CN115861397A - Point cloud registration method based on improved FPFH-ICP - Google Patents
Point cloud registration method based on improved FPFH-ICP Download PDFInfo
- Publication number
- CN115861397A CN115861397A CN202211489427.3A CN202211489427A CN115861397A CN 115861397 A CN115861397 A CN 115861397A CN 202211489427 A CN202211489427 A CN 202211489427A CN 115861397 A CN115861397 A CN 115861397A
- Authority
- CN
- China
- Prior art keywords
- point
- key
- point cloud
- registration
- points
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The invention relates to a point cloud registration method based on improved FPFH-ICP, which comprises the following steps: s1, performing down-sampling on two pieces of point clouds to be registered through voxel filtering to obtain a template point cloud P and a target point cloud Q; s2, extracting key points from the point cloud after down-sampling through an ISS3D algorithm, introducing the gravity center of the point cloud to constrain the key points, and eliminating error point pairs to obtain a key point set and a key point set y (ii) a S3, using the improved FPFH algorithm to respectively construct a feature descriptor FPFH (fuzzy predictive playing function) based on the key points in the key point set and the key point set Pi And FPFH Qi (ii) a S4, preliminarily estimating the pose relationship of the two frames of point clouds by an SCA-IA algorithm to obtain a rough matching pose transformation matrix, and acquiring rough registration point clouds according to the matrix; s5, optimizing two frames of point cloud data by using an improved ICP (inductively coupled plasma) algorithm to obtain an accurate attitude transformation matrix, and S6, converting the rough registration point cloud into a coordinate system of a target point cloud by using the accurate attitude transformation matrixAnd (4) forming fine registration. The invention can obviously improve the precision and efficiency of point cloud registration.
Description
Technical Field
The invention belongs to the field of three-dimensional registration, and particularly relates to a point cloud registration method based on improved FPFH-ICP.
Background
In recent years, with rapid development of three-dimensional laser scanning technology and visual capture technology, difficulty in obtaining three-dimensional point cloud data is remarkably reduced. Compared with two-dimensional data, the three-dimensional point cloud contains richer geometric information, can describe the information of a target object more accurately, and is widely applied to the fields of three-dimensional reconstruction, reverse engineering, industrial manufacturing and the like.
However, in the actual process of acquiring point cloud data, the required complete point cloud data cannot be acquired at one time due to the influence of the acquisition angle, the acquisition distance, the object shielding and other factors, and usually, the complete surface information of the object can be acquired by multi-angle and multi-batch acquisition. Therefore, an appropriate rotation matrix and an appropriate translation matrix need to be determined so as to uniformly convert the data acquired from multiple viewing angles into the same coordinate system, namely, the registration of the point cloud data.
The essence of point cloud registration is to find an optimal transformation matrix, so that two groups of point clouds in different coordinate systems can be matched as much as possible after matrix transformation. The point cloud registration process can be generally divided into two steps of coarse registration and fine registration, wherein the coarse registration refers to that under the condition that the relative poses of the target point cloud and the reference point cloud are completely unknown, an approximate transformation array is quickly found, and the two point clouds are registered to approximately correct positions. The coarse registration method is generally realized through local feature descriptor similarity, and has strong resistance to scale change and robustness. However, there are some problems, for example, for some adjacent points, a situation that a plurality of feature points are similar to the features of one sampling point may occur, which may result in mismatching and cause a great negative effect on the subsequent solution of the transformation matrix.
The point cloud fine registration refers to registration based on coarse registration to obtain a more accurate transformation matrix, so that the two groups of point clouds are overlapped as much as possible. At present, an Iterative Closest Point (ICP) algorithm is most widely used for point cloud precise registration, and is simple and easy to implement, and the solution precision is high. But ICP is sensitive to the initial poses of the original point cloud and the target point cloud, is easy to fall into a local optimal solution, and has large calculation amount and low calculation speed.
Disclosure of Invention
In view of the above, the present invention aims to provide a point cloud registration method based on an improved FPFH-ICP, which shortens the time consumption of the ICP stage, improves the accuracy and speed of registration, and obtains a better registration result.
In order to achieve the purpose, the invention adopts the following technical scheme:
a point cloud registration method based on improved FPFH-ICP comprises the following steps:
s1, performing down-sampling on two pieces of point clouds to be registered through voxel filtering to obtain a template point cloud P and a target point cloud Q;
s2, extracting key points from the point cloud after down-sampling through an ISS3D algorithm, introducing the gravity center of the point cloud to constrain the key points, and eliminating error point pairs to obtain a key point set P key And key point set Q key ；
S3, respectively using the improved FPFH algorithm to set the key points P key And key point set Q key Key point construction feature descriptor FPFH Pi And FPFH Qi ；
S4, preliminarily estimating the pose relationship of the two frames of point clouds by an SCA-IA algorithm to obtain a rough matching pose transformation matrix, and acquiring rough registration point clouds according to the matrix;
s5, optimizing the matching effect of the two frames of point cloud data by using an improved ICP algorithm to obtain an accurate attitude transformation matrix;
and S6, converting the rough registration point cloud into a coordinate system of a target point cloud by using the accurate attitude transformation matrix to complete the fine registration.
Further, the extracting key points of the point cloud after down sampling by the ISS3D algorithm specifically comprises:
for each point P in the point cloud P a Establishing local coordinatesIs determined and a neighborhood radius r is set p ；
Calculating a keypoint P according to equation (1) a The Euclidean distance from each point in the neighborhood and the weight w are set ab ；
Calculating each key point P according to equation (2) a Covariance matrix cov (p) with all points in the neighborhood a )；
Cov (p) is obtained according to formula (2) a ) All eigenvalues of (a) a 1 ,λ a 2 ,λ a 3 In which λ is a 1 ＞λ a 2 ＞λ a 3 ；
Setting a threshold value epsilon a And ε b If it satisfies the formula (3), it is the ISS key point
Further, the introduced point cloud gravity center restrains the key points and eliminates wrong point pairs, specifically comprising:
for each selected key point P i 、Q i Calculating the center of gravity of neighborhood space of each key point to obtain C Pi And C Qi The calculation formula is shown as formula (4):
wherein a represents a key point, and k represents the number of neighborhood points;
for any key point Q in the target point cloud i Searching corresponding points from key points in the template point cloud,if P i As the corresponding point, the following condition is satisfied:
wherein epsilon is a set threshold;
if the corresponding point pair { P i ,Q i If the condition is not met, rejecting the key point pair;
through the steps, a key point set P is obtained key And key point set Q key 。
Further, the step S3 specifically includes:
step S3.1: for the key point set P key And key point set Q key Each of the key points P in i 、Q i Calculating the characteristic relationship between each key point and k neighborhoods thereof to obtain a simplified point characteristic histogram (SPFH) Pi And SPFH Qi ；
Step S3.2: the SPFH of the key points and the SPFH of the weighted neighborhood points are counted to obtain the final FPFH Pi And FPFH Qi The calculation formula is shown in formula (6):
wherein, P i Representing a key point, P j Is a neighborhood point, k denotes the number of neighborhood points, SPFH is a simplified histogram of point features, w j And the weight coefficient represents the distance between the key point and the jth neighborhood point.
Further, step S4 specifically includes:
step S4.1: from a template point cloud key point set P key Selecting n sampling points, wherein the distance between every two selected sampling points is larger than a set threshold value, and then obtaining the FPFH (field-programmable gate flash) characteristics of the selected sampling points;
step S4.2: from a set of key points Q of the target point cloud key Finding out the key points with similar FPFH characteristics with the sampling points, and storing the key points with similar characteristics in oneIn the set, corresponding point pairs are constructed with sampling points in a random extraction mode;
step S4.3: calculating a rigid body transformation matrix between the template point cloud P and the target point cloud Q through the constructed corresponding point pairs, then carrying out rough point cloud registration by using the transformation matrix, calculating registration error through a Huber penalty function, and recording the registration error as
Wherein H (e) i ) Is represented by equation (7):
wherein e is i Distance difference after matrix transformation for the ith group of corresponding points, l e Is a set threshold value;
step S4.4: repeating the steps S4.1-S4.3 until reaching the preset iteration number K, and selecting the transformation matrix M with the minimum registration error from all the transformation matrices 0 And the method is used for final coarse registration transformation to obtain a coarse registration point cloud P'.
Further, the step S5 specifically includes:
step S5.1: establishing a geometric topological relation between the discrete point clouds by using a KD tree, and searching by a nearest point pair based on a spatial index relation;
step S5.2: the point-to-surface error function is used for replacing the original ICP midpoint-to-point error function, nonlinear calculation in the ICP algorithm is simplified into linearity, and the point-to-surface error function calculation formula is shown in the formula (8):
wherein M is opt To minimize the results, p i As matching points, q i Is a target point, n i Is the normal vector of the plane where the target point is located, and M is a transformation matrix;
step S5.3: solving the transformation moment from point cloud P' to point cloud Q by SVD methodMatrix M 1 ；
Step S5.4: according to a transformation matrix M 1 Updating the pose of the point cloud P', and calculating a registration error by using the formula (4); step S5.5: repeating the steps S5.3-S5.4 until the registration error of the two point clouds is smaller than a set threshold or the iteration frequency reaches an upper limit, stopping iteration, and obtaining a final transformation matrix M 1 Will be based on the transformation matrix M 1 And converting the rough registration point cloud into a coordinate system of the target point cloud, thereby finishing the fine registration.
Compared with the prior art, the invention has the following beneficial effects:
1. in the key point selection stage, the key points are constrained by introducing the gravity centers of the point clouds, and error point pairs are eliminated, so that the error matching probability is reduced; in the FPFH characteristic description stage, compared with a distance linear function in an original FPFH calculation formula, the distance quadratic function used in the invention can further improve the weight of a point which is closer to the query point in the neighborhood, and simultaneously weaken the weight of a point which is farther away, so that the characteristic description of the query point is more reasonable;
2. in the ICP fine registration stage, the KD tree algorithm is used for optimizing the search efficiency, the speed of searching the closest point is increased, the point-to-surface error function is used for replacing the original ICP midpoint-to-point error function, the nonlinear calculation in the ICP algorithm is simplified into linear calculation, the convergence process is accelerated, the time consumption in the ICP stage is shortened to a great extent, the registration precision and speed are increased, and a better registration result is obtained.
Drawings
FIG. 1 is a flow chart of a point cloud registration algorithm designed by the present invention;
FIG. 2 is a schematic diagram of a local coordinate system constructed in the present invention;
FIG. 3 is an initial pose graph of a template point cloud and a target point cloud in the present invention;
FIG. 4 is a pose diagram of the template point cloud and the target point cloud in the present invention.
Detailed Description
The invention is further explained below with reference to the drawings and the embodiments.
Referring to fig. 1, an embodiment of the present invention provides a point cloud registration method based on an improved FPFH-ICP, including the following steps:
step 1: loading two point clouds to be registered: and loading point cloud source and point cloud target which are acquired from different perspectives and have similarity.
And 2, step: down-sampling the point cloud source and the point cloud target to obtain a template point cloud P and a target point cloud Q:
respectively creating minimum three-dimensional voxel grids according to a point cloud source and a point cloud target, wherein a calculation formula of the side length L of a cubic grid is shown as a formula (1), then taking the size of L as a dividing basis, decomposing the three-dimensional voxel grid, after the dividing work is finished, putting point cloud data into a grid corresponding to the point cloud data, deleting grids which do not contain data, taking a data point which is closest to the gravity center of the grid as reserved data in a small grid with data, thereby realizing the down-sampling processing of the point cloud data, obtaining point clouds P and Q after down-sampling through voxel filtering, and showing the pose as shown in FIG. 3, wherein the left side is a template point cloud, and the right side is a target point cloud.
Where α and s are scaling factors and g is the number of point clouds in the grid.
And step 3: and extracting key points from the point cloud P and the point cloud Q by using an ISS3D algorithm.
In this embodiment, specifically, the step 3 includes:
step 3.1: for each point P in the point cloud P a Establishing a local coordinate system and setting a neighborhood radius r p ；
Step 3.2: calculating a Key Point P according to equation (2) a Euclidean distance from each point in neighborhood and setting weight w ab ；
Step 3.3: calculating each key point P according to equation (3) a Covariance matrix cov (p) with all points in the neighborhood a )；
Step 3.4: cov (p) is obtained according to formula (3) a ) All eigenvalues of (a) a 1 ,λ a 2 ,λ a 3 In which λ is a 1 ＞λ a 2 ＞λ a 3 ；
Step 3.5: setting a threshold value epsilon a And ε b If it satisfies the formula (4), it is the ISS key point
And 4, step 4: introducing the point cloud gravity center to constrain the key points, and eliminating error point pairs to obtain a key point set P key And key point set Q key 。
In this embodiment, preferably, step S4 specifically includes:
step 4.1: for selecting each key point P i 、Q i Calculating the center of gravity of neighborhood space of each key point to obtain C Pi And C Qi The calculation formula is shown as formula (5):
wherein a represents a key point, and k represents the number of neighborhood points;
step 4.2: for any key point Q in the target point cloud i Finding corresponding points from key points in the template point cloud, if P is i To correspond to the points, the following conditions should be satisfied:
wherein epsilon is a set threshold value.
If the corresponding point pair { P i ,Q i If the condition is not met, rejecting the key point pair;
through the above steps, set point set P key 、Q key All eligible keypoints are stored.
And 5: respectively aligning the key point sets P by using an improved FPFH algorithm key And key point set Q key Key point construction feature descriptor FPFH Pi And FPFH Qi 。
Specifically, in this embodiment, step 5 includes:
step 5.1: for feature point set P key Each of which is a key point P i Determining the neighborhood range and P through the set neighborhood radius j Establishing a group of point pairs;
step 5.2: as shown in fig. 2, for each keypoint P i Establishing a local coordinate system, wherein three axes of the coordinate system are u, v and w respectively, and the calculation formula is shown as formula (7):
wherein n is i Is a key point P i Normal vector, P j Is a neighborhood point;
step 5.3: calculating each key point P under a local coordinate system i And three characteristic elements between other adjacent points, and the calculation formula is shown as formula (8):
wherein n is j Is a neighborhood point P j The number of normal vectors, alpha,
and θ is the normal deviation;
step 5.4:by computing each keypoint P only i And three characteristic elements between adjacent domains to obtain SPFH Pi ；
Step 5.5: the characteristic point P i SPFH of Pi And weighted neighborhood point P j SPFH of Pj Making statistics to obtain final FPFH Pi The calculation formula is shown as formula (9):
wherein, P i Representing a key point, P j Is a neighborhood point, k denotes the number of neighborhood points, SPFH is a simplified histogram of point features, w j Representing the distance between the key point and the jth neighborhood point as a weight coefficient;
similarly, according to the above steps, the FPFH Qi 。
Step 6: preliminarily estimating the pose relationship of two point clouds based on an SCA-IA algorithm to obtain a rough matching pose transformation matrix M 0 And a coarse registration point cloud P'.
In this embodiment, specifically, step 6 includes:
step 6.1: from a template point cloud key point set P key Selecting n sampling points, wherein in order to ensure that the sampling points can better cover the point cloud to be registered, the distance between every two selected sampling points is greater than a set threshold value, and then obtaining the FPFH (field programmable gate flash) characteristics of the selected sampling points;
the coordinate transformation relationship between two point clouds can be expressed by equation (10):
wherein (x) i ,y i ,z i ) For the key point P in the template point cloud i (x) of (C) i’ ,y i’ ,z i’ ) For key point Q in target point cloud i M is a transformation matrix.
As can be seen from equation (11), the matrix M has 12 parameters, and a pair of point clouds can provide 3 equations, so at least 4 sets of point pairs are required to solve the transformation matrix, i.e. n should be greater than or equal to 4.
Step 6.2: from the target point cloud key point set Q key Searching key points with similar FPFH characteristics to the sampling points, storing the key points with similar characteristics in a set, and constructing corresponding point pairs with the sampling points in a random extraction mode;
step 6.3: calculating a rigid transformation matrix between the template point cloud P and the target point cloud Q through the constructed corresponding point pairs, then carrying out rough point cloud registration by using the transformation matrix, calculating registration error through a Huber penalty function, and recording the registration error as
Wherein H (e) i ) Is represented by equation (11):
wherein e is i Distance difference after matrix transformation for the ith group of corresponding points, l e Is a set threshold value;
step 6.4: repeating the steps 6.1-6.3 until reaching the preset iteration number K, and selecting the transformation matrix M with the minimum registration error from all the transformation matrices 0 And the method is used for obtaining a coarse registration point cloud P' in the final coarse registration transformation.
The SAC-IA-based point cloud rough registration method has certain anti-interference performance on noise data, and meanwhile, the registration of the point cloud data in a global range can be realized, so that the point cloud registration is prevented from entering a local optimal solution.
And 7: obtaining an accurate attitude transformation matrix M using an improved ICP algorithm 1 According to a transformation matrix M 1 And converting the rough registration point cloud into a coordinate system of the target point cloud to finish the fine registration.
In this embodiment, preferably, step S7 specifically includes:
step 7.1: the KD tree is used for establishing a geometric topological relation between the discrete point clouds, the nearest point pair search based on the spatial index relation is realized, the speed of the nearest point search is improved, and the tree establishment process of the KD tree is as follows:
firstly, taking an X axis as a division dimension, sorting all points in a point set according to the X axis direction, dividing the point set into two subsets in the X axis direction by taking the point as a dividing point after finding a midpoint, then sorting the two subsets according to the Y axis and the Z axis direction respectively, and continuing to divide the subsets according to the midpoint. And dividing the directions of the X axis, the Y axis and the Z axis in a circulating manner, and finally adding all points into a tree structure to finish the tree building process.
Step 7.2: the point-to-surface error function is used for replacing the original ICP midpoint-to-point error function, nonlinear calculation in the ICP algorithm is simplified into linearity, the convergence process is accelerated, and the point-to-surface error function calculation formula is shown in formula (12):
wherein M is opt To minimize the results, p i Matching point, q i Is a target point, n i Is the normal vector of the plane where the target point is located, and M is a transformation matrix;
as can be seen from equation (12), when M is transformed, p i And then with q i And performing difference to obtain a vector, wherein the vector represents the displacement difference in the Euclidean space. This vector is then reconciled with point q i The normal vector of (a) is dot-multiplied, and the obtained result is used for evaluating the distance between the point and the tangent plane of the corresponding point. When all p are i After transformation according to the matrix M, the accumulated error of the formula (12) becomes minimum, and M is M at this time opt 。
Step 7.3: solving a transformation matrix from the point cloud P' to the point cloud Q by an SVD method:
for a transformation matrix M of displacement rotation, it is composed of a translation component T (T) x ,t y ,t z ) And a rotating moiety R (α, β, γ), which may be specifically expressed by formula (13) to formula (15):
M＝T(t x ,t y ,t z )·R(α,β,γ)(13)
wherein, t x ,t y ,t z Displacement amounts in the X-axis direction, the Y-axis direction and the Z-axis direction, respectively, rotation angles around the X-axis, the Y-axis and the Z-axis, alpha, beta and gamma, respectively, and r ij For the matrix multiplier component, it can be represented by equation (16):
as can be seen from equation (16), the rotation component is non-linear, making the ICP algorithm a non-linear optimization problem.
After rough matching, if the pose of the source point set is approximate to that of the target point set, the point-to-surface ICP algorithm can be used for approaching the nonlinear problem to the linear problem, so that the mathematical form of the algorithm is simpler, and the running speed is higher
Specifically, in this embodiment, the trigonometric function in R is equivalently substituted by using the equivalent similarity of the trigonometric function, so that sin α ≈ α and cos α ≈ 1, and the pose transformation matrix M may be represented as:
bringing formula (17) into formula (12), then M opt The unfolding can be simplified as follows:
wherein, a, x and b are simplified components, and can be specifically expressed as:
equation (12) can be written as:
converting the solved least square nonlinear model into a least square linear solving model through the transformation:
M opt ＝x opt ＝argmin x |Ax-b| 2 (21)
then SVD decomposition is performed on equation (21) to obtain:
M opt ＝x opt ＝SVD|Ax-b| 2 ＝AΣ + b (22)
step 7.4: according to a transformation matrix M 1 Updating the pose of the point cloud P', and calculating a registration error by using the formula (12);
step 7.5: repeating the steps 7.3-7.4 until the registration error of the two point clouds is smaller than a set threshold or the iteration frequency reaches an upper limit, stopping iteration to obtain a final transformation matrix M 1 And converting the rough registration point cloud into a coordinate system of the target point cloud so as to complete fine registration, wherein the registration result is shown in figure 4.
The above description is only a preferred embodiment of the present invention, and all equivalent changes and modifications made in accordance with the claims of the present invention should be covered by the present invention.
Claims (6)
1. A point cloud registration method based on improved FPFH-ICP is characterized by comprising the following steps:
s1, performing down-sampling on two pieces of point clouds to be registered through voxel filtering to obtain a template point cloud P and a target point cloud Q;
s2, extracting key points from the point cloud after down sampling through an ISS3D algorithm, introducing the gravity center of the point cloud to constrain the key points, and eliminating error point pairs to obtain a key point set P key And key point set Q key ；
S3, respectively identifying a key point set P by using an improved FPFH algorithm key And key point set Q key Key point construction feature descriptor FPFH Pi And FPFH Qi ；
S4, preliminarily estimating the pose relationship of the two point clouds by an SCA-IA algorithm to obtain a rough matching pose transformation matrix, and acquiring rough registration point clouds according to the matrix;
s5, optimizing the matching effect of the two frames of point cloud data by using an improved ICP algorithm to obtain an accurate attitude transformation matrix;
and S6, converting the rough registration point cloud into a coordinate system of a target point cloud by using the accurate attitude transformation matrix to complete the fine registration.
2. The method according to claim 1, wherein the key points are extracted from the down-sampled point cloud by ISS3D algorithm, specifically:
for each point P in the point cloud P a Establishing a local coordinate system and setting a neighborhood radius r p ；
Calculating a keypoint P according to equation (1) a The Euclidean distance from each point in the neighborhood and the weight w are set ab ；
Calculating each key point P according to equation (2) a Covariance matrix cov (p) with all points in the neighborhood a )；
Cov (p) is obtained according to formula (2) a ) All eigenvalues of (a) a 1 ,λ a 2 ,λ a 3 In which λ is a 1 ＞λ a 2 ＞λ a 3 ；
Setting a threshold value epsilon a And ε b If it satisfies the formula (3), it is the ISS key point
3. The method according to claim 1, wherein the introduced point cloud barycenter constrains key points and rejects pairs of error points, specifically:
for each selected key point P i 、Q i Calculating the center of gravity of neighborhood space of each key point to obtain C Pi And C Qi The calculation formula is shown as formula (4):
wherein a represents a key point, and k represents the number of neighborhood points;
for any key point Q in the target point cloud i Finding corresponding points from key points in the template point cloud, if P is i As the corresponding point, the following condition is satisfied:
wherein epsilon is a set threshold;
if the corresponding point pair { P i ,Q i If the condition is not met, rejecting the key point pair;
through the steps, a key point set P is obtained key And key point set Q key 。
4. The point cloud registration method based on the improved FPFH-ICP as claimed in claim 1, wherein the step S3 is specifically:
step S3.1: for the key point set P key And key point set Q key Each of the key points P in i 、Q i Calculating the characteristic relationship between each key point and k neighborhoods thereof to obtain a simplified point characteristic histogram (SPFH) Pi And SPFH Qi ；
Step S3.2: the SPFH of the key points and the SPFH of the weighted neighborhood points are counted to obtain the final FPFH Pi And FPFH Qi The calculation formula is shown in formula (6):
wherein, P i Representing a key point, P j Is a neighborhood point, k denotes the number of neighborhood points, SPFH is a simplified histogram of point features, w j And the weight coefficient represents the distance between the key point and the jth neighborhood point.
5. The method for point cloud registration based on improved FPFH-ICP as recited in claim 1, wherein the step S4 specifically comprises:
step S4.1: from a template point cloud key point set P key Selecting n sampling points, wherein the distance between every two selected sampling points is larger than a set threshold value, and then obtaining the FPFH (field-programmable gate flash) characteristics of the selected sampling points;
step S4.2: from a set of key points Q of the target point cloud key Searching key points with approximate FPFH (field-programmable gate flash) characteristics with the sampling points, storing the key points with similar characteristics in a set, and constructing corresponding point pairs with the sampling points in a random extraction mode;
step S4.3: calculating a rigid body transformation matrix between the template point cloud P and the target point cloud Q through the constructed corresponding point pairs, then carrying out rough point cloud registration by using the transformation matrix, calculating registration error through a Huber penalty function, and recording the registration error as
Wherein H (e) i ) Is calculated byThe formula is shown in formula (7):
wherein e is i Distance difference after matrix transformation for the ith group of corresponding points, l e Is a set threshold value;
step S4.4: repeating the steps S4.1-S4.3 until reaching the preset iteration number K, and selecting the transformation matrix M with the minimum registration error from all the transformation matrices 0 And the method is used for final coarse registration transformation to obtain a coarse registration point cloud P'.
6. The point cloud registration method based on the improved FPFH-ICP as claimed in claim 1, wherein the step S5 is specifically:
step S5.1: establishing a geometric topological relation between the discrete point clouds by using a KD tree, and searching by a nearest point pair based on a spatial index relation;
step S5.2: the point-to-surface error function is used for replacing the original ICP midpoint-to-point error function, nonlinear calculation in the ICP algorithm is simplified into linearity, and the point-to-surface error function calculation formula is shown in the formula (8):
wherein M is opt To minimize the results, p i As matching points, q i Is a target point, n i Is the normal vector of the plane where the target point is located, and M is a transformation matrix;
step S5.3: solving transformation matrix M from point cloud P' to point cloud Q through SVD method 1 ；
Step S5.4: according to a transformation matrix M 1 Updating the pose of the point cloud P', and calculating a registration error by using the formula (4);
step S5.5: repeating the steps S5.3-S5.4 until the registration error of the two point clouds is smaller than a set threshold or the iteration frequency reaches the upper limit, stopping the iteration,obtaining the final transformation matrix M 1 Will be based on the transformation matrix M 1 And converting the rough registration point cloud into a coordinate system of the target point cloud, thereby finishing the fine registration.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211489427.3A CN115861397A (en) | 2022-11-25 | 2022-11-25 | Point cloud registration method based on improved FPFH-ICP |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211489427.3A CN115861397A (en) | 2022-11-25 | 2022-11-25 | Point cloud registration method based on improved FPFH-ICP |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115861397A true CN115861397A (en) | 2023-03-28 |
Family
ID=85666364
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211489427.3A Pending CN115861397A (en) | 2022-11-25 | 2022-11-25 | Point cloud registration method based on improved FPFH-ICP |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115861397A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116452648A (en) * | 2023-06-15 | 2023-07-18 | 武汉科技大学 | Point cloud registration method and system based on normal vector constraint correction |
| CN116468764A (en) * | 2023-06-20 | 2023-07-21 | 南京理工大学 | Multi-view industrial point cloud high-precision registration system based on super-point space guidance |
| CN116935013A (en) * | 2023-09-14 | 2023-10-24 | 武汉工程大学 | Circuit board point cloud large-scale splicing method and system based on three-dimensional reconstruction |
| CN117437269A (en) * | 2023-12-22 | 2024-01-23 | 深圳大学 | Tree point cloud non-rigid registration method and related equipment |
-
2022
- 2022-11-25 CN CN202211489427.3A patent/CN115861397A/en active Pending
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116452648A (en) * | 2023-06-15 | 2023-07-18 | 武汉科技大学 | Point cloud registration method and system based on normal vector constraint correction |
| CN116452648B (en) * | 2023-06-15 | 2023-09-22 | 武汉科技大学 | Point cloud registration method and system based on normal vector constraint correction |
| CN116468764A (en) * | 2023-06-20 | 2023-07-21 | 南京理工大学 | Multi-view industrial point cloud high-precision registration system based on super-point space guidance |
| CN116935013A (en) * | 2023-09-14 | 2023-10-24 | 武汉工程大学 | Circuit board point cloud large-scale splicing method and system based on three-dimensional reconstruction |
| CN116935013B (en) * | 2023-09-14 | 2023-11-28 | 武汉工程大学 | Circuit board point cloud large-scale splicing method and system based on three-dimensional reconstruction |
| CN117437269A (en) * | 2023-12-22 | 2024-01-23 | 深圳大学 | Tree point cloud non-rigid registration method and related equipment |
| CN117437269B (en) * | 2023-12-22 | 2024-04-16 | 深圳大学 | Tree point cloud non-rigid registration method and related equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115861397A (en) | Point cloud registration method based on improved FPFH-ICP | |
| CN109887015B (en) | Point cloud automatic registration method based on local curved surface feature histogram | |
| CN111080684B (en) | Point cloud registration method for point neighborhood scale difference description | |
| CN107038717B (en) | A method of 3D point cloud registration error is automatically analyzed based on three-dimensional grid | |
| CN112017220B (en) | Point cloud accurate registration method based on robust constraint least square algorithm | |
| CN108376408B (en) | Three-dimensional point cloud data rapid weighting registration method based on curvature features | |
| CN114170279A (en) | Point cloud registration method based on laser scanning | |
| CN114972459B (en) | Point cloud registration method based on low-dimensional point cloud local feature descriptor | |
| CN106250881A (en) | A kind of target identification method based on three dimensional point cloud and system | |
| CN110807781B (en) | Point cloud simplifying method for retaining details and boundary characteristics | |
| CN106373118A (en) | A complex curved surface part point cloud reduction method capable of effectively keeping boundary and local features | |
| CN109523582B (en) | Point cloud coarse registration method considering normal vector and multi-scale sparse features | |
| CN113628263A (en) | Point cloud registration method based on local curvature and neighbor characteristics thereof | |
| CN111652855B (en) | Point cloud simplification method based on survival probability | |
| CN111986219A (en) | Matching method of three-dimensional point cloud and free-form surface model | |
| CN114663373A (en) | Point cloud registration method and device for detecting surface quality of part | |
| CN112861983A (en) | Image matching method, image matching device, electronic equipment and storage medium | |
| CN112396641A (en) | Point cloud global registration method based on congruent two-baseline matching | |
| CN117274339A (en) | Point cloud registration method based on improved ISS-3DSC characteristics combined with ICP | |
| CN115601574A (en) | Unmanned aerial vehicle image matching method for improving AKAZE characteristics | |
| CN114358166B (en) | Multi-target positioning method based on self-adaptive k-means clustering | |
| CN113706381A (en) | Three-dimensional point cloud data splicing method and device | |
| CN113409332A (en) | Building plane segmentation method based on three-dimensional point cloud | |
| CN109035311A (en) | A kind of curved bone fracture autoregistration and internal fixation steel plate pre-bending modeling method | |
| CN112581511A (en) | Three-dimensional reconstruction method and system based on approximate vertical scanning point cloud rapid registration |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |