Thanks to Guangdong Metal Manufacturing Co., Ltd. for choosing Sinowon Digital Rockwell Hardness Tester DigiRock DR3
profile measurement adopting binocular active vision with normalization object of vector orthogonality
by：Sinowon
2020-01-29
Active-vision-
Measurement plays an important role in the research of contour detection.
In the active vision system, the passive vision of both eyes is adopted to give full play to its advantages.
The laser plane was initially calibrated by two 2D targets without texture.
Then, an L target with feature points is designed to construct an orthogonal object of two vectors.
In order to accurately pairactive-
In the visual system, the feature points on the L target are constructed by two cameras and parameterized by the laser plane.
Unlike the optimization method based on the distance object, the laser plane passes through the distance-angle object.
Therefore, an optimization function is created considering the specification and angle of the vector.
However, the scale of the distance is different from the scale of the angle.
Therefore, the optimization function is enhanced by standardizing the process to balance different scales.
The comparative experiments show that the dual-eye active vision with vector orthogonal normalization has reached the reduced distance error of 25%, 22%, 13% and 4% and the reduced angle error of 23%, 20%, 14% and 4%, which indicates an accurate measurement of the reconstructed object profile.
Contour measurement is one of the most important technologies in optical detection and vision.
In recent years, the information acquisition of 3D contour has received wide attention in the fields of automobile industry, robot vision, medical diagnosis, reverse engineering and so on.
However, the camera can only generate two-dimensional images without depth information, which limits the contours of the objects we acquire and perceive in the real world.
In recent years, many progress has been reported to solve the problem of contour measurement.
According to the way of obtaining 3D information, the development of contour measurement can be divided into two main parts.
The first method is to provide 3D information by another camera, I . E. e.
Or stereo vision. Faugerass O.
Provides classic work for calculating stereo in images
Based on reconstruction.
The features of stereo vision, including polar geometry, are summarized according to the camera projection model.
The fundamental matrix of stereo vision is also estimated in the literature. Grimson W. E. L.
This paper proposes an unsupervised learning method through multiplecamera system.
The framework will Group tracks corresponding to the same activity.
Then extract the path from the view of the camera.
Thirdly, the probability model is used to detect abnormal activities.
The activity model is suitable for scene structure recognition.
The improvement is to learn from cameras that are not marked in the method and activities that are not supervised.
The framework is focused on the identification of the activity model and the scene structure.
However, only the position and direction are extracted as features of the trajectory.
This paper does not reconstruct the exact contours and dimensions of the object. Ren Z.
A 3D measurement method for small mechanical parts based on the vision of the eyes is proposed.
Complex noise effects are considered in the test.
By multi-scale decomposition of images and virtual chains, nonlinear features are accurately extracted.
The 3D curve is reconstructed by the projection of the two cameras. Ambrosch K.
Real hardware is provided
Stereo vision of time.
For the stereo matching problem, a new algorithm of absolute difference and census transform is proposed. Wang Z.
This paper introduces a calibration method for the parameters of the two-eye vision system.
The test is performed by small targets in a large field of vision. The re-
The projection error and Zhang\'s method are used to optimize the parameters.
Although the implementation of the dual vision is simple, feature matching is still a complex problem for objects without texture.
Therefore, in the field of 3D Attitude Estimation, visual navigation, etc.
The second contour measurement method is active vision, which generates 3D information through structural light.
Light patterns in active vision include points, lines, and grating. Xu G.
An overview of a model used to reconstruct 3D laser projection points on an object under test.
Combine the point laser projector on the plane target and export the projection invariant on the plane target. Rodríguez J. A. M.
A method of 3D vision using mobile camera and laser line is proposed.
An approximate network for calibration scanning system is constructed.
Grating is a common structural light in active vision. Chen X.
Construction Camera-
Projector system.
Projection of four sine grating on the object under test.
The fourth-order radial and tangent lens distortion is considered in the system calibration model.
Active vision contributes to contour measurement because the structure light is an idea marker on an object.
Therefore, active vision overcomes the matching problem of objects without texture. Chen, H.
An improved calibration method is proposed, which can improve the calibration accuracy of the camera and projector at the same time.
Previous works also covered the study of the combination of active vision and eye vision.
Due to the problem of camera occlusion in a single camera, Zhang S.
A shape measurement method based on one grating projector and two cameras is developed.
Iterative registration of recent system models and 3D data
The point algorithm is described in the method. Vilaça J. L.
A laser scanning system with two cameras is reported for contour the surface of a 3D object.
Two cameras realize laser line detection for structural light calibration.
Since the grating pattern is susceptible to illumination in the environment, the measurement based on the laser line is suitable for accurate non-
Contact measurement. Jin Z. S.
A reconstruction method for arc welding of Gas Metal is proposed (GMAW)pool.
The visual measurement system consists of two cameras and a laser generator.
The point projection is constructed from a small hole model.
The two cameras and the laser plane were calibrated respectively.
This method provides the results of 3D reconstruction of the weld surface.
However, there is no mention in reconstruction of polar geometry related to points on one camera and lines on another.
In addition, this paper only models the solutions of closed forms.
The method also does not report the optimization process based on real linear dimensions and angle dimensions.
In order to realize the contour measurement, a method of active vision of both eyes is proposed.
In traditional research, two main points of view are reported in the literature.
The first is based on 2D weight.
Projection error, generated by the image, and parameterized by the internal and external matrix of the active visual system.
Obviously, the accuracy of 3D reconstruction is not the optimization goal of this idea.
Therefore, it is not suitable for 3D contour reconstruction.
Another method uses the error of 3D reconstruction.
The 3D distance is reconstructed by using the parameters of the active visual system.
In the optimization process, the measurement accuracy of linear size is improved.
However, the measurement accuracy of the angle size is not considered in the optimization.
In order to improve the measurement accuracy of the actual linear dimension and angle dimension at the same time, we focus on the orthogonal of the vector generated from the L target.
Geometrically, two vectors are derived from feature points on the L target.
The angle between the specification of the vector and the vector is parameterized by the laser plane coordinates, the internal and external matrix of the camera.
According to the known angle and specification of the vector on the L target, the modeling angle reconstruction error and distance reconstruction error are integrated in the optimization target function.
In the study, the measurement accuracy is improved by optimizing the process.
By this method, both distance and angle are considered in the optimization object.
In addition, for accurate measurement, different scales of distance and angle are normalized.
The rest of the paper is divided into three parts: the second section describes the solution model of the active vision of the eyes with the normalized object of vector orthogonal.
Then, in section 3, a comprehensive comparative experiment is carried out to verify the method.
The fourth section summarizes the paper.
Measurement plays an important role in the research of contour detection.
In the active vision system, the passive vision of both eyes is adopted to give full play to its advantages.
The laser plane was initially calibrated by two 2D targets without texture.
Then, an L target with feature points is designed to construct an orthogonal object of two vectors.
In order to accurately pairactive-
In the visual system, the feature points on the L target are constructed by two cameras and parameterized by the laser plane.
Unlike the optimization method based on the distance object, the laser plane passes through the distance-angle object.
Therefore, an optimization function is created considering the specification and angle of the vector.
However, the scale of the distance is different from the scale of the angle.
Therefore, the optimization function is enhanced by standardizing the process to balance different scales.
The comparative experiments show that the dual-eye active vision with vector orthogonal normalization has reached the reduced distance error of 25%, 22%, 13% and 4% and the reduced angle error of 23%, 20%, 14% and 4%, which indicates an accurate measurement of the reconstructed object profile.
Contour measurement is one of the most important technologies in optical detection and vision.
In recent years, the information acquisition of 3D contour has received wide attention in the fields of automobile industry, robot vision, medical diagnosis, reverse engineering and so on.
However, the camera can only generate two-dimensional images without depth information, which limits the contours of the objects we acquire and perceive in the real world.
In recent years, many progress has been reported to solve the problem of contour measurement.
According to the way of obtaining 3D information, the development of contour measurement can be divided into two main parts.
The first method is to provide 3D information by another camera, I . E. e.
Or stereo vision. Faugerass O.
Provides classic work for calculating stereo in images
Based on reconstruction.
The features of stereo vision, including polar geometry, are summarized according to the camera projection model.
The fundamental matrix of stereo vision is also estimated in the literature. Grimson W. E. L.
This paper proposes an unsupervised learning method through multiplecamera system.
The framework will Group tracks corresponding to the same activity.
Then extract the path from the view of the camera.
Thirdly, the probability model is used to detect abnormal activities.
The activity model is suitable for scene structure recognition.
The improvement is to learn from cameras that are not marked in the method and activities that are not supervised.
The framework is focused on the identification of the activity model and the scene structure.
However, only the position and direction are extracted as features of the trajectory.
This paper does not reconstruct the exact contours and dimensions of the object. Ren Z.
A 3D measurement method for small mechanical parts based on the vision of the eyes is proposed.
Complex noise effects are considered in the test.
By multi-scale decomposition of images and virtual chains, nonlinear features are accurately extracted.
The 3D curve is reconstructed by the projection of the two cameras. Ambrosch K.
Real hardware is provided
Stereo vision of time.
For the stereo matching problem, a new algorithm of absolute difference and census transform is proposed. Wang Z.
This paper introduces a calibration method for the parameters of the two-eye vision system.
The test is performed by small targets in a large field of vision. The re-
The projection error and Zhang\'s method are used to optimize the parameters.
Although the implementation of the dual vision is simple, feature matching is still a complex problem for objects without texture.
Therefore, in the field of 3D Attitude Estimation, visual navigation, etc.
The second contour measurement method is active vision, which generates 3D information through structural light.
Light patterns in active vision include points, lines, and grating. Xu G.
An overview of a model used to reconstruct 3D laser projection points on an object under test.
Combine the point laser projector on the plane target and export the projection invariant on the plane target. Rodríguez J. A. M.
A method of 3D vision using mobile camera and laser line is proposed.
An approximate network for calibration scanning system is constructed.
Grating is a common structural light in active vision. Chen X.
Construction Camera-
Projector system.
Projection of four sine grating on the object under test.
The fourth-order radial and tangent lens distortion is considered in the system calibration model.
Active vision contributes to contour measurement because the structure light is an idea marker on an object.
Therefore, active vision overcomes the matching problem of objects without texture. Chen, H.
An improved calibration method is proposed, which can improve the calibration accuracy of the camera and projector at the same time.
Previous works also covered the study of the combination of active vision and eye vision.
Due to the problem of camera occlusion in a single camera, Zhang S.
A shape measurement method based on one grating projector and two cameras is developed.
Iterative registration of recent system models and 3D data
The point algorithm is described in the method. Vilaça J. L.
A laser scanning system with two cameras is reported for contour the surface of a 3D object.
Two cameras realize laser line detection for structural light calibration.
Since the grating pattern is susceptible to illumination in the environment, the measurement based on the laser line is suitable for accurate non-
Contact measurement. Jin Z. S.
A reconstruction method for arc welding of Gas Metal is proposed (GMAW)pool.
The visual measurement system consists of two cameras and a laser generator.
The point projection is constructed from a small hole model.
The two cameras and the laser plane were calibrated respectively.
This method provides the results of 3D reconstruction of the weld surface.
However, there is no mention in reconstruction of polar geometry related to points on one camera and lines on another.
In addition, this paper only models the solutions of closed forms.
The method also does not report the optimization process based on real linear dimensions and angle dimensions.
In order to realize the contour measurement, a method of active vision of both eyes is proposed.
In traditional research, two main points of view are reported in the literature.
The first is based on 2D weight.
Projection error, generated by the image, and parameterized by the internal and external matrix of the active visual system.
Obviously, the accuracy of 3D reconstruction is not the optimization goal of this idea.
Therefore, it is not suitable for 3D contour reconstruction.
Another method uses the error of 3D reconstruction.
The 3D distance is reconstructed by using the parameters of the active visual system.
In the optimization process, the measurement accuracy of linear size is improved.
However, the measurement accuracy of the angle size is not considered in the optimization.
In order to improve the measurement accuracy of the actual linear dimension and angle dimension at the same time, we focus on the orthogonal of the vector generated from the L target.
Geometrically, two vectors are derived from feature points on the L target.
The angle between the specification of the vector and the vector is parameterized by the laser plane coordinates, the internal and external matrix of the camera.
According to the known angle and specification of the vector on the L target, the modeling angle reconstruction error and distance reconstruction error are integrated in the optimization target function.
In the study, the measurement accuracy is improved by optimizing the process.
By this method, both distance and angle are considered in the optimization object.
In addition, for accurate measurement, different scales of distance and angle are normalized.
The rest of the paper is divided into three parts: the second section describes the solution model of the active vision of the eyes with the normalized object of vector orthogonal.
Then, in section 3, a comprehensive comparative experiment is carried out to verify the method.
The fourth section summarizes the paper.
Custom message
Related Products
