the rejecting that the RANSAC method is missed match point.Two dimension attributes of this constraint condition image application have been rejected a part of mistake match point as the foundation of judging.
(4) to match point to be confirmed to carrying out the gray scale correlation computations, be lower than the match point to be confirmed of gray scale similarity measure threshold value to rejecting for similarity measure.
(5) once oppositely mate again, reach not the purpose of coupling by mistake.
(6) correct match point is mated carrying out least square, make matching precision reach sub-pixel, be conducive to the precision of subsequent point cloud registration.
For the robustness that guarantees to mate, adopt the strategy of pyramid classification coupling.Pyramid classification coupling is a kind of matching strategy from coarse to fine, is the method for often using in the present image coupling, and the present invention also adopts this method to improve the accuracy of coupling.It carries out image down-sampled according to certain resolution, the interval of sampling is generally 3 pixels or 9 pixels, and the sampling number of plies is generally 3 layers.
The detailed process of pyramid coupling is: at first generate the pyramid diagram picture, use the SIFT coupling at the top layer pyramid, and then oppositely mate, to its further purification, for the first time the match point to be confirmed of coupling pair should be disallowable with oppositely to mate the inconsistent match point of the match point centering to be confirmed second time that obtains right.Because the top layer image is less, the compatibility of image is good, generally through not occurring the phenomenon of mistake coupling after the SIFT coupling.On following layer image, the result who uses the upper strata coupling carries out range information constraint, RANSAC constraint, gray scale related constraint and reverse matching constraint, until the bottom image.Be the n layer in lower image, and under the situation of n 〉=0, need successively mate the n layer lower image of adjacent two websites, to reject wrong match point gradually.And, when using reverse matching constraint, be earlier with by the left image order of image to the right, successively n layer lower image mated, obtain that match point to be confirmed is right for the third time, and then with by the right image order of image left, successively n layer lower image mated, it is right to obtain the 4th match point to be confirmed, then, match point to be confirmed is to right with the 4th the inconsistent match point of match point centering to be confirmed for the third time in rejecting, and it is right to obtain correct match point.In order to improve matching precision and follow-up registration accuracy, on bottom layer image, mate carrying out the least square high precision at last.Matching process is seen Fig. 8.Fig. 7 is the reference images matching result under constraint condition that adopts pyramid classification matching strategy in the one embodiment of the invention.
4, the registration of laser point cloud data
Same place by above-mentioned reflection strength images match is right, can obtain the right three-dimensional information of same place, calculates the space conversion parameter.The space similarity transformation is 1 zoom factor, 3 rotation parameters and 3 translation parameterss, because the engineer's scale of the some cloud of two websites equates that so zoom factor is 1, its space similarity transformation model is:
Above-mentioned model can be considered as the transfer problem of three-dimensional rectangular coordinate, uses traditional Eulerian angle and resolves, and can cause the instability of solution when the anglec of rotation between coordinate system is big, and the key that addresses this problem is the textural of rotation matrix.At present, mainly contain following several method: based on the resolving of four-tuple, resolving and direct structured approach based on Rodrigo's matrix.Wherein the method based on Rodrigo's matrix is converted to linear function to the rotation matrix nonlinear function, and computing velocity is fast, is suitable for registration at any angle.
In the following formula
With
Be respectively the center of gravity coordinate of two same places of same place centering.The parameter that application calculates (a, b, c) and Rodrigo's matrix computations go out angle parameter between two coordinate systems, calculate translation parameters according to first formula again.Carry out least square adjustment under the situation of excess observation and calculate having, thereby finish registration.Fig. 9 is the autoregistration result of the laser point cloud data of one embodiment of the invention.
Although embodiment of the present invention are open as above, but it is not restricted to listed utilization in instructions and the embodiment, it can be applied to various suitable the field of the invention fully, for those skilled in the art, can easily realize other modification, therefore do not deviating under the universal that claim and equivalency range limit, the present invention is not limited to specific details and illustrates here and the legend of describing.
