CN109341530A - Object point positioning method and system in a kind of binocular stereo vision - Google Patents

Abstract

The invention discloses object point positioning method in a kind of binocular stereo vision, specifically: enabling left and right side camera first shoots same target, carries out Feature Points Matching to the left and right image that shooting obtains；Calculated separately under world coordinates again left and right side camera photocentre to match point unit direction vector；And then according to the optical center coordinate of two cameras, the direction vector of optical center to characteristic point solves the three-dimensional coordinate of object point.The present invention also provides the systems for realizing the above method.The invention avoids a large amount of squares, the operations such as evolution and modulus have speed fast, calculate stable feature, there is higher application value in three-dimensional reconstruction field.

Description

Object point positioning method and system in binocular stereo vision

Technical Field

The invention relates to the technical field of machine vision, in particular to a three-dimensional point positioning method and system in binocular stereo vision.

Background

In the triangulation step of binocular stereo vision, after feature matching and pose estimation are usually performed on pictures obtained by shooting the same scene at different positions by two cameras, the intersection point of extension lines of the two cameras is obtained by homonymous rays of the two cameras, so that the object point coordinates of the point in a world coordinate system are obtained, and the three-dimensional dense point cloud construction and other work are completed by utilizing the information.

The spatial position of the object point is generally calculated by adopting a method of a common perpendicular line midpoint method, which relates to operations such as cross product and module calculation, and has high calculation complexity and much time consumption.

Disclosure of Invention

Aiming at the defects or improvement requirements of the prior art, the invention provides a three-dimensional point positioning method in binocular stereo vision, which ensures the accuracy and simplifies the calculation complexity.

A three-dimensional point positioning method in binocular stereo vision specifically comprises the following steps:

(1) starting a left camera and a right camera to shoot the same target, and performing feature point matching on the shot left image and the shot right image to obtain a feature point coordinate p under a left camera coordinate system_LCoordinate p of characteristic point with same name under coordinate system of camera on right side_R；

(2) Respectively calculating unit direction vectors N from the optical centers of the left camera and the right camera to the matching points under world coordinates₁And N₂：

Wherein Z is_L、Z_RRepresenting coefficients unitizing the direction vectors of the left and right cameras, R_LW、R_RWRespectively representing the rotation matrix of the left and right camera coordinate systems with respect to the world coordinate system, K₁、K₂Representing the internal parameters of the left camera and the right camera respectively, and the superscript T represents transposition;

(3) calculating the three-dimensional coordinate P of the target object point:

wherein,P₁₀,P₂₀respectively represent the optical center O of the left camera_LAnd right camera optical center O_RThree-dimensional coordinates in the world coordinate system (refer to fig. 1), P₁、P₂Are respectively ray O_Lp_LAnd O_Rp_RA foot hanging from the male vertical line.

A three-dimensional point positioning system in binocular stereo vision comprises:

the double cameras comprise a left camera and a right camera and are used for shooting the same target;

a homonymous feature point matching module for matching the feature points of the left and right images to obtain the feature point coordinate p in the left camera coordinate system_LCoordinate p of characteristic point with same name under coordinate system of camera on right side_R；

A unit direction vector calculation module for calculating unit direction vectors N from the left and right camera optical centers to the matching points respectively under world coordinates₁And N₂：

Wherein Z is_L、Z_RRepresenting coefficients unitizing the direction vectors of the left and right cameras, R_LW、R_RWRespectively representing the rotation matrix of the left and right camera coordinate systems with respect to the world coordinate system, K₁、K₂Representing the internal parameters of the left camera and the right camera respectively, and the superscript T represents transposition;

a positioning module, configured to calculate a three-dimensional coordinate P of the target object point:

wherein,P₁₀,P₂₀respectively represent the optical center O of the left camera_LAnd right camera optical center O_RThree-dimensional coordinates in the world coordinate system (refer to fig. 1), P₁、P₂Are respectively ray O_Lp_LAnd O_Rp_RA foot hanging from the male vertical line.

In general, compared with the original object point positioning method, the positioning method greatly reduces the times of multiplication, addition, square and evolution calculation, and the calculation formula of the invention achieves the same calculation precision as the original calculation formula under the condition of the same camera parameters through verification, but the calculation time is reduced by at least 24.24%.

Drawings

FIG. 1 is a schematic illustration of the positioning principle of the present invention;

FIG. 2 is a schematic view of an out-of-plane straight line projection.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

The method comprises the following steps:

(1) starting a left camera and a right camera to shoot the same target, extracting key points and calculating descriptors of the left image and the right image obtained by shooting by using SIFT or ORB operators, and obtaining a feature point coordinate p under a left camera coordinate system_LCoordinate p of characteristic point with same name under coordinate system of camera on right side_R；

(2) Respectively calculating unit direction vectors N from the optical centers of the left camera and the right camera to the matching points under world coordinates₁And N₂：

Wherein Z is_L、Z_RRepresenting coefficients unitizing the direction vectors of the left and right cameras, R_LW、R_RWRespectively representing the rotation matrix of the left and right camera coordinate systems with respect to the world coordinate system, K₁、K₂Representing the internal parameters of the left camera and the right camera respectively, and the superscript T represents transposition;

(3) calculating the three-dimensional coordinate P of the target object point:

wherein,P₁₀,P₂₀respectively represent the optical center O of the left camera_LAnd right camera optical center O_RThree-dimensional coordinates in the world coordinate system (refer to fig. 1), P₁、P₂Are respectively ray O_Lp_LAnd O_Rp_RThe foot hanging from the common vertical line (see fig. 2).

Simulation experiment:

1. adopting MATLAB to carry out simulation experiment, and selecting a left camera coordinate system O in the experiment_L-X_LY_LZ_LAs world coordinate system, O_L＝(0,0,0)^T，R_LW＝I₃Is a third order identity matrix, R_RWIs expressed as follows:

2. in the first type of experiment, let f_x＝f_y＝1000，u₀＝v₀300, 3 degrees, 2 degrees, 3 degrees, and the real coordinate of the object point in the world coordinate system is (0,0,0) at the center of the sphere^TOn a spherical surface with a radius of 400, x_w∈[-100,100]，y_w∈[-100,100]，The 81 object points are uniformly taken, and the coordinate of the right camera in the world coordinate system is set as O_R＝(10,10,0)^T。

3. In the second experiment, the rotation matrix R was set_RWRotating 90 degrees along the positive direction of the y axis and the x axis, and taking the real coordinate of the object point in the world coordinate system as the circle center (1000,0,0)^TOn a circle with radius 300, x_w＝1000,And (4) taking 81 uniformly distributed object points, wherein the rest parameters are the same as those of the first type of experiment, and after the parameters are set, designing the experiment steps.

4. Parameters are first substituted into the camera projection model:

p_L＝K₁R_LW(P'-P₁₀)

p_R＝K₂R_RW(P'-P₂₀)

wherein, P' represents the real coordinates of the object point obtained in the step 3.

Obtaining true homogeneous coordinates p of projected image points_L＝(u_L,v_L,1)^TAnd p_R＝(u_R,v_R,1)^TThen select the right pixel point p_ROr adding white Gaussian noise with standard deviation of 1 and 2 to the right camera pose (shown in tables 2 and 3), and obtaining a ray O_Lp_LAnd O_Rp_RTo obtain a unit direction vector N₁And N₂Taking the coordinates (i.e. origin) of the optical center of the left camera as point P₁₀Taking the optical center O of the right camera_RAs point P₂₀The coordinate of the model can be used for respectively calculating the midpoints P of the plumb lines of different models through the formula of the invention.

TABLE 1 Algorithm complexity comparison

TABLE 2 comparison of first type experiments

TABLE 3 comparison of the second type of experiment

It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (2)

1. A three-dimensional point positioning method in binocular stereo vision is characterized by comprising the following steps:

(1) starting a left camera and a right camera to shoot the same target, and performing feature point matching on the shot left image and the shot right image to obtain a feature point coordinate p under a left camera coordinate system_LCoordinate p of characteristic point with same name under coordinate system of camera on right side_R；

(2) Respectively calculating unit direction vectors N from the optical centers of the left camera and the right camera to the matching points under world coordinates₁And N₂：

Wherein Z is_L、Z_RRepresenting coefficients unitizing the direction vectors of the left and right cameras, R_LW、R_RWRespectively representing the rotation matrix of the left and right camera coordinate systems with respect to the world coordinate system, K₁、K₂Representing the internal parameters of the left camera and the right camera respectively, and the superscript T represents transposition;

(3) calculating the three-dimensional coordinate P of the target object point:

wherein, b is N₁ ^TN₂，P₁₀,P₂₀Respectively represent the optical center O of the left camera_LAnd right camera optical center O_RThree-dimensional coordinates in the world coordinate system (refer to fig. 1), P₁、P₂Are respectively ray O_Lp_LAnd O_Rp_RA foot hanging from the male vertical line.

2. A three-dimensional point positioning system in binocular stereo vision is characterized by comprising:

the double cameras comprise a left camera and a right camera and are used for shooting the same target;

a homonymous feature point matching module for matching the feature points of the left and right images to obtainFeature point coordinates p in the left camera coordinate system_LCoordinate p of characteristic point with same name under coordinate system of camera on right side_R；

A unit direction vector calculation module for calculating unit direction vectors N from the left and right camera optical centers to the matching points respectively under world coordinates₁And N₂：

Wherein Z is_L、Z_RRepresenting coefficients unitizing the direction vectors of the left and right cameras, R_LW、R_RWRespectively representing the rotation matrix of the left and right camera coordinate systems with respect to the world coordinate system, K₁、K₂Representing the internal parameters of the left camera and the right camera respectively, and the superscript T represents transposition;

a positioning module, configured to calculate a three-dimensional coordinate P of the target object point:

wherein, b is N₁ ^TN₂，P₁₀,P₂₀Respectively represent the optical center O of the left camera_LAnd right camera optical center O_RThree-dimensional coordinates in the world coordinate system (refer to fig. 1), P₁、P₂Are respectively ray O_Lp_LAnd O_Rp_RA foot hanging from the male vertical line.