Background technology
The manufacturing approach of real object simulated computer image has important use in actual production, its key is the precision of multi-view angle three-dimensional point cloud data registration.Three dimensional optical measuring equipment once can only be seen a side of object, obtain the three-dimensional data of complete object, need measure to cover whole body surface object from a plurality of angles.Because the coordinate system of three-dimensional data is different under the resulting all angles, thereby, need each view be carried out registration, make it unified under the coordinate system of an overall situation.Obviously, registration accuracy has directly determined the measuring accuracy of total system.
Traditional method for registering has subsides positioning label and the high-precision rotary table of use on testee.It is invisible that fixedly ball and positioning label can make the part object, generally is attached to more smooth surf zone relatively.The flow-type optical scanner is to label special circularly on the object or on the instrument of fixed object, and its effect is equivalent to fixedly ball.According to 2 view in front and back more than 3 or 3 not the common tag of conllinear come data are carried out registration.Utilize rotary table directly to carry out registration, but the data point of object bottom and bottom can't be gathered to measurement data.More than these traditional registration technique methods simpler, but can't realize accurate location, and automaticity is lower.
In the autoregistration of three dimensional point cloud; Most widely used ICP (the Iterative Closest Point) algorithm that in document A method for registration of 3-D shape, proposes for Besl etc., it is put the nearest point of model point set and sets up the corresponding relation between point set through constantly searching in the data point set each.But ICP requires the subclass that data set is another data set, and initial position is had relatively high expectations, and each iteration all will be calculated the corresponding point that impact point concentrates each point to concentrate in RP, and calculated amount is big, and efficient is low.Chen etc. propose the utilization direction of normal in document Object modeling by registration of multiple range images distance replaces point-to-point apart from the evaluation function as coupling; Point in such two views need not corresponding one by one; But this method requires lsqnonlin, and counting yield is low.Propositions such as Masuda are carried out stochastic sampling to point set, and as measurement criterion, but this method all need sampled after the iteration each time again with minimum intermediate value square error.Soon-Yang etc. have proposed a kind of CCP (Contrative Projection Point) algorithm, but still can not solve the complicated orientation problem of looking more.Above-mentioned algorithm mostly only is suitable for having the data point set of obvious corresponding relation and the registration between the model, and initial position is all had higher requirements.Therefore, how addressing the above problem is research contents of the present invention.
Summary of the invention
The object of the present invention is to provide a kind of manufacturing approach of real object simulated computer image, this method not only helps the user carries out efficient accurate three-dimensional modelling to testee and describes, and registration accuracy is high.
Its solution is: this new invention has proposed a kind of based on curvature; And the character of comprehensive use resultant curvature; Seek the corresponding data point to carrying out registration with Hausdorff distance between the point set of visual angle and method arrow invariant, realize the manufacturing approach of real object simulated computer image.It is characterized in that being undertaken: calculate each curvature variation of data points by following three steps; The searching that corresponding data point is right; Computational transformation matrix R, T.
The invention has the advantages that:
(1) the present invention's geometric properties of utilizing data point self to be had carries out the registration of data point, realizes data point robotization registration, and registration accuracy is high;
(2) the present invention also need not relation of inclusion for treating that the co-registration data points cloud does not have status requirement, only needs abundant overlapping region;
(3) the present invention can realize allocative effect preferably for the bigger surface points cloud of curved transition under the little situation in overlapping region;
(4) the present invention need not label, and telltale marks such as place kick need not high-precision rotary table.
Embodiment
With reference to figure 1, Fig. 2, Fig. 3 and Fig. 4, a kind of manufacturing approach of real object simulated computer image is characterized in that carrying out according to the following steps:
1. calculate each curvature variation of data points
1.1 on the plural at least vision orientation of model material object, confirm the three-dimensional coordinate of corresponding orientation; On each three-dimensional coordinate, determine the coordinate parameters of each data point of model surface respectively, form corresponding visual angle point set respectively by the data point of model surface on each orientation, visual angle;
1.2 for the point set under each different visual angles of above-mentioned acquisition, the coordinate parameters that selected K neighbour's point around point is concentrated each data point, and neighbour respectively ordered;
1.3 utilize K neighbour to put the curvature of calculating each data point
2. the right searching of corresponding data point
2.1 in an above-mentioned visual angle point set A, choose a data points p
i
Satisfy the Euclidean distance DL (p that method is vowed invariant 2.2 in the point set B of visual angle, choose m
i, q
J)<the data point of ε if can not find any satisfied data point, is returned step 2.1;
2.3 calculate the data point p among the above-mentioned visual angle point set A respectively
iWith the Hausdorff distance of the curvature of m data points among the point set B of visual angle, data point p
iWith each data point q in the m data points
jHausdorff distance calculation process following:
(1) at above-mentioned data point p
iAround selected K neighbour's point, formation neighbour point set N bhd (p
i), at above-mentioned data point q
jAround selected K neighbour's point, formation neighbour point set N bhd (q
j);
(2) adopt formula || p-q||=|K
p-K
q| calculate above-mentioned neighbour's point set N bhd (p respectively
i) in each neighbour put p
IkTo neighbour's point set N bhd (q
j) in all neighbours point between distance, wherein K is that two neighbours put the Gaussian curvature that concentrated neighbour order, selection neighbour point set N bhd (q
j) the middle distance neighbour puts p
IkNearest neighbour puts q
Jn, the neighbour puts p
IkPut q with the neighbour
JnCorresponding, constitute a pair of neighbour's point to (p
Ik, q
Jn);
(3) getting above-mentioned all neighbours point, the maximum a pair of neighbour of middle distance is put right distance value is the data point p among the point set A of visual angle
iWith data point q among the point set B of visual angle
jThe unidirectional Hausdorff distance h (p of curvature
i, q
j);
(4) in like manner can get data point q among the point set B of visual angle
jWith the data point p among the point set A of visual angle
iThe unidirectional Hausdorff distance h (q of curvature
j, p
i);
(5) according to above-mentioned h (p
i, q
j) and h (q
j, p
i), selecting maximal value is data point p among the point set A of visual angle
iWith data point q among the point set B of visual angle
jThe Hausdorff distance H (p of curvature
i, q
j);
2.4 according to m the Hausdorff distance that aforementioned calculation obtains, get minimum value and be designated as min, if min satisfies min<ξ, then corresponding data point is to (p
i, q
j) be that a pair of corresponding data point of visual angle point set A and visual angle point set B is to (b
t, b
t'), otherwise, return step 2.1;
2.5 in above-mentioned visual angle point set A, choose next data point p
I+1, to 2.4, calculate p according to step 2.2
I+1With corresponding data point q among the point set B of visual angle
J+1, another of formation visual angle point set A and visual angle point set B is to corresponding data point (b
T+1, b
T+1'), each that so calculates visual angle point set A and visual angle point set B one by one is to the corresponding data point;
The process flow diagram of step 2 is as shown in Figure 1;
3. computational transformation matrix R, T
3.1 the corresponding data point according to above-mentioned acquisition is obtained the coordinate transform on the corresponding three-dimensional coordinate on two different visions orientation to adopting the four-tuple method to find the solution, and obtains rotation matrix R and translation matrix T;
3.2 rotation matrix R and translation matrix T according to above-mentioned acquisition; Data point among data point among the point set A of visual angle and the visual angle point set B is unified under the coordinate system at reference data point set place; Accomplish the registered placement of data point under two visual angles, thereby form analog image on computers.
Practical implementation example of the present invention is following:
Fig. 2 and Fig. 3 are a kind of two visual angles of material object, generate two visual angle point sets.With the data entry program of two visual angle point sets, system just can carry out curvature automatically and calculate the searching of corresponding data point; Operations such as coordinate conversion; Good three-dimensional data and rotation matrix R and the translation matrix T of final output registration, thus analog image formed on computers, and its image is as shown in Figure 4; Registration error: 0.0812mm, registration accuracy is high.