Claims (9)
1. the autoegistration method based on the laser point cloud data of datum plane image is characterized in that, may further comprise the steps:
Step 1, utilize laser scanner scans target buildings, and there is lap in laser scanner in the scanning of adjacent two websites, to obtain the target buildings respectively at the laser point cloud data of adjacent two websites;
Step 2, calculate the reference images of the laser point cloud data of adjacent two websites respectively, wherein the computation process of the reference images of the laser point cloud data of each website is:
(1) extract all plane characteristics of the laser point cloud data of a website, utilize all plane characteristics to calculate the point set of all reference fields and each reference field correspondence,
(2) all reference fields are calculated, wherein the computation process of each reference field is: utilize a two-dimentional regular grid that a reference field is divided into a plurality of grid units, each grid unit a plurality of points that point that this reference field correspondence should be arranged is concentrated then, in described a plurality of points, with in the point between this reference field and laser scanner to this reference field apart from the reflection strength information of the point of the maximum gray-scale value as this grid unit, with the three-dimensional coordinate information of this some three-dimensional coordinate information as the center of this grid unit, calculate the gray-scale value of possessive case net unit, each grid unit is as a pixel, thereby generates a reference images of this reference field correspondence;
Step 3, the reference images of the laser point cloud data of adjacent two websites is mated, the same place of reference images that obtains adjacent two websites is right;
Step 4, utilize the right three-dimensional coordinate information of same place of the reference images of adjacent two websites, calculate the space conversion parameter of adjacent two websites.
2. the autoegistration method of the laser point cloud data based on the datum plane image as claimed in claim 1, it is characterized in that, in (1) of described step 2, the approximately equalised a plurality of plane characteristics of normal direction in all plane characteristics are merged into a reference field, with the point set of a plurality of plane characteristic correspondences point set as this reference field correspondence.
3. the autoegistration method of the laser point cloud data based on the datum plane image as claimed in claim 2, it is characterized in that, with the method for average of a plurality of plane characteristics to the normal direction as this reference field, with the center of gravity of the point set of this reference field correspondence center of gravity as this reference field.
4. the autoegistration method of the laser point cloud data based on the datum plane image as claimed in claim 2, it is characterized in that, in (2) of described step 2, the mesh spacing of the two-dimentional regular grid of any reference field correspondence draws by following process: calculate the mean distance between adjacent two points in the laser point cloud data of any website, and participate in the point that calculates to the distance of laser scanner less than first distance threshold.
5. as the autoegistration method of each described laser point cloud data based on the datum plane image in the claim 1 to 4, it is characterized in that, in the described step 3, adopt pyramid classification matching strategy that the reference images of adjacent two websites is mated, detailed process is: (1) generates the pyramid diagram picture of the reference images of adjacent two websites respectively, the pyramid diagram of the reference images of each website looks like to comprise one deck top layer images, n layer lower image and one deck bottom layer image, n 〉=0, (2) top layer images of adjacent two websites is carried out the SIFT coupling, it is right to obtain all match points to be confirmed, (3) successively on the n of adjacent two websites layer lower image, utilize constraint condition to all match points to be confirmed to calculating, right to reject wrong match point, it is right to obtain correct match point, (4) on the bottom layer image of adjacent two websites, utilize least square method to described correct match point to mating, right with the match point that is optimized, the match point of optimization is to being that the same place of reference images of adjacent two websites is right.
6. the autoegistration method of the laser point cloud data based on the datum plane image as claimed in claim 5, it is characterized in that, in (3) of described step 3, described constraint condition comprises the distance restraint condition, described distance restraint condition is: calculate each match point of any match point centering to be confirmed to the distance of its reference field, and calculate distance poor that two match point branches are clipped to its reference field, and the difference of this distance is greater than a second distance threshold value, and it is right to reject this match point to be confirmed.
7. the autoegistration method of the laser point cloud data based on the datum plane image as claimed in claim 5 is characterized in that in (3) of described step 3, described constraint condition comprises RANSAC constraint condition, and detailed process is:
1. initial setting up is established sample number k for infinitely great, and it is 0 that Sample Counter is counted t,
2. randomly draw 3 match points to be confirmed to setting up system of equations from all match point centerings to be confirmed, calculate 6 parameters of affined transformation function, wherein, the affined transformation function is:
x＝a
0+a
1x′+a
2y′
y＝b
0+b
1x′+b
2y′
3. calculate all the other all match points to be confirmed pair and its candidate matches point between distance, wherein the right computation process of each match point to be confirmed is, calculate one of them match point to be confirmed of this match point centering to be confirmed through the point after the affined transformation, calculate the distance between this point and another match point to be confirmed again, if this distance is less than the 3rd distance threshold, then this match point to be confirmed is to being interior point, otherwise be exterior point, wherein, interior point is right for the correct match point of RANSAC constraint checking, exterior point is that the match point of mistake of RANSAC constraint checking is right, and the right ratio of the match point of miscount:
It is exactly to be confirmed some match point logarithm that feature is counted,
4. calculate sample numerical value
η gets 0.99,
5. Sample Counter t adds 1,
6. repeat 2., 3., 4. and 5., when k＜t, stop circulation,
7. choose and have the maximum set of interior number of spots as some set in best, it is right then to reject the match point to be confirmed that does not belong to some set in this best.
8. the autoegistration method of the laser point cloud data based on the datum plane image as claimed in claim 5, it is characterized in that, in (3) of described step 3, described constraint condition comprises the grayscale restraint condition, described grayscale restraint condition is: calculate the right gray scale similarity measure of any match point to be confirmed, when this this gray scale similarity measure was lower than gray scale similarity measure threshold value, then this match point to be confirmed of rejecting was right.
9. the autoegistration method of the laser point cloud data based on the datum plane image as claimed in claim 5, it is characterized in that, in (2) of described step 3, with an order top layer images of adjacent two websites is carried out the SIFT coupling earlier, obtain that for the first time match point to be confirmed is right, wherein, a described order is to another website by a website, in reverse order the top layer images of adjacent two websites is carried out the SIFT coupling again, obtain that for the second time match point to be confirmed is right, wherein, described opposite order is to a described website by described another website, match point to be confirmed is to right with the inconsistent match point of the match point centering to be confirmed second time for the first time in rejecting, and it is right to obtain all match points to be confirmed; In (3) of described step 3, earlier with an order successively on the n layer lower image to adjacent two websites, utilize constraint condition to all match points to be confirmed to calculating, obtain that match point to be confirmed is right for the third time, wherein, a described order is to described another website by a described website, again in reverse order on the n layer lower image to adjacent two websites, utilize constraint condition to all match points to be confirmed to calculating, it is right to obtain the 4th match point to be confirmed, wherein, described opposite order is to a described website by described another website, match point to be confirmed is to right with the 4th the inconsistent match point of match point centering to be confirmed for the third time in rejecting, and it is right to obtain correct match point.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310226380.6A CN103295239B (en) | 2013-06-07 | 2013-06-07 | A kind of autoegistration method of the laser point cloud data based on datum plane image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310226380.6A CN103295239B (en) | 2013-06-07 | 2013-06-07 | A kind of autoegistration method of the laser point cloud data based on datum plane image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103295239A true CN103295239A (en) | 2013-09-11 |
| CN103295239B CN103295239B (en) | 2016-05-11 |
Family
ID=49096044
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310226380.6A Expired - Fee Related CN103295239B (en) | 2013-06-07 | 2013-06-07 | A kind of autoegistration method of the laser point cloud data based on datum plane image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103295239B (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103646156A (en) * | 2013-12-30 | 2014-03-19 | 北京建筑大学 | Ball target detection-based automatic registration method for laser point cloud data |
| CN103744086A (en) * | 2013-12-23 | 2014-04-23 | 北京建筑大学 | High-precision registration method for ground laser radar and close-range photography measurement data |
| CN104700393A (en) * | 2013-10-22 | 2015-06-10 | 莱卡地球系统公开股份有限公司 | Registration of multiple laser scans |
| CN104766302A (en) * | 2015-02-05 | 2015-07-08 | 武汉大势智慧科技有限公司 | Method and system for optimizing laser scanning point cloud data by means of unmanned aerial vehicle images |
| CN105469388A (en) * | 2015-11-16 | 2016-04-06 | 集美大学 | Building point cloud registration algorithm based on dimension reduction |
| CN105866764A (en) * | 2015-12-01 | 2016-08-17 | 中国科学院上海技术物理研究所 | On-satellite laser altimeter rough error elimination method integrated with multi-source data |
| CN106954012A (en) * | 2017-03-29 | 2017-07-14 | 武汉嫦娥医学抗衰机器人股份有限公司 | A kind of high definition polyphaser full-view stereo imaging system and method |
| CN108828613A (en) * | 2018-05-24 | 2018-11-16 | 北京石油化工学院 | A kind of method and harmful influence storage laser scanning device of removal noise |
| CN109214982A (en) * | 2018-09-11 | 2019-01-15 | 大连理工大学 | A kind of three-dimensional point cloud imaging method based on bicylindrical projection model |
| CN109544607A (en) * | 2018-11-24 | 2019-03-29 | 上海勘察设计研究院(集团)有限公司 | A kind of cloud data registration method based on road mark line |
| CN110084840A (en) * | 2019-04-24 | 2019-08-02 | 百度在线网络技术（北京）有限公司 | Point cloud registration method, device, server and computer-readable medium |
| CN110322492A (en) * | 2019-07-03 | 2019-10-11 | 西北工业大学 | A kind of extraterrestrial target three-dimensional point cloud method for registering based on global optimization |
| CN110443836A (en) * | 2019-06-24 | 2019-11-12 | 中国人民解放军战略支援部队信息工程大学 | A kind of point cloud data autoegistration method and device based on plane characteristic |
| CN111292364A (en) * | 2020-01-21 | 2020-06-16 | 天目爱视（北京）科技有限公司 | Method for rapidly matching images in three-dimensional model construction process |
| CN111445472A (en) * | 2020-03-26 | 2020-07-24 | 达闼科技成都有限公司 | Laser point cloud ground segmentation method and device, computing equipment and storage medium |
| CN112489099A (en) * | 2020-12-11 | 2021-03-12 | 北京航空航天大学 | Point cloud registration method and device, storage medium and electronic equipment |
| CN113793370A (en) * | 2021-01-13 | 2021-12-14 | 北京京东叁佰陆拾度电子商务有限公司 | Three-dimensional point cloud registration method and device, electronic equipment and readable medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2010093675A1 (en) * | 2009-02-13 | 2010-08-19 | Harris Corporation | Registration of 3d point cloud data to 2d electro-optical image data |
| CN102169579A (en) * | 2011-03-31 | 2011-08-31 | 西北工业大学 | Rapid and accurate registration method of dense point cloud model |
| CN102959355A (en) * | 2010-07-05 | 2013-03-06 | 株式会社拓普康 | Point cloud data processing device, point cloud data processing system, point cloud data processing method, and point cloud data processing program |
-
2013
- 2013-06-07 CN CN201310226380.6A patent/CN103295239B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2010093675A1 (en) * | 2009-02-13 | 2010-08-19 | Harris Corporation | Registration of 3d point cloud data to 2d electro-optical image data |
| CN102959355A (en) * | 2010-07-05 | 2013-03-06 | 株式会社拓普康 | Point cloud data processing device, point cloud data processing system, point cloud data processing method, and point cloud data processing program |
| CN102169579A (en) * | 2011-03-31 | 2011-08-31 | 西北工业大学 | Rapid and accurate registration method of dense point cloud model |
Cited By (33)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10198533B2 (en) | 2013-10-22 | 2019-02-05 | Hexagon Technology Center Gmbh | Registration of multiple laser scans |
| CN104700393A (en) * | 2013-10-22 | 2015-06-10 | 莱卡地球系统公开股份有限公司 | Registration of multiple laser scans |
| CN104700393B (en) * | 2013-10-22 | 2018-03-27 | 莱卡地球系统公开股份有限公司 | The registration of multiple laser scannings |
| CN103744086A (en) * | 2013-12-23 | 2014-04-23 | 北京建筑大学 | High-precision registration method for ground laser radar and close-range photography measurement data |
| CN103744086B (en) * | 2013-12-23 | 2016-03-02 | 北京建筑大学 | A kind of high registration accuracy method of ground laser radar and close-range photogrammetry data |
| CN103646156B (en) * | 2013-12-30 | 2016-08-03 | 北京建筑大学 | A kind of laser point cloud data autoegistration method based on the detection of ball target |
| CN103646156A (en) * | 2013-12-30 | 2014-03-19 | 北京建筑大学 | Ball target detection-based automatic registration method for laser point cloud data |
| CN104766302A (en) * | 2015-02-05 | 2015-07-08 | 武汉大势智慧科技有限公司 | Method and system for optimizing laser scanning point cloud data by means of unmanned aerial vehicle images |
| CN104766302B (en) * | 2015-02-05 | 2017-11-24 | 武汉大势智慧科技有限公司 | A kind of method and system using unmanned plane image optimization Point Cloud of Laser Scanner |
| CN105469388A (en) * | 2015-11-16 | 2016-04-06 | 集美大学 | Building point cloud registration algorithm based on dimension reduction |
| CN105469388B (en) * | 2015-11-16 | 2019-03-15 | 集美大学 | Building point cloud registration method based on dimensionality reduction |
| CN105866764A (en) * | 2015-12-01 | 2016-08-17 | 中国科学院上海技术物理研究所 | On-satellite laser altimeter rough error elimination method integrated with multi-source data |
| CN106954012A (en) * | 2017-03-29 | 2017-07-14 | 武汉嫦娥医学抗衰机器人股份有限公司 | A kind of high definition polyphaser full-view stereo imaging system and method |
| CN108828613A (en) * | 2018-05-24 | 2018-11-16 | 北京石油化工学院 | A kind of method and harmful influence storage laser scanning device of removal noise |
| CN108828613B (en) * | 2018-05-24 | 2022-07-08 | 北京石油化工学院 | Method for removing noise and hazardous chemical storage laser scanning device |
| CN109214982A (en) * | 2018-09-11 | 2019-01-15 | 大连理工大学 | A kind of three-dimensional point cloud imaging method based on bicylindrical projection model |
| CN109214982B (en) * | 2018-09-11 | 2022-09-20 | 大连理工大学 | Three-dimensional point cloud imaging method based on double-cylinder projection model |
| CN109544607A (en) * | 2018-11-24 | 2019-03-29 | 上海勘察设计研究院(集团)有限公司 | A kind of cloud data registration method based on road mark line |
| CN109544607B (en) * | 2018-11-24 | 2021-07-06 | 上海勘察设计研究院(集团)有限公司 | Point cloud data registration method based on road sign line |
| CN110084840A (en) * | 2019-04-24 | 2019-08-02 | 百度在线网络技术（北京）有限公司 | Point cloud registration method, device, server and computer-readable medium |
| CN110084840B (en) * | 2019-04-24 | 2022-05-13 | 阿波罗智能技术(北京)有限公司 | Point cloud registration method, device, server and computer readable medium |
| US11158071B2 (en) | 2019-04-24 | 2021-10-26 | Baidu Online Network Technology (Beijing) Co., Ltd. | Method and apparatus for point cloud registration, and computer readable medium |
| CN110443836B (en) * | 2019-06-24 | 2022-03-18 | 中国人民解放军战略支援部队信息工程大学 | Point cloud data automatic registration method and device based on plane features |
| CN110443836A (en) * | 2019-06-24 | 2019-11-12 | 中国人民解放军战略支援部队信息工程大学 | A kind of point cloud data autoegistration method and device based on plane characteristic |
| CN110322492B (en) * | 2019-07-03 | 2022-06-07 | 西北工业大学 | Space three-dimensional point cloud registration method based on global optimization |
| CN110322492A (en) * | 2019-07-03 | 2019-10-11 | 西北工业大学 | A kind of extraterrestrial target three-dimensional point cloud method for registering based on global optimization |
| CN111292364A (en) * | 2020-01-21 | 2020-06-16 | 天目爱视（北京）科技有限公司 | Method for rapidly matching images in three-dimensional model construction process |
| CN111445472A (en) * | 2020-03-26 | 2020-07-24 | 达闼科技成都有限公司 | Laser point cloud ground segmentation method and device, computing equipment and storage medium |
| CN111445472B (en) * | 2020-03-26 | 2023-05-12 | 达闼机器人股份有限公司 | Laser point cloud ground segmentation method, device, computing equipment and storage medium |
| CN112489099A (en) * | 2020-12-11 | 2021-03-12 | 北京航空航天大学 | Point cloud registration method and device, storage medium and electronic equipment |
| CN112489099B (en) * | 2020-12-11 | 2023-01-20 | 北京航空航天大学 | Point cloud registration method and device, storage medium and electronic equipment |
| CN113793370A (en) * | 2021-01-13 | 2021-12-14 | 北京京东叁佰陆拾度电子商务有限公司 | Three-dimensional point cloud registration method and device, electronic equipment and readable medium |
| CN113793370B (en) * | 2021-01-13 | 2024-04-19 | 北京京东叁佰陆拾度电子商务有限公司 | Three-dimensional point cloud registration method and device, electronic equipment and readable medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103295239B (en) | 2016-05-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103295239B (en) | A kind of autoegistration method of the laser point cloud data based on datum plane image | |
| CN111486855B (en) | Indoor two-dimensional semantic grid map construction method with object navigation points | |
| WO2021233029A1 (en) | Simultaneous localization and mapping method, device, system and storage medium | |
| CN105184801B (en) | It is a kind of based on multi-level tactful optics and SAR image high-precision method for registering | |
| CN100430690C (en) | Method for making three-dimensional measurement of objects utilizing single digital camera to freely shoot | |
| CN103136525B (en) | A kind of special-shaped Extended target high-precision locating method utilizing Generalized Hough Transform | |
| CN106485690A (en) | Cloud data based on a feature and the autoregistration fusion method of optical image | |
| CN103268729A (en) | Mobile robot cascading type map creating method based on mixed characteristics | |
| CN104134216B (en) | The laser point cloud autoegistration method described based on 16 dimensional features and system | |
| CN101398886A (en) | Rapid three-dimensional face identification method based on bi-eye passiveness stereo vision | |
| CN104536009A (en) | Laser infrared composite ground building recognition and navigation method | |
| CN106023298A (en) | Point cloud rigid registration method based on local Poisson curved surface reconstruction | |
| CN104077760A (en) | Rapid splicing system for aerial photogrammetry and implementing method thereof | |
| CN113223068A (en) | Multi-modal image registration method and system based on depth global features | |
| CN101246595A (en) | Multi-view point data splitting method of optical three-dimensional scanning system | |
| CN105550994A (en) | Satellite image based unmanned aerial vehicle image rapid and approximate splicing method | |
| CN102446356A (en) | Parallel and adaptive matching method for acquiring remote sensing images with homogeneously-distributed matched points | |
| Wei et al. | A concentric loop convolutional neural network for manual delineation-level building boundary segmentation from remote-sensing images | |
| CN106780577A (en) | A kind of matching line segments method based on group feature | |
| CN104484647A (en) | High-resolution remote sensing image cloud height detection method | |
| CN104156952A (en) | Deformation resisting image matching method | |
| Jende et al. | Low-level tie feature extraction of mobile mapping data (mls/images) and aerial imagery | |
| CN113705731A (en) | End-to-end image template matching method based on twin network | |
| Sun | Deep learning-based building extraction using aerial images and digital surface models | |
| Wang et al. | Stereo Rectification Based on Epipolar Constrained Neural Network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20160511 Termination date: 20180607 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |