CN107330911B

CN107330911B - Pure rotation motion judgment method based on intersection constraint

Info

Publication number: CN107330911B
Application number: CN201710318964.4A
Authority: CN
Inventors: 武元新; 蔡奇; 郁文贤
Original assignee: Shanghai Jiaotong University
Current assignee: Shanghai Jiaotong University
Priority date: 2017-05-08
Filing date: 2017-05-08
Publication date: 2022-01-11
Anticipated expiration: 2037-05-08
Also published as: CN107330911A

Abstract

The invention provides a pure rotational motion judgment method based on intersection constraint, which comprises the following steps: the method comprises the following steps: extracting correct feature point pairs (x) from the dual view_i,x′_i) I is 1,2 … m; step two: defining a matrix; step three: performing singular value decomposition on the matrix; step four: defining a vector; step five: if p/m falls within a certain interval with 0.5 as the center, the pure rotation motion is determined; if p/m is close to 0 or close to 1, non-pure rotational motion is determined. The invention is suitable for various feature point structures; compared with a homography matrix method, the pure rotation judgment of the method is more precise and accurate.

Description

Pure rotation motion judgment method based on intersection constraint

Technical Field

The invention relates to a pure rotational motion judgment method, in particular to a pure rotational motion judgment method based on intersection constraint.

Background

The calculation of pose and three-dimensional reconstruction by using double views is a key problem in computer vision and is also a basic technology for forming a complex vision system, such as instant positioning and map building (SLAM). In a specific scene, the more obvious the motion between the two views is, the better the pose and the three-dimensional reconstruction effect are, and the more beneficial the normal work of the SLAM system is. Three-dimensional reconstruction is almost impossible when there is only pure or nearly pure rotational (small displacement/depth of field ratio) motion between the dual views. The pure rotation motion is accurately judged, and the method has significance for improving the working robustness of a visual system.

At present, the computer vision field usually adopts the interior point proportion of a Homography (homographic) model to realize the judgment of pure rotation motion, but the threshold value selection has larger arbitrariness and fuzziness, and the practical effect of the engineering is not good.

Disclosure of Invention

Aiming at the defects in the prior art, the invention aims to provide a pure rotational motion judgment method based on Intersection Constraint, which utilizes the Intersection Constraint (Intersection Constraint) of dual-view imaging geometry, and the theoretical value of the Intersection Constraint is zero under pure rotational motion. The constraint is made equal in probability to the number of pairs greater than zero and less than zero, taking into account the effect of random noise.

According to an aspect of the present invention, there is provided a pure rotational motion determination method based on intersection constraints, comprising:

the method comprises the following steps: extracting correct feature point pairs (x) from the dual view_i,x_i'), i is 1,2, m, wherein the number m of the characteristic point pairs is more than or equal to 4, x_iAnd x_i' normalized homogeneous coordinates of the feature point pairs, respectively;

step two: defining a matrix, taking a singular vector of the matrix corresponding to the minimum singular value, and constructing a 3 x 3 matrix from the singular vector according to the sequence of columns;

step three: performing singular value decomposition on the 3 x 3 matrix;

step four: defining a vector;

step five: the number of elements larger than zero in the counting vector is p, and if p/m falls in a certain interval with 0.5 as the center, the counting vector is judged to be pure rotational motion; if p/m is close to 0 or close to 1, non-pure rotational motion is determined.

Compared with the prior art, the invention has the following beneficial effects: the invention is suitable for various characteristic point structures, such as plane characteristic point structures; compared with a homography matrix method, the pure rotation judgment of the method is more precise and accurate.

Detailed Description

The present invention will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the invention, but are not intended to limit the invention in any way. It should be noted that variations and modifications can be made by those skilled in the art without departing from the spirit of the invention. All falling within the scope of the present invention.

The pure rotation motion judgment method based on the intersection constraint comprises the following steps:

the method comprises the following steps: extracting correct feature point pairs (x) from the dual view_i,x_i'), i is 1,2, m, wherein the number m of the characteristic point pairs is more than or equal to 4, x_iAnd x_i' normalized coordinates of the dual views, respectively;

the feature points are extracted from the dual-view image, various mainstream feature extraction algorithms such as SIFT, SURF or ORB can be adopted, and the error matching point pairs are removed by using a robust algorithm such as random sample consensus (RANSAC).

Step two: defining a matrix, as in formula (1):

wherein

Expressed as the kronecker product. Taking a singular vector Q of the matrix A corresponding to the minimum singular value, and constructing a 3 x 3 matrix Q from the singular vector Q according to the sequence of columns; the superscript T represents the transpose of a matrix or vector.

Step three: singular Value Decomposition (SVD) is performed on the 3 x 3 matrix Q, i.e. Q ═ U Σ V^T＝(u₁,u₂,u₃)Σ(v₁,v₂,v₃)^TWherein Σ is a diagonal matrix composed of singular values; u and V are orthogonal matrices obtained by singular value decomposition, U_iAnd v_i(i ═ 1,2,3) column vectors for U and V, respectively;

step four: vector M is defined, as in formula (2):

wherein, | | · |, represents the length of the vector, det (·) represents the determinant of the matrix;

step five: the number of elements larger than zero in the counting vector M is p, and if p/M falls in a certain interval with 0.5 as the center, the counting vector M is judged to be pure rotational motion; if p/m is close to 0 or close to 1, non-pure rotational motion is determined.

The length setting of the interval depends on the specific decision accuracy requirement. Taking intervals according to past experience^[0.4,0.6]Pure or near-pure rotation can be effectively determined.

The foregoing description of specific embodiments of the present invention has been presented. It is to be understood that the present invention is not limited to the specific embodiments described above, and that various changes and modifications may be made by one skilled in the art within the scope of the appended claims without departing from the spirit of the invention.

Claims

1. A pure rotation motion judgment method based on intersection constraint is characterized by comprising the following steps:

the method comprises the following steps: extracting correct feature point pairs (xi, x' i) from the double-view map, wherein i is 1,2, m, the number m of the feature point pairs is more than or equal to 4, and x is larger than or equal to 4_iAnd x_i' normalized coordinates of the dual views, respectively;

step two: defining a matrix, taking a singular vector of the matrix corresponding to the minimum singular value, and constructing a 3 x 3 matrix Q from the singular vector according to the sequence of columns;

step three: performing singular value decomposition on the 3 x 3 matrix Q;

Q＝UΣV^T＝(u₁,u₂,u₃)Σ(v₁,v₂,v₃)^Twherein Σ is a diagonal matrix composed of singular values; u and V are orthogonal matrices obtained by singular value decomposition, U_iAnd v_i(i ═ 1,2,3) column vectors for U and V, respectively;

step four: defining a vector;

step five: the number of elements larger than zero in the counting vector is p, and if p/m falls in a certain interval with 0.5 as the center, the counting vector is judged to be pure rotational motion; if p/m is close to 0 or close to 1, determining that the rotation is not pure rotation;

the second step is as follows: defining a matrix, as in formula (1):

wherein

Expressed as the kronecker product; taking a singular vector Q of the matrix A corresponding to the minimum singular value, and constructing a 3 x 3 matrix Q from the singular vector Q according to the sequence of columns; superscript T represents the transpose of a matrix or vector;

the fourth step is that: vector M is defined, as in formula (2):

where | l | · | |, represents the length of the vector, det (·) represents the determinant of the matrix.