General 3D surface matching process based on point-to-point model
Technical field
The present invention relates to technical field of mapping, more specifically, relate to a kind of based on the general least square 3D of point-to-point model surface matching process.
Background technology
3D surface coupling (generally also claiming surface registration) is that the forward position in computer vision, digital photogrammetry field studies a question; Its fundamental purpose is in order to establish coordinate system transformational relation between two three-dimensional surfaces; Thereby the formed cloud data point set of geometric jacquard patterning unit surface in mapping to object carries out registration, is to realize the seamless spliced key step of some cloud curved surface.This transformational relation generally adopts all or part of factor in 7 conformality transformation parameters such as yardstick, rotation and translation to describe.The key factor of implementing registration has 2 points: 1) select suitable mathematical model to describe geometrical-restriction relation between the 3D surface, and then can accurately estimate geometric parameter between the surface; 2) establish between the 3D surface same place to search rule, and then can utilize computing machine to carry out integral body and resolve.
Mainly there are 3 kinds of main flow 3D surface matching algorithms in this field at present: 1) iterative closest point (Iterative Closest Point, ICP) registration Algorithm (Besl and McKay, 1992; Chenand Medioni, 1992; Zhang, 1994), this algorithm is through nearest true in as same place on the space between two 3D point sets of search, and is 6 conformality transformation parameters between geometrical constraint condition estimation point set with Euclidean distance minimum between true same place; 2) minimum normal direction is apart from (Least Normal Distance, LND) registration Algorithm (Robert, 2004; Gruen and Akca, 2005) then be to serve as that foundation is carried out 7 conformality transformation parameter estimation between point set with the local surfaces normal direction apart from minimum with point; 3) (Least Z-Difference, LZD) registration Algorithm (Rosenholm, 1988) then is to find the solution 7 conformality transformation parameters between point set through the discrepancy in elevation quadratic sum that minimizes between all corresponding same places to minimum elevation distance.
In the ICP algorithm, select 6 positivity transformation parameters to carry out geometric relationship statement between the surface, lack scale factor, therefore can't carry out effective registration to different resolution 3D surface; This method adopts the linear iteration computing method, and speed of convergence is slower.
In the LND algorithm, although these class methods finally can obtain the good registration precision, existing LND registration Algorithm all adopts different adjustment mathematical models, and mathematical model differs greatly the adjustment estimation model that lack of uniform is general in form.
In the LZD algorithm, only consider elevation variance factor between corresponding same place, therefore in implementing registration process, be prone to produce the local convergence phenomenon, the registration initial value is had relatively high expectations.
In sum; Existing main flow 3D surface registration method also fails to form unified general adjustment mathematical model; Lack a kind of general registration model in the prior art; Can carry out organic unity to existing method for registering in theory, therefore need seek a kind of general 3D surface registration model, and can adapt to different same place definition rules.
Summary of the invention
To above-mentioned deficiency of the prior art; The purpose of this invention is to provide a kind of general 3D surface matching process based on point-to-point model; Can not only unify surperficial matching algorithm theoretically; And can have excellent extensibility according to the same place rule of correspondence of treating that the selection of registration data characteristics is fit to.More importantly, the adjustment Models that this method for registering is corresponding is very suitable for computer code and realizes possessing good application potential.
General 3D of the present invention surface matching process is a kind of high-quality point cloud, curved surface method for registering; Be to realize some cloud, the seamless spliced key step of curved surface; This registration Algorithm can directly be used for airborne laser analyzing spot cloud registration, territorial laser scanning point cloud registration, any 3D point set registration at random, and the registration application between the digital elevation model (DEM).
General 3D of the present invention surface matching process, its concrete technical scheme is following:
Step 1 is searched for the overlapping region between the surperficial cloud data collection of 3D fast, and overlapping region point collection is carried out mark;
Step 2 is carried out data organization to the cloud data collection of reference surface, sets up the reference surface cloud data and concentrates the topological relation between each node data;
Step 3 according to point-to-point model, selects the specific same place rule of correspondence to search for, and sets up the same place one-to-one relationship between registration surface and reference surface;
Step 4, the same place that obtains to search be to setting up error equation, and according to the error equation of setting up, set up normal equation according to the principle of least square, and resolve normal equation and obtain likelihood registration parameter valuation;
Step 5 according to the registration parameter that calculates, is treated registration surface and is carried out coordinate conversion.
Preferably, said method also comprises: before step 1, at first the surperficial cloud data collection of said 3D is carried out data filtering and handle.
Preferably, said step 2 through TIN (TIN), regular grid (Grid) or and the KD tree said cloud data collection is carried out data organization to set up said topological relation.
Preferably, the selectable said same place rule of correspondence comprises in the step 3: along the conjugation same place of surface normal direction, along in the nearest conjugation same place of Euclidean distance on the conjugation same place of surperficial vertical line direction, the space any one.
Preferably, the error equation of step 4 foundation is:
Wherein, t
x, t
y, t
zBe three translational components along coordinate axis,
ω, κ are three angle parameters that rotate around coordinate axis, and m is the change of scale factor, and V is the residual error that the local expansion of Taylor series is introduced.
Preferably, coordinate conversion comprises described in the step 5: treat on the registration surface arbitrfary point be P (z), conjugate points is Q (x ', y ', z ') on the corresponding reference surface for x, y, and conversion formula is following:
Wherein, t
x, t
y, t
zBe three translational components along coordinate axis,
ω, κ are three angle parameters that rotate around coordinate axis, and m is the change of scale factor.
General 3D surface matching process based on point-to-point model of the present invention is not only unified surperficial matching algorithm theoretically, and can be had excellent extensibility according to the same place rule of correspondence of treating that the selection of registration data characteristics is fit to.More importantly; The adjustment Models that this method for registering is corresponding; Being very suitable for computer code realizes; Can directly be used for airborne laser analyzing spot cloud registration, territorial laser scanning point cloud registration, any 3D point set registration at random, and the registration application between the digital elevation model (DEM), good application potential possessed.
Description of drawings
Figure 1A-C is the synoptic diagram of three kinds of same place rules of correspondence between the 3D surface of the embodiment of the invention;
Fig. 2 A-F is by the section effect contrast figure before and after the matching process registration of the embodiment of the invention.
Embodiment
By specifying technology contents of the present invention, structural attitude, realized purpose and effect, give explanation below in conjunction with embodiment and conjunction with figs. are detailed.
General 3D of the present invention surface matching process is a kind of high-quality point cloud, curved surface method for registering; Be to realize some cloud, the seamless spliced key step of curved surface; This registration Algorithm can directly be used for airborne laser analyzing spot cloud registration, territorial laser scanning point cloud registration, any 3D point set registration at random, and the registration application between the digital elevation model (DEM).
Be example with 3D cloud data registration below, describe of the present invention based on the concrete performing step of point-to-point model 3D surface matching process.Suppose to have two overlapped 3D surface data collection, one is reference surface, and another is for treating registration surface, then:
Step 1 if the data of handling are airborne LiDAR data, need to carry out data filtering to adjacent ribbons earlier and handles, i.e. millet cake and non-ground point discretely, and the reservation millet cake carries out the subsequent registration processing then.Search for the overlapping region between the surperficial cloud data collection of 3D fast, and overlapping region point collection is carried out mark.
Step 2 is carried out data organization to the cloud data collection of reference surface, sets up the reference surface cloud data and concentrates the topological relation between each node data.Generally can adopt spatial data structure such as TIN (TIN), regular grid (Grid) or KD tree that the cloud data collection is carried out said data organization.
Step 3 according to point-to-point model, selects the specific same place rule of correspondence to search for, and sets up the same place one-to-one relationship between registration surface and reference surface.Alternative three kinds of same place rule of correspondence synoptic diagram are as shown in Figure 1, comprising: Figure 1A is the nearest conjugation same place of Euclidean distance on the space for conjugation same place, the Figure 1B along the surface normal direction for conjugation same place, Fig. 1 C along surperficial vertical line direction.
Step 4, the least square adjustment that adopts point-to-point adjustment Models to carry out geometric transformation parameter between the surface is resolved.Specifically, the same place that obtains to search is to setting up error equation, and according to the error equation of setting up, sets up normal equation according to the principle of least square, and resolves normal equation and obtain likelihood registration parameter valuation.Said adjustment mathematical model is:
D=(x-x′)
2+(y-y′)
2+(z-z′)
2
To same place to the error equation set up as shown in the formula:
Wherein, t
x, t
y, t
zBe three translational components along coordinate axis,
ω, κ are three angle parameters that rotate around coordinate axis, and m is the change of scale factor.V is the residual error that the local expansion of Taylor series is introduced; From the statistical theory angle; Stochastic variable
its expectation value E (V)=0,
therefore above-mentioned estimation model are typical Gauss-Markov estimation models.
7 positivity transformation parameters that utilize above-mentioned steps to calculate are treated the registration surface data and are carried out coordinate conversion, carry out the purpose of registration thereby reach reference surface with treating registration surface.
Suppose to treat arbitrfary point P on the registration surface (z), conjugate points is Q (x ', y ', z ') on the corresponding reference surface for x, y, and conversion formula is following:
Wherein, t
x, t
y, t
zBe three translational components along coordinate axis,
ω, κ are three angle parameters that rotate around coordinate axis, and m is the change of scale factor.
Fig. 2 A-F is by several sections effect contrast figure before and after the matching process registration of the embodiment of the invention.The left side of every width of cloth accompanying drawing is the preceding overlay region of a band registration sectional view, and the right side is the overlay region sectional view behind the band registration.Through relatively seeing that before the registration, two band data differences of left side sectional view are bigger, to eliminate and the difference quilt is good behind the registration, strip data is in coincidence status basically.
We utilize said general 3D surface matching process based on point-to-point model to develop the corresponding software module; Adopt this software module to carry out the experiment of airborne laser scanning band adjustment; Wherein, Two groups of strip data are obtained by the airborne LiDAR system (ALS6.0) of come card company, and the phase closing precision is reported as follows:
Data set 1:
The overlay region always count into: 1,892,407
Ground point count into: 1,040,538
The equalization point spacing is: 1.13meter
Error is in the weight unit that the ICP same place rule of correspondence obtains: 0.71meter
Error is in the weight unit that the LND same place rule of correspondence obtains: 0.43meter
Error is in the weight unit that the LZD same place rule of correspondence obtains: 0.97meter
Data set 2:
The overlay region always count into: 2,034,542
Ground point count into: 1,132,670
The equalization point spacing is: 0.72meter
Error is in the weight unit that the ICP same place rule of correspondence obtains: 0.38meter
Error is in the weight unit that the LND same place rule of correspondence obtains: 0.27meter
Error is in the weight unit that the LZD same place rule of correspondence obtains: 0.42meter
Above-mentioned experimental result shows that the general 3D surface matching process based on point-to-point model of the present invention can carry out registration according to three kinds of different same place definition rules, is a kind of general registration model really.
The above is merely embodiments of the invention; Be not so limit claim of the present invention; Every equivalent structure or equivalent flow process conversion that utilizes instructions of the present invention and accompanying drawing content to be done; Or directly or indirectly be used in other relevant technical fields, all in like manner be included in the scope of patent protection of the present invention